JP2018055597A - Vehicle type discrimination device and vehicle type discrimination method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a vehicle type discrimination device and a vehicle type discrimination method, which can provide various services by identifying a specific vehicle type by utilizing existing facilities.SOLUTION: A vehicle type discrimination device 103 comprises a storage unit 95, a detection unit 92, and a discrimination unit 95. The storage unit stores a specific vehicle type in a vehicle type classification of a vehicle in association with a reference image for each face as a discrimination reference for each of the plurality of faces of the vehicle. The detection unit detects a plurality of faces from a captured image captured by an image capturing device of the vehicle. The discrimination unit collates each of the plurality of faces detected by the detection unit with the reference image for each face to obtain a similarity degree indicating the degree of similarity with the reference image for each face, and performs weighting addition by multiplying the degree of similarity of each face by weighting coefficients to obtain a vehicle type similarity which is a degree of similarity for each type of vehicle, and discriminates a model of the captured vehicle based on the vehicle type similarity.SELECTED DRAWING: Figure 2

Description

  Embodiments described herein relate generally to a vehicle type identification device and a vehicle type identification method.

  2. Description of the Related Art Conventionally, a technique for discriminating vehicle types such as ordinary vehicles, medium-sized vehicles, and large-sized vehicles based on the vehicle width, vehicle length, and vehicle height estimated from captured images is known. In addition, a technique for discriminating a vehicle type classification using information registered in an ETC card of a vehicle traveling on a road is known.

  Conventionally, a technique for counting the number of vehicles traveling on a road from an image captured by a low-resolution camera such as a CCTV (Closed-Circuit Television) camera is known. In the conventional techniques such as these, it is required to discriminate a specific vehicle type with high accuracy.

JP2013-148971A

  However, from the vehicle width, vehicle length, and vehicle height information, it is possible to discriminate up to vehicle types such as ordinary vehicles, medium-sized vehicles, and large vehicles, but the specific vehicle types belonging to each vehicle type category It was difficult to identify. Also, the ETC card contains information on the vehicle type classification for determining the toll class of the expressway, but since specific vehicle type information is not registered, the specific vehicle type is selected from the information on the ETC card. It was also difficult to identify.

  In general, an image captured by a CCTV camera or the like has a low resolution, and the angle at which the vehicle is imaged differs depending on the installation position of each CCTV camera or the like. Therefore, the reference image for specifying a specific vehicle type is accurate. It was difficult to match. On the other hand, it may be difficult to newly install a high-resolution camera for the purpose of discriminating a specific vehicle type.

  The vehicle type identification device of the embodiment includes a storage unit, a detection unit, and a determination unit. A memory | storage part matches and memorize | stores the specific vehicle type in the vehicle type classification | category of a vehicle, and the reference | standard image for every surface used as the determination criterion of each of the several surface of a vehicle. The detection unit detects a plurality of surfaces from a captured image obtained by imaging the vehicle by the imaging device. The discriminating unit collates each of the plurality of surfaces detected by the detecting unit with the reference image for each surface, obtains a similarity indicating the degree of similarity with the reference image for each surface, and weights the similarity for each surface By multiplying the coefficient and performing weighted addition, a vehicle type similarity that is a similarity for each vehicle type is obtained, and the vehicle type of the imaged vehicle is determined based on the vehicle type similarity.

FIG. 1 is a diagram illustrating an example of a configuration of a vehicle type identification system according to an embodiment. FIG. 2 is a block diagram illustrating an example of a functional configuration of the vehicle type identification device according to the embodiment. FIG. 3A is a diagram illustrating an example of captured images continuously captured according to the embodiment. FIG. 3B is a diagram illustrating an example of detection of the outline of the vehicle according to the embodiment. FIG. 4 is a diagram illustrating an example of setting a virtual plane according to the embodiment. FIG. 5A is a diagram illustrating an example of a shape of a vehicle in a captured image according to the embodiment. FIG. 5B is a diagram illustrating another example of the shape of the vehicle in the captured image according to the embodiment. FIG. 5C is a diagram illustrating another example of the shape of the vehicle in the captured image according to the embodiment. FIG. 5D is a diagram illustrating another example of the shape of the vehicle in the captured image according to the embodiment. FIG. 6 is a diagram illustrating an example of detection of the front portion of the outline of the vehicle using the virtual plane according to the embodiment. FIG. 7 is a diagram illustrating an example of vehicle width measurement according to the embodiment. FIG. 8A is a diagram illustrating an example of a circumscribed rectangular parallelepiped of the vehicle according to the embodiment. FIG. 8B is a diagram for explaining details of each side constituting the circumscribed rectangular parallelepiped of the vehicle according to the embodiment. FIG. 9 is a diagram illustrating an example of conversion of the shape of each surface of the vehicle according to the embodiment. FIG. 10 is a diagram illustrating an example of a vehicle type classification database according to the embodiment. FIG. 11 is a diagram illustrating an example of a dictionary database according to the embodiment. FIG. 12 is a flowchart illustrating an example of a flow of a vehicle type determination process according to the embodiment. FIG. 13 is a diagram illustrating an example of the effect of the vehicle type identification device according to the embodiment. FIG. 14 is a diagram illustrating another example of the effect of the vehicle type identification device according to the embodiment. FIG. 15A is a diagram illustrating an example of a captured image in a normal state according to a modified example. FIG. 15B is a diagram illustrating an example of a captured image when the illuminance according to the modification is low. FIG. 15C is a diagram illustrating an example of a captured image when the illuminance according to the modification is very low. FIG. 16 is a diagram illustrating an example of collation between a plurality of reference images and captured images according to the resolution according to the modification.

  FIG. 1 is a diagram illustrating an example of a configuration of a vehicle type identification system 1 according to the present embodiment. As shown in FIG. 1, the vehicle type identification system 1 of this embodiment includes an imaging device 101, an ETC radio wave receiver 102, and a vehicle type identification device 103.

  The imaging device 101 of the present embodiment is a monocular camera or the like that images the vehicle 10 traveling on the road. As the imaging device 101, for example, an existing CCTV camera installed for a traffic counter that images a group of vehicles traveling on a road and counts the number of vehicles 10 may be employed. The resolution of the captured image captured by the image capturing apparatus 101 does not have to be high, and it is only necessary to secure enough pixels to count the number of vehicles 10.

The imaging device 101 is installed, for example, at a height of about 7 m from the road, and images the vehicle 10 from obliquely above. As another example, the imaging apparatus 101 may be installed at a position where the vehicle 10 traveling on the road is photographed from the side, and imaging conditions such as an installation position, an installation angle, and an angle of view of the imaging apparatus 101 are provided. There are no strict restrictions on the.
The imaging device 101 transmits a captured image obtained by capturing the vehicle 10 to the vehicle type identification device 103.

  The ETC radio wave receiver 102 is a device that receives information registered in the ETC card of the vehicle 10 traveling on the road. The ETC radio wave receiver 102 receives the vehicle type classification of the vehicle 10 registered in the ETC card of the vehicle 10 and transmits it to the vehicle type discrimination device 103 in real time.

  The vehicle type identification device 103 is an electronic computer that executes vehicle type identification processing. The vehicle type identification device 103 is electrically connected to the imaging device 101 and the ETC radio wave receiver 102.

  The vehicle type identification device 103 inputs a captured image from the imaging device 101. Then, the vehicle type determination device 103 determines the vehicle type classification and specific vehicle type of the vehicle 10 from the input captured image in real time. Further, the vehicle type identification device 103 receives the vehicle type classification of the vehicle 10 registered in the ETC card of the vehicle 10 from the ETC radio receiver 102, and improves the accuracy of the determination of the vehicle type classification.

  Here, the vehicle type classification in the present embodiment is a classification for classifying the type of the vehicle 10. For example, according to the classification used in the highway toll standard, the type of vehicle 10 is classified into a light vehicle (including a two-wheeled vehicle), a normal vehicle, a medium-sized vehicle, a large vehicle, and an extra large vehicle. Classify.

  The vehicle type is a type name that more specifically specifies the type of the vehicle 10 included in the vehicle type division. For example, the vehicle type classification “large vehicle” includes vehicle types such as ordinary trucks, medium-sized buses, and tractors. The vehicle type classification “extra large vehicle” includes vehicle types such as large buses and semi-trailers. Hereinafter, when simply referred to as “vehicle type”, this specific vehicle type is meant.

  The vehicle type discriminating apparatus 103 transmits the discriminated vehicle type classification and the vehicle type to a computer or the like of a management center (hereinafter referred to as “center”) that centrally manages vehicle information. The vehicle type identification device 103 of this embodiment is installed on the side of the road in FIG. 1, but is not limited to this, and a configuration installed in the center may be adopted.

Next, a functional configuration of the vehicle type identification device 103 will be described.
FIG. 2 is a block diagram illustrating an example of a functional configuration of the vehicle type identification device 103 according to the present embodiment. As shown in FIG. 2, the vehicle type identification device 103 of this embodiment includes an input unit 90, a communication unit 91, a detection unit 92, a conversion unit 93, an adjustment unit 94, a determination unit 95, and a storage unit 96. And comprising.

  As shown in FIG. 2, the storage unit 96 stores a vehicle type classification database 961 (hereinafter referred to as “vehicle type classification DB 961”) and a dictionary database 962 (hereinafter referred to as “dictionary DB 962”). Details of each database will be described later. The storage unit 96 is a storage medium such as a hard disk drive (HDD), a solid state drive (SSD), or a memory.

  The input unit 90 receives an input of a captured image obtained by capturing the vehicle 10 from the imaging device 101. Further, the input unit 90 sends the input captured image to the detection unit 92.

  The communication unit 91 acquires the vehicle type classification of the vehicle 10 registered in the ETC card of the vehicle 10 from the ETC radio wave receiver 102. The communication unit 91 sends the acquired vehicle type classification to the determination unit 95. The communication unit 91 transmits the vehicle type classification and the vehicle type of the vehicle 10 determined by the determination unit 95 described later to the center in real time via a network such as the Internet.

  The detection unit 92 detects the outline of the vehicle 10 from the captured image captured by the imaging device 101. Specifically, first, the detection unit 92 extracts a feature point of the vehicle 10 from the captured image. The feature point is, for example, an edge (corner portion) of the vehicle 10 in the captured image.

Further, when the entire view of the vehicle 10 is not included in the range captured by the imaging device 101 at a time, the detection unit 92 detects the outline of the vehicle 10 from a plurality of captured images that are continuously captured.
FIG. 3A is a diagram illustrating an example of captured images continuously captured according to the present embodiment. The captured images 301 to 304 illustrated in FIG. 3A are images of the moving vehicle 10 continuously captured by the imaging device 101. In the picked-up images 301 to 304, the portion from the head portion to the rear portion of the vehicle 10 is picked up as time passes. The detection unit 92 detects the feature points of the vehicle 10 from the captured images 301 to 304, and tracks the movement of the feature points accompanying the movement of the vehicle 10, thereby identifying the outline of the vehicle 10 included in the captured images 301 to 304. By connecting them, the entire outline of the vehicle 10 is detected.

  FIG. 3B is a diagram illustrating an example of detection of the outer shell 305 of the vehicle according to the present embodiment. The detection unit 92 detects the outline 305 of the vehicle 10 from the feature points extracted from the captured images 301 to 304 as shown in FIG. 3B.

Further, the detection unit 92 detects a change vector (luminance gradient) of the luminance value in the small region in the captured images 301 to 304 shown in FIG. 3A, and detects the outline 305 of the vehicle 10 from the position where the luminance contour line becomes a closed loop. It may be detected. Further, when the imaging angle of the imaging device 101 is specified, the detection unit 92 may detect the outline 305 of the vehicle 10 using pattern matching with a previously registered image.
Further, the captured images 301 to 304 shown in FIG. 3A have a constant imaging angle with respect to the vehicle 10, but the configuration is not limited to this, and the imaging apparatus 101 adopts a configuration in which the imaging angle is changed as the vehicle 10 moves. You may do it.

  The detection unit 92 further uses a virtual plane in order to improve the accuracy of detection of the outer shell 305 of the vehicle 10. Since the captured image captured by the image capturing apparatus 101 according to the present embodiment is assumed to have low resolution, the outline 305 of the vehicle 10 may not be clearly identified from the feature points detected from the captured image. In this case, the detection unit 92 uses the virtual plane to detect the outline 305 of the vehicle 10 from the captured image with higher accuracy.

In the present embodiment, the virtual plane is a plane virtually set at a predetermined position on the road in image processing.
The detection unit 92 sets a virtual plane provided in a direction intersecting the traveling direction of the vehicle 10 on the road on which the vehicle 10 passes. The detection unit 92 obtains a measured value of the vehicle width of the vehicle 10 by comparing the length of the bottom of the virtual plane and the length of the bottom of the front portion of the vehicle 10 on the captured image. In addition, the detection unit 92 detects the circumscribed cuboid of the vehicle 10 by translating each side constituting the virtual plane and obtaining the position coordinates in contact with the outline 305 of the vehicle 10 in the captured image.

The setting of the virtual plane in the present embodiment will be specifically described below.
FIG. 4 is a diagram illustrating an example of setting a virtual plane according to the present embodiment. An image 401 is an example of a captured image captured by the imaging device 101. As shown in an image 401, the captured image includes a road on which the vehicle 10 passes. Since it is assumed that the road width is known to be about 3.5 m and that white lines are drawn at both ends of the road, the detection unit 92 is a line indicating both ends of the road as shown in an image 402. L100 and L101 are detected. Further, as shown in an image 403, the detection unit 92 provides virtual horizontal lines L1 to L4 that intersect with the lines L100 and L01 indicating both ends of the road at regular intervals so as to cross the road. Further, as shown in the image 404, the detection unit 92 uses the horizontal lines L1 to L4 as the bottom sides, and virtual vertical lines standing in the vertical direction with respect to the road from both ends of the horizontal lines L1 to L4 as the left and right sides. Virtual planes R1 to R4 are set with the upper side being a line parallel to the horizontal lines L1 to L4 connecting the two vertical lines. As shown in the image 405, the virtual planes R1 to R4 are provided in a direction perpendicular to the road on the road on which the vehicle 10 passes. Here, the vertical direction is an example of a direction intersecting the road, and the installation angles of the virtual planes R1 to R4 are not limited to this.

  Since the traveling direction of the vehicle 10 is assumed to be parallel to the lines L100 and L101 indicating both ends of the road, the virtual planes R1 to R4 are provided in a direction perpendicular to the traveling direction of the vehicle 10. Therefore, as shown in the image 405, when the virtual plane R2 is in contact with the leading portion of the vehicle 10 and the virtual plane R4 is in contact with the position of the rearmost portion of the vehicle 10, the space on the road divided by the virtual planes R2 and R4 However, it becomes the shape of a rectangular parallelepiped surrounding the vehicle 10. By setting the virtual planes R1 to R4 in this way, changes in geometric information such as deformation of the shape of the vehicle 10 on the captured image as the vehicle 10 travels are considered by considering the orientation of the virtual planes R1 to R4. It becomes possible to predict.

  For the setting of the virtual planes R1 to R4, it is preferable to use information such as white lines painted on the roads or utility poles. However, for roads with few such landmarks, the width and height, such as container vehicles, may be used. The virtual planes R <b> 1 to R <b> 4 may be created by tracking the movement of the vertices of the standardized shape of the vehicle 10 for each time series.

  Further, the installation positions of the virtual planes R1 to R4 may determine the optimal position on the road for imaging the vehicle 10 for vehicle type discrimination. For example, when a plurality of continuous captured images include the entire view of the vehicle 10 multiple times, the captured image at the time when the head portion of the vehicle 10 is in contact with the specific virtual planes R1 to R4 among the virtual planes R1 to R4. May be used for vehicle type discrimination. Moreover, the installation position and the number of installation of virtual plane R1-R4 shown in FIG. 4 are examples, and do not limit the number of installations.

  5A to 5D are diagrams respectively showing examples of the shape of the vehicle 10 in the captured images 501 to 504 according to the present embodiment. The captured images 501 to 504 shown in FIGS. 5A to 5D are all images of the same vehicle 10, but the shape of the vehicle 10 in the captured images changes as the vehicle 10 moves. For example, in the captured image 501 illustrated in FIG. 5A, the contour of the vehicle 10 is not clear because the vehicle 10 is far from the imaging device 101. On the other hand, in the captured image 504 shown in FIG. 5D, since the vehicle 10 is close to the imaging device 101, the outline of the vehicle 10 is clear, but the shape of the vehicle 10 is deformed. A captured image 503 illustrated in FIG. 5C has a clear outline of the vehicle 10 and is less deformed on the shape of each surface of the vehicle 10 than other captured images.

  In such a case, the detection unit 92 compares the preset virtual planes R1 to R4 with the feature points of the vehicle 10 of the plurality of captured images 501 to 504, so that the optimal position for vehicle type determination is obtained. You may employ | adopt the structure which selects the captured image 503 which the vehicle 10 reached | attained.

Next, detection of the front portion of the outer shell 305 of the vehicle 10 using the virtual plane R2 in contact with the head portion of the vehicle 10 among the virtual planes R1 to R4 set as described above will be described. The detection unit 92 detects the front portion of the outer shell 305 of the vehicle 10 detected from the feature points more clearly using the virtual plane R2.
FIG. 6 is a diagram illustrating an example of detection of a front portion of the outer shell 305 of the vehicle 10 by the virtual plane R2 according to the present embodiment. As shown in FIG. 6, the leading portion of the vehicle 10 in the captured image is in contact with the virtual plane R2. The detection unit 92 translates the bottom side of the virtual plane R2, and obtains position coordinates in contact with the bottom surface of the outer shell 305 of the vehicle 10 in the captured image.

  Specifically, the detection unit 92 translates the bottom of the virtual plane R <b> 2 and obtains position coordinates that contact the bottom surface of the outer shell 305 of the vehicle 10 detected from the feature points of the vehicle 10. Similarly, with respect to the other sides, the detection unit 92 translates each side of the virtual plane R <b> 2 and obtains position coordinates in contact with the outline 305 of the vehicle 10 in the captured image. In this way, the detection unit 92 more clearly detects the outline 305 of the front portion of the vehicle 10 using the virtual plane R2.

In FIG. 6, the outline of the vehicle 10 is clear for the sake of explanation. However, when the resolution of the captured image is low, an unclear case such as a feature point of the vehicle 10 being interrupted is assumed. Even in such a case, by using the virtual plane R2, it is possible to interpolate the feature points and detect the outline 305 of the vehicle 10 with higher accuracy.
Further, R2 ′ shown in FIG. 6 is obtained by converting the shape of the conversion unit 93 described later on the front surface of the vehicle 10 on the captured image. The conversion will be described later.

  Further, the detection unit 92 measures the vehicle width measurement value of the vehicle 10 by detecting the front portion of the outer shell 305 of the vehicle 10. Specifically, since the width of the road is known, the length of the bottom side of the virtual plane R2 is known. The detection unit 92 measures the measured value of the vehicle width of the vehicle 10 by comparing the length of the bottom of the known virtual plane R2 with the length of the bottom of the detected front portion of the outline 305.

  FIG. 7 is a diagram illustrating an example of vehicle width measurement according to the present embodiment. As illustrated in FIG. 7, the detection unit 92 obtains the vehicle width by measuring the lengths of the line segments [1] a to [1] c corresponding to the bottom sides of the front portions of the vehicles 10a to 10c on the captured images. The data is sent to the determination unit 95. The determination unit 95 determines the vehicle type classification of the vehicles 10a to 10c from the vehicle width. Details of the discrimination of the vehicle type will be described later.

Next, detection of the circumscribed cuboid of the vehicle 10 will be described.
FIG. 8A is a diagram illustrating an example of a circumscribed rectangular parallelepiped of the vehicle 10 according to the present embodiment. As shown in FIG. 8A, the detection unit 92 detects a circumscribed cuboid 800 that circumscribes the outer shell 305 of the vehicle 10 on the captured image. This is because even when the resolution of the captured image is low, the image of the vehicle 10 can be cut out from the entire captured image by detecting the circumscribed cuboid 800 that circumscribes the outer shell 305 of the vehicle 10. In particular, as shown in FIG. 8A, when the vehicle 10 is a bus, the shape of the vehicle body approximates a rectangular parallelepiped, and therefore the vehicle 10 can be cut out from the captured image with high accuracy by detecting the circumscribed rectangular parallelepiped 800. .

FIG. 8B is a diagram illustrating details of each side constituting the circumscribed rectangular parallelepiped 800 of the vehicle 10 according to the present embodiment. In the image processing according to the present embodiment, the detection unit 92 obtains the position coordinates of the line segments [1] to [9] and the intersection points P1 to P7 constituting the circumscribed rectangular parallelepiped 800.
First, as described with reference to FIG. 6, the detection unit 92 translates the base of the virtual plane R <b> 2 with which the top portion of the vehicle 10 is in contact with the upper direction, and obtains the position coordinates in contact with the outline 305 of the vehicle 10 obtained from the feature points. The line segment [1] corresponding to the bottom side of the circumscribed rectangular parallelepiped 800 is detected.
Then, the detection unit 92 translates the vertical side on the right side of the virtual plane R2 in the captured image in the left direction of the screen, and obtains the position coordinates in contact with the side surface of the outer shell 305 of the vehicle 10, so that the circumscribed cuboid 800 is A line segment [2] corresponding to the front vehicle height is detected. At this time, by determining the intersection P1 between the line segment [1] and the line segment [2], the end point of the line segment [1] and the start point of the line segment [2] are determined.

The detection unit 92 detects line segment [3] by translating a line L101 indicating the right end of the road in the captured image in the left direction of the screen and obtaining a position coordinate in contact with the upper side surface of the outer surface 305 of the vehicle 10. When the detection unit 92 detects the line segment [3], an intersection point P2 between the line segment [2] and the line segment [3] is determined. The intersection point P2 is the end point of the line segment [2] and the start point of the line segment [3].
Moreover, the detection part 92 may detect line segment [3] by detecting the line segment which connects the vertex of the upper right of continuous virtual plane R1-R4 instead of the line L101, and translating.

Further, the detection unit 92 detects a virtual plane that contacts the rear portion of the vehicle 10. For example, it is assumed that the virtual plane R4 is in contact with the rear portion of the vehicle 10 as an image 405 in FIG. The detection unit 92 translates the upper side of the virtual plane R4 in the downward direction and obtains the position coordinates in contact with the upper side of the rear part of the outer shell 305 of the vehicle 10, thereby obtaining a line segment [4 corresponding to the upper side of the rear part of the circumscribed rectangular parallelepiped 800. ] Is detected. The detection unit 92 sets a point where the line segment [3] and the line segment [4] intersect as an intersection point P3.
Alternatively, when the detection unit 92 can obtain the position coordinates of the intersection point P3 that is the starting point of the line segment [4] from the feature point of the vehicle 10, the line segment [3] is obtained by connecting the intersection point P2 and the intersection point P3. It may be detected.

The detection unit 92 translates the vertical side on the left side of the virtual plane R4 in the right direction of the screen, and obtains the position coordinates in contact with the side surface of the outer shell 305 of the vehicle 10, so that the vehicle height at the rear part of the circumscribed rectangular parallelepiped 800 is obtained. Corresponding line segment [5] is detected. At this time, by determining the intersection P4 between the line segment [4] and the line segment [5], the end point of the line segment [4] and the start point of the line segment [5] are determined.
Alternatively, the detection unit 92 may detect the line segment [5] by translating the previously detected line segment [2] and obtaining the position coordinates in contact with the side surface of the outer shell 305 of the vehicle 10.

  The detection unit 92 detects the line segment [6] by translating the previously detected line segment [3] and obtaining the position coordinates in contact with the bottom of the side surface of the outer shell 305 of the vehicle 10. At this time, the intersection P5 of the line segment [5] and the line segment [6] is determined, and the end point of the line segment [5] and the start point of the line segment [6] are determined. Further, the intersection point P6 of the line segment [1] and the line segment [6] is determined, and the start point of the line segment [1] and the end point of the line segment [6] are determined.

  Alternatively, the detection unit 92 translates the line L100 indicating the left end of the road in the captured image in the right direction of the screen and obtains the position coordinates in contact with the bottom side of the side surface of the outer shell 305 of the vehicle 10, thereby obtaining a line segment [6]. May be detected. Moreover, the detection part 92 may detect line segment [6] by detecting the line segment which connects the lower left vertex of continuous virtual plane R1-R4 instead of the line L100, and translating.

  Since the line segments [7] to [9] are located inside the vehicle 10 in the captured image, the position coordinates are obtained from the outline 305 of the vehicle 10 with higher accuracy than the line segments [1] to [8]. Is more difficult. The line segment [7] is a line segment corresponding to the upper side of the front surface of the vehicle 10. The line segment [8] is a line segment corresponding to the upper side of the side surface of the vehicle 10. Line [9] is a line corresponding to the height of the front of the vehicle 10. Basically, line [7] is a line [1], line [8] is a line [3], and line [9] is a line that is parallel to line [2]. However, depending on the imaging angle of the imaging device 101 and the size of the vehicle 10, the shape of the vehicle 10 in the captured image may be deformed. In such a case, even if a line segment parallel to the line segments [1] to [3] is obtained, the line segments [7] to [9] do not cross one point at the intersection P7, and the circumscribed rectangular parallelepiped 800 is detected. May be difficult.

  Therefore, in the present embodiment, the allowed parallel scores for the line segment [7] and the line segment [1], the line segment [8] and the line segment [3], and the line segment [9] and the line segment [2], respectively. This range is set in advance. The parallel score is a value indicating how close the parallel relationship of the two line segments is. For example, if two line segments are completely parallel, the parallel score may be “0”, and if the angle that intersects when two line segments are expanded is 10 °, the parallel score may be “10”. . The detection unit 92 determines that the two line segments whose parallel scores are within a predetermined range are substantially parallel. The threshold value indicating the predetermined range of the parallel score may be stored in the storage unit 96.

  The detection unit 92 has a parallel score of the line segment [7] and the line segment [1], a parallel score of the line segment [8] and the line segment [3], and a parallel score of the line segment [9] and the line segment [2]. By determining the position coordinates of the intersection point P7 that are all within the predetermined range, the position coordinates of the appropriate line segments [7] to [9] are determined.

  In this way, the detection unit 92 detects the line segments [1] to [9] constituting the circumscribed rectangular parallelepiped 800 from the captured image. That is, the detection unit 92 sets virtual planes R1 to R4 provided in a direction perpendicular to the traveling direction of the vehicle 10 on the road on which the vehicle 10 passes, and sets each side constituting the virtual planes R1 to R4. The circumscribed cuboid 800 is detected by moving in parallel and obtaining the position coordinates in contact with the outline 305 of the vehicle 10 in the captured image.

  The above-described detection of the line segments [1] to [9] is an example, and a configuration in which the circumscribed rectangular parallelepiped 800 of the vehicle 10 is detected from the captured image by another method may be employed.

  Next, the detection unit 92 cuts out a plurality of surfaces of the circumscribed rectangular parallelepiped 800 from the captured image. Specifically, the detection unit 92 sets the surface surrounded by the line segment [1], the line segment [2], the line segment [7], and the line segment [9] illustrated in FIG. 8B as the front surface of the circumscribed rectangular parallelepiped 800. cut. Further, the detection unit 92 cuts out a surface surrounded by the line segment [3], the line segment [4], the line segment [7], and the line segment [8] as the upper surface of the circumscribed rectangular parallelepiped 800. In addition, the detection unit 92 cuts out a surface surrounded by the line segment [5], the line segment [6], the line segment [8], and the line segment [9] as a side surface of the circumscribed rectangular parallelepiped 800. Since the traveling direction of the vehicle 10 is known from the road information, it can be estimated which surface of the circumscribed cuboid 800 corresponds to which surface of the vehicle 10.

  As shown in FIG. 8A, the front, side, and top surfaces of the circumscribed cuboid 800 of the vehicle 10 correspond to the front, side, and top surfaces of the vehicle 10 in the captured image, respectively. That is, the detection unit 92 detects the circumscribed cuboid 800 of the vehicle 10 from the captured image, and detects the front, side, and upper surface of the vehicle 10 by cutting out the front, side, and upper surface of the detected circumscribed cuboid 800. 8A and 8B, the front, side, and top surfaces of the vehicle 10 are detected. However, the present invention is not limited to this, and when the imaging device 101 images the vehicle 10 from behind, the detection unit 92 detects the back surface of the vehicle 10. You may employ | adopt the structure to detect.

  By detecting the circumscribed cuboid 800 of the vehicle 10, the detection unit 92 can cut out the image of the vehicle 10 from the captured image even when the resolution of the captured image is low and the outline 305 of the vehicle 10 is unclear. . Moreover, the detection part 92 can detect the several surface of the vehicle 10 from the same captured image by cutting out the several surface of the circumscribed rectangular parallelepiped 800. FIG.

  Returning to FIG. 2, the detection unit 92 sends the detected front, side, and top surfaces of the vehicle 10 to the conversion unit 93. The conversion unit 93 converts the shape of each surface detected by the detection unit 92 from the captured image into a rectangular shape. As the conversion method, a known method is used.

  FIG. 9 is a diagram illustrating an example of conversion of the shape of each surface of the vehicle 10 according to the present embodiment. In the present embodiment, the imaging device 101 does not always capture each surface of the vehicle 10 from the front. When the imaging device 101 images the vehicle 10 from an oblique angle or the like, the shape of each surface of the vehicle 10 in the captured image is deformed. For example, the upper surface A of the circumscribed rectangular parallelepiped 800 of the vehicle 10 shown in FIG. 9 is deformed to a shape close to a parallelogram. In this state, it is difficult for the determination unit 95 described later to accurately collate images.

  Therefore, as illustrated in FIG. 9, the conversion unit 93 converts the upper surface A, the side surface B, and the front surface C of the circumscribed rectangular parallelepiped 800 of the vehicle 10 into a rectangular upper surface A ′, side surface B ′, and front surface C ′, respectively. The conversion unit 93 may use the virtual planes R1 to R4 in converting the shape of each surface of the circumscribed rectangular parallelepiped 800. For example, as shown in an image 405 in FIG. 4, when the virtual plane R2 has a shape close to a parallelogram, the conversion unit 93 converts the shape of the virtual plane R2 so that all the internal angles of the virtual plane R2 are right angles. To do. And the conversion part 93 measures the change of the position of the line segment which comprises virtual plane R2, and applies the change of the position of the measured line segment to each line segment of the circumscribed rectangular parallelepiped 800, and thereby the circumscribed rectangular parallelepiped 800 of the vehicle 10 is changed. The shape of the front C is converted into a rectangular shape. The conversion unit 93 may adopt a configuration using the virtual planes R1 to R4 even when the shapes of the upper surface A and the side surface B are converted into a rectangle.

  Returning to FIG. 2, the conversion unit 93 sends the converted upper surface A ′, side surface B ′, and front surface C ′ to the determination unit 95. Further, the conversion unit 93 sends the degree of shape conversion of the upper surface A ′, the side surface B ′, and the front surface C ′ to the adjustment unit 94. The degree of shape conversion is, for example, a numerical value that represents the difference in shape between the surfaces before and after conversion, and may be a value obtained by quantifying the angle of the four line segments surrounding the image, the amount of movement of the feature points, and the like. .

The determination unit 95 determines the vehicle type classification and the vehicle type of the vehicle 10.
First, the determination of the vehicle type classification will be described. As described with reference to FIG. 7, the determination unit 95 acquires the vehicle width of the vehicle 10 detected by the detection unit 92. Then, the determination unit 95 searches the vehicle type division DB 961 for the vehicle type division associated with the vehicle width of the vehicle 10 and determines the vehicle type division of the vehicle 10.

  FIG. 10 is a diagram illustrating an example of the vehicle type classification DB 961 according to the present embodiment. As shown in FIG. 10, the vehicle type division DB 961 stores a vehicle type division and a determination condition for the vehicle type division in association with each other. The vehicle type classification determination condition in the present embodiment is the length of the vehicle 10. That is, the vehicle type division DB 961 stores a vehicle type division and a threshold value of the vehicle width of the vehicle 10 in the vehicle type division in association with each other. For example, when the vehicle width of the vehicle 10 is n1 meters or less, the determination unit 95 determines the vehicle type classification of the vehicle 10 as “light vehicle (including a two-wheeled vehicle)”. Further, when the vehicle width of the vehicle 10 is longer than n3 meters and equal to or less than n4 meters, the determination unit 95 determines the vehicle type classification of the vehicle 10 as “large vehicle”. In addition, when the vehicle width of the vehicle 10 is longer than n4 meters, the determination unit 95 determines the vehicle type classification of the vehicle 10 as “extra large vehicle”. The determination unit 95 also determines the vehicle type divisions “ordinary vehicle” and “medium-sized vehicle” based on the length of the vehicle width.

  The determination unit 95 may send the vehicle type classification determined in this way to the detection unit 92. In this case, the detection part 92 is good also as what detects the above-mentioned circumscribed rectangular parallelepiped 800 only when the vehicle 10 is a specific vehicle type division. For example, a configuration in which the detection unit 92 detects the circumscribed cuboid 800 only when the determination unit 95 determines that the vehicle type classification of the vehicle 10 is “large vehicle” or “extra large vehicle” may be adopted. When this configuration is employed, the processing load due to the detection of the circumscribed cuboid 800 of the detection unit 92 described above can be reduced.

  Further, the determination unit 95 acquires information on the vehicle type classification registered in the ETC card of the vehicle 10 from the ETC radio receiver 102. The determination unit 95 increases the accuracy of determination of the vehicle type classification by matching the vehicle type classification determined from the vehicle width of the vehicle 10 with the vehicle type classification registered in the ETC card. For example, when at least one of the vehicle type classification determined by the determination unit 95 based on the vehicle width and the vehicle type classification registered in the ETC card is “large vehicle” or “extra-large vehicle”, the above-described detection unit 92 includes the circumscribed cuboid 800. A configuration for performing detection may be employed. For example, even if the vehicle type classification determined by the determination unit 95 from the vehicle width is “medium-sized vehicle”, if the vehicle type classification registered in the ETC card is “large vehicle”, the detection unit 92 detects the circumscribed cuboid 800. Assumed to be performed. Even when the resolution of the captured image is low and the vehicle width cannot be detected accurately, the determination unit 95 can improve the accuracy of determination of the vehicle type classification by using the information of the vehicle type classification registered in the ETC card. .

  Depending on the installation location of the imaging apparatus 101, there may be a case where the ETC radio wave receiver 102 cannot be installed. Even in such a case, the determination unit 95 can determine the vehicle type classification from the vehicle width. The determination unit 95 sends the determined vehicle type classification to the communication unit 91.

Moreover, the discrimination | determination part 95 collates the upper surface A ', the side surface B', the front C 'which the conversion part 93 converted, and the reference | standard image for every surface memorize | stored in dictionary DB962, and determines the vehicle type of the vehicle 10. Determine.
FIG. 11 is a diagram showing an example of the dictionary DB 962 according to the present embodiment. As shown in FIG. 11, the dictionary DB 962 stores vehicle types and reference images of front, side, and top surfaces corresponding to the vehicle types. The reference image is an image obtained by previously capturing each surface of the vehicle 10 corresponding to each vehicle type, and serves as a determination criterion for each of the plurality of surfaces of the vehicle 10. The reference image may be an image in which feature points are detected in advance so that matching is easy. In the example shown in FIG. 11, the dictionary DB 962 stores reference images of the front, side, and top surfaces of the vehicle types “large bus”, “semi-trailer”, and “normal truck”, respectively. There may be multiple types of reference images for each vehicle type. For example, although the front reference image associated with the vehicle type “large bus” is one image in FIG. 11, a plurality of front reference images may be associated and stored.

  The discriminating unit 95 collates (matches) the upper surface A ′, the side surface B ′, the front surface C ′ with the reference image for each surface stored in the dictionary DB 962, and shows the similarity indicating the degree of similarity with the reference image for each surface. Find the degree. The determination unit 95 calculates the similarity for each surface by a technique such as template matching. For example, the similarity may be increased as the ratio of the feature points of the upper surface A ′, the side surface B ′, the front surface C ′ and the reference image for each surface overlaps is higher.

  Furthermore, the determination unit 95 calculates the vehicle type similarity that is the similarity for each vehicle type by weighting and adding the calculated similarity for each surface, and based on the obtained vehicle type similarity, the captured vehicle 10 Identify the vehicle type. Specifically, the determination unit 95 determines the vehicle type having the highest vehicle type similarity as the vehicle type of the vehicle 10 captured by the imaging device 101. The vehicle type similarity is a weighted addition value of the similarity for each surface calculated by the equation (1).

  Vehicle model similarity = αS1 + βS2 + γS3 (1)

  In equation (1), S1 is the similarity between the upper surface A ′ and the upper reference image, S2 is the similarity between the side B ′ and the side reference image, and S1 is the similarity between the front C ′ and the front reference image. It is. Α, β, and γ are weighting factors for each surface.

  The greater the weighting factor, the greater the influence of the similarity of the surface on the vehicle type discrimination. For example, the similarity between the upper surface A ′, the side surface B ′, the front surface C ′ and the reference image of the upper surface, the side surface, and the front surface of the vehicle type “large bus” stored in the dictionary DB 962 is “15”, “20”, “10”, respectively. It is assumed that the determination unit 95 obtains “ Further, the similarities between the top surface A ′, the side surface B ′, the front surface C ′, and the reference image of the top surface, the side surface, and the front surface of the vehicle type “semi-trailer” stored in the dictionary DB 962 are “10”, “10”, and “20”, respectively. Similarly, it is assumed that the similarities with the reference images of the top, side, and front of the vehicle type “ordinary truck” are “5”, “5”, and “20”, respectively.

When the top surface weighting factor α is “1”, the side surface weighting factor β is “2”, and the frontal weighting factor γ is “1”, the vehicle type similarity of the vehicle type “large bus” calculated by the equation (1) Is “65”, the vehicle type similarity of the vehicle type “semi-trailer” is “50”, and the vehicle type “normal truck” is “35”. In this case, since the vehicle type having the highest vehicle type similarity is “large bus”, the determination unit 95 determines that the vehicle type of the imaged vehicle 10 is “large bus”. The similarity compared to the front C ′ is larger in the “semi-trailer” and “ordinary truck”, but since the weight coefficient β on the side is the largest, the similarity compared to the side B ′ is higher than that on the other side. The “large bus” is selected with a strong influence on the determination result.
The determination unit 95 sends the determined vehicle type to the communication unit 91.

Further, the discriminating unit 95 may adopt a configuration for further classifying the load when the vehicle type of the vehicle 10 is a vehicle type on which a load is loaded such as a “semi-trailer” or a “normal truck”.
As shown in FIG. 11, the dictionary DB 962 stores a plurality of types of reference images for each type of vehicle. For example, in the vehicle type “semi-trailer”, a reference image of the upper surface of the load and a reference image of the side surface associated with a plurality of types of loads such as “load 1” and “load 2” are stored. When the determination unit 95 determines that the vehicle type of the vehicle 10 is “semi-trailer”, the determination unit 95 compares the upper surface A ′ acquired from the conversion unit 93 with the reference image of the upper surface associated with “load 1” of “semi-trailer”. And obtain the similarity. Further, the determination unit 95 collates the side surface B ′ with the reference image of the side surface associated with the “load 1” of the “semi-trailer”, and obtains the similarity. Then, the determination unit 95 weights and adds the similarities of the upper surface and the side surface to calculate a weighted addition value. The weighted addition value of the similarity with the reference image of each surface of the load is defined as the load similarity. Similarly, the determination unit 95 also calculates the load similarity compared with the reference image of the load after “Load 2”, and determines the type of the load having the maximum load similarity as the load of the vehicle 10.
The determination unit 95 sends the determined classification of the load of the vehicle 10 to the communication unit 91.

  In FIG. 11, both the upper surface A ′ and the side surface B ′ are used for classification of the load, but a configuration using only one of the surfaces may be adopted. That is, the discriminating unit 95 discriminates the vehicle type of the imaged vehicle 10 and further collates the upper surface A ′ or side surface B ′ of the vehicle 10 detected by the detecting unit 92 with the reference image of the upper surface or side surface of the load. Thus, the load of the vehicle 10 is classified. Moreover, when imaging the back surface of the vehicle 10, you may employ | adopt the structure which classifies a load using the image of a back surface.

In addition, the dictionary DB 962 shown in FIG. 11 stores the upper and side reference images of the load separately from the upper and side reference images for determining the vehicle type of the vehicle 10. You may employ | adopt the structure which memorize | stores multiple types of the image of the side surface and upper surface for every vehicle model. In the case of adopting the configuration, for example, after the determination unit 95 determines the vehicle type, the adjustment unit 94 described later increases the weight coefficient of the upper surface and the side surface, and the determination unit 95 performs the determination again to classify the load. You may employ | adopt the structure to do.
In addition, a configuration may be adopted in which the reference image of the cargo is stored in a database different from the dictionary DB 962.

  In addition, when the vehicle type is composed of a front part with a driver's seat such as a “semi-trailer” or “ordinary truck” and a rear part on which the load is loaded, the discrimination part 95 is separated from the front part and the rear part. The cargo may be classified by collating the upper surface A ′ and the side surface B ′ of the rear portion with the reference image.

  By classifying the type of cargo, the weight of the vehicle 10 can be estimated. Further, by determining the type of cargo, the vehicle type of the vehicle 10 can be further divided and determined. For example, the determination unit 95 can identify a vehicle type “pole trailer” that transports a long object (pole), which is a kind of trailer. Furthermore, if the type of the load that is likely to collapse is specified in advance, the determination unit 95 can specify the vehicle 10 loaded with the load classified as such a load.

Returning to FIG. 2, the adjustment unit 94 adjusts the values of the weighting coefficients α, β, and γ.
As an example, the adjustment unit 94 may adopt a configuration that determines the values of the weighting factors α, β, and γ according to the use of the vehicle type identification system 1. Specifically, it is assumed that the use of the vehicle type discrimination system 1 is to discriminate a bus. As shown in FIG. 11, rectangular windows are arranged on the side of the bus. The side surface of the bus is more characteristic than the front and top surfaces, and the similarity of the side surface is important for discriminating between other vehicle types and the bus. In this case, the adjustment unit 94 increases the weighting coefficient β for the side surface as compared with the weighting factors α and γ of the other surfaces.

  If the use of the vehicle type identification system 1 is to classify the load of the vehicle 10, the adjustment unit 94 may increase the weighting coefficient α for the upper surface and the weighting coefficient β for the side surface as compared with γ for the front surface. good. In particular, in the case of a truck with an open upper surface of the loading platform, the image viewed from the upper surface is important in the classification of the load. Therefore, the adjustment unit 94 increases the weight coefficient α for the upper surface as compared with other surfaces. May be.

  As another example, the adjustment unit 94 has a weighting coefficient α, a value that increases as the degree of shape conversion by the conversion unit 93 is smaller among a plurality of surfaces cut out from the circumscribed rectangular parallelepiped 800 of the vehicle 10. You may employ | adopt the structure which determines (beta) and (gamma).

  In the example of conversion by the conversion unit 93 shown in FIG. 9, the degree of shape conversion from the front C to the front C ′ is the smallest, and then the degree of shape conversion from the upper surface A to the upper surface A ′ is small. Moreover, the degree of the shape conversion from the side surface B to the side surface B ′ is larger than that of other surfaces.

  As the degree of shape conversion when the shape of the deformed surface is changed to a rectangle is larger, the amount of correction by image processing increases, so the reliability of similarity on the surface decreases. For this reason, in the example shown in FIG. 9, the similarity obtained by comparing the front C ′ with the reference image is more reliable than the similarity of the upper surface A ′ and the side surface B ′. In this case, the adjustment unit 94 adjusts the value of the weighting factor so that the weighting factor γ with respect to the front surface is the largest and the weighting factor α with respect to the upper surface is then increased.

  When imaging the vehicle 10 using the existing imaging apparatus 101, the vehicle 10 may be inclined or distorted in the captured image depending on the installation position and installation angle of the imaging apparatus 101. In such a case, by adjusting the values of the weighting factors α, β, and γ according to the degree of conversion of the shape of each surface, it is possible to discriminate the vehicle type due to the difference in imaging conditions such as the installation position of the imaging device 101. The influence can be reduced and various imaging conditions can be handled.

  The weighting factor may be newly calculated by the adjustment unit 94 for each vehicle type determination process, or the storage unit 96 stores the initial value of the weighting factor, and the adjustment unit 94 changes the initial value. May be.

  The weighting factor may be a fixed value stored in advance in the storage unit 96. For example, when the use of the vehicle type identification system 1 is determined in advance, the storage unit 96 stores in advance a weighting factor corresponding to the determined use. Further, when the degree of deformation of the shape of each surface of the vehicle 10 in the captured image is estimated in advance by the installation position and the installation angle of the imaging device 101, the degree of shape conversion by the conversion unit 93 for each surface can be estimated. . In such a case, a configuration may be employed in which the storage unit 96 stores a weighting factor that is set in advance such that the smaller the degree of conversion of the estimated shape is, the larger the value is. When adopting a configuration in which the weighting factor is a fixed value, the vehicle type identification device 103 does not need to include the adjustment unit 94.

Next, the vehicle type discrimination process according to this embodiment configured as described above will be described.
FIG. 12 is a flowchart illustrating an example of the flow of the vehicle type determination process according to the present embodiment. First, the input unit 90 acquires a captured image obtained by capturing the vehicle 10 from the imaging device 101 (S1). The input unit 90 sends the acquired captured image to the detection unit 92.

  Next, the detection unit 92 extracts feature points of the vehicle 10 from the captured image (S2). The detection unit 92 detects the outline 305 of the vehicle 10 from the extracted feature points. In addition, the detection unit 92 sets virtual planes R1 to R4 perpendicular to the traveling direction of the vehicle 10 on the captured image.

  The detection unit 92 detects the positional relationship between the outline 305 detected from the feature points of the vehicle 10 and the virtual planes R1 to R4. The detection unit 92 translates each side of the virtual plane R2 in contact with the head portion of the vehicle 10 and obtains position coordinates in contact with the outline 305 of the vehicle 10 in the captured image, whereby the front of the outline 305 of the vehicle 10 is obtained. Detect parts more accurately. The detection unit 92 detects the actual measured value of the vehicle width of the vehicle 10 from the captured image by comparing the length of the bottom of the known virtual plane R2 and the length of the bottom of the front portion of the outline 305 of the vehicle 10 detected. (S3). The detection unit 92 sends the detected vehicle width to the determination unit 95.

  The determination unit 95 acquires the vehicle width of the vehicle 10 detected by the detection unit 92. Then, the determination unit 95 searches the vehicle type division associated with the vehicle width of the vehicle 10 from the vehicle type division DB 961 of the storage unit 96 to determine the vehicle type division of the vehicle 10 (S4).

  When the vehicle type classification determined from the vehicle width of the vehicle 10 is not “Large Car” or “Extra Large Car” (S5 “No”), that is, the vehicle type classification is “light vehicle”, “normal vehicle”, “medium-sized vehicle” If it is determined that the vehicle type is either, the determination unit 95 transmits the determined vehicle type classification to the center (S6). In this case, the vehicle type determination process ends.

  On the other hand, when the vehicle type classification determined from the vehicle width of the vehicle 10 is “large vehicle” or “extra-large vehicle” (S5 “Yes”), the determination unit 95 sends the determination result to the detection unit 92.

  Here, the communication unit 91 may receive the vehicle type classification of the vehicle 10 registered in the ETC card of the vehicle 10 captured by the imaging device 101 from the ETC radio wave receiver 102 and send it to the determination unit 95. In this case, in the process of S5, the determination unit 95 collates the vehicle type classification of the vehicle 10 registered in the ETC card with the vehicle type classification determined from the vehicle width of the vehicle 10, and at least one of them is “large vehicle” or “ In the case of an “oversized vehicle”, the determination unit 95 may send the determination result to the detection unit 92.

  In the present embodiment, the specific vehicle type is determined by narrowing the range to a specific vehicle type classification, so that the detection of the circumscribed cuboid 800 by the detection unit 92, the conversion by the conversion unit 93, and the vehicle type determination load by the determination unit 95 And reducing the load of creating a reference image used for collation.

  When the vehicle type classification is “large vehicle” or “extra large vehicle” (S5 “Yes”), the detection unit 92 translates each side constituting the virtual planes R1 to R4, and the vehicle 10 in the captured image. The circumscribed cuboid 800 of the vehicle 10 is detected from the captured image by obtaining the position coordinates in contact with the outer contour (S7). The detection unit 92 cuts out the upper surface A, the side surface B, and the front surface C from the circumscribed cuboid 800 and sends them to the conversion unit 93.

  The conversion unit 93 converts the upper surface A, the side surface B, and the front surface C of the circumscribed cuboid 800 of the vehicle 10 into a rectangular upper surface A ′, side surface B ′, and front surface C ′, respectively (S8). The conversion unit 93 sends a value indicating the degree of conversion of the shape of each surface to the adjustment unit 94.

  The adjustment unit 94 adjusts the value of the weighting factor so that the weighting factor for the surface having a small value indicating the degree of shape conversion acquired from the conversion unit 93 is increased (S9). The adjustment unit 94 may change the weighting coefficient according to the use of the vehicle type identification system 1. The adjustment unit 94 sends the adjusted weight coefficient to the determination unit 95.

  The discriminating unit 95 collates the images of the upper surface A ′, the side surface B ′, and the front surface C ′ converted by the converting unit 93 with the reference image of each surface for each vehicle type stored in the dictionary DB 962, and the similarity for each surface Is calculated (S10).

  The discriminating unit 95 calculates the vehicle type similarity by weighting and adding the calculated similarity for each surface according to the equation (1) (S11). Until the images of the upper surface A ′, the side surface B ′, and the front surface C ′ converted by the conversion unit 93 are collated with the reference images of all surfaces of all vehicle types stored in the dictionary DB 962 (S12 “No”), the determination unit 95 Repeats the processing of S10 to S12.

  When the images of the upper surface A ′, the side surface B ′, and the front surface C ′ converted by the conversion unit 93 are collated with the reference images of each surface of all vehicle types stored in the dictionary DB 962 (S12 “Yes”), the determination unit 95 Determines the vehicle type having the highest vehicle type similarity as the vehicle type of the vehicle 10 imaged by the imaging device 101 (S13).

  The determination unit 95 transmits the determined vehicle type classification and vehicle type to the center (S14). This is the end of the processing of this flowchart.

  As described above, according to the vehicle type identification device 103 of the present embodiment, the detection unit 92 detects a plurality of surfaces of the vehicle 10 from the captured image, so that even a captured image captured from one direction is used as a reference image. Can be verified. Further, the result of collating a plurality of surfaces of the vehicle 10 with the image for each vehicle type is weighted and added by a weighting coefficient to obtain the vehicle type similarity, thereby integrating the collation results of the plurality of surfaces and improving the accuracy of the collation. Therefore, even when the resolution of the captured image is low, a specific vehicle type in the vehicle type classification can be determined with high accuracy.

  In conventional vehicle type determination, a reference image is collated using a captured image obtained by capturing the vehicle 10 from a specific direction. For example, in the technique of discriminating the vehicle type using an image of a specific surface such as the front of the vehicle, it is necessary to capture the vehicle traveling from a specific angle. In order to capture such a captured image, it is necessary to install the imaging device 101 at a predetermined specific position and angle. Therefore, an existing CCTV camera or the like that is installed without the purpose of vehicle type discrimination is used. It was difficult to discriminate the vehicle type. Although there is a technique for storing reference images corresponding to various imaging angles and resolutions in advance and collating them with the captured images, the load for creating the reference images individually corresponding to the installation conditions of each imaging device 101 is high, and general-purpose It has been difficult to use as an effective discrimination method.

  On the other hand, according to the vehicle type identification device 103 of the present embodiment, the detection unit 92 detects a plurality of surfaces of the vehicle 10 from the captured image, so that the vehicle 10 depends on the installation position and the installation angle of the existing imaging device 101. Even if the imaging angles of are different, the captured image can be collated with the reference image. Even if the resolution of the captured image is low, the accuracy of vehicle type discrimination can be improved by collating a plurality of surfaces with the reference image.

  Further, conventionally, a plurality of surfaces of the vehicle 10 are imaged using a plurality of imaging devices 101, laser scanners, and the like. However, according to the vehicle type identification device 103 of this embodiment, one imaging device 101 captures images. Since a plurality of surfaces of the vehicle 10 can be detected from the captured image, it is easy to utilize an existing CCTV camera or the like.

  According to the vehicle type identification device 103 of the present embodiment, various services can be provided by specifying a specific vehicle type of the vehicle 10 by utilizing the existing equipment in this way.

As an example of various services that can be provided by the vehicle type identification device 103 specifying the vehicle type, there is a service provided by estimating the number of passengers in the vehicle 10.
FIG. 13 is a diagram illustrating an example of the effect of the vehicle type identification device 103 according to the present embodiment. Two trucks are parked in the parking lot of the highway service area 1002a shown in FIG. In addition, two buses are parked in the parking area of the service area 1002b. The vehicles 10 parked in the service areas 1002a and 1002b are classified into the same vehicle type classification.

  However, even if the vehicle type classification is the same, the expected number of passengers estimated for trucks and buses differs greatly. In the conventional discrimination based on the vehicle type registered in the ETC card, the number of vehicles 10 can be counted for each vehicle type, but it is difficult to estimate the number of passengers because the vehicle type information cannot be obtained. there were.

  Therefore, by inputting the captured image captured by the imaging device 101 such as a CCTV camera installed at the entrance of the expressway or at the entrance / exit of the service areas 1002a and 1002b to the vehicle type discrimination device 103 of the present embodiment, The vehicle type can be determined. By counting the number of vehicles 10 and the vehicle type discrimination results at the center 1001, a service of providing an estimated value of the number of users to the service areas 1002a and 1002b can be performed.

  In particular, the number of users using the service areas 1002a and 1002b can be estimated more accurately by determining a large bus such as a sightseeing bus. As a result, the service areas 1002a and 1002b can cope with personnel assignment according to the degree of congestion. Further, by estimating the number of users in each of the service areas 1002a and 1002b, it is possible to provide a service for guiding the driver of the vehicle 10 to the service area 1002a with fewer users. In this way, an information providing service for averaging the degree of congestion of users in a plurality of service areas 1002a, 1002b and parking areas is also possible.

  FIG. 14 is a diagram illustrating another example of the effect of the vehicle type identification device 103 according to the present embodiment. In general, semi-trailers are heavy and have a high load on road infrastructure such as roads and bridges. Conventionally, however, it is difficult to accurately determine the weight load on a road because it is difficult to distinguish the type of vehicle.

  On the other hand, according to the vehicle type identification device 103 according to the present embodiment, as shown in FIG. 14, the vehicle type of the vehicle 10 passing through is identified from the captured image captured by the existing imaging device 101 installed on the road. Then, the result can be transmitted to the center 1001. For this reason, the results of discriminating vehicle types are aggregated by the computer of the center 1001, and the number of semi-trailers that have passed through a certain section and the frequency of traffic are used as reference values for estimating the degree of deterioration of road infrastructure. Can be provided.

Further, as a traffic flow survey for traffic jam prediction or the like, the vehicle type discrimination result determined by the vehicle type discrimination device 103 of the present embodiment may be used.
The various services described above are examples, and the services that can be provided by applying the vehicle type identification device 103 of the present embodiment are not limited to these.

  As described above, according to the vehicle type identification device 103 of the present embodiment, a new imaging device 101 is installed by using the low-resolution imaging device 101 that has already been installed to determine a specific vehicle type with high accuracy. Therefore, it is possible to provide various services by estimating the number of passengers of the vehicle 10 and the weight load on the road.

  Further, according to the vehicle type identification device 103 of the present embodiment, a circumscribed cuboid 800 of the vehicle 10 is detected from the captured image, and a plurality of surfaces of the detected circumscribed cuboid 800 are cut out to detect a plurality of surfaces of the vehicle 10. Therefore, each surface of the vehicle 10 can be accurately detected and collated with the reference image even from a captured image with low resolution.

  Furthermore, according to the vehicle type identification device 103 of the present embodiment, the shape of the plurality of surfaces of the vehicle 10 detected from the captured image is converted into a rectangular shape. Even if the shape of the image is deformed, it can be collated with the reference image.

  Further, in the vehicle type identification device 103 according to the present embodiment, when the weighting coefficient is set to a larger value as the surface of the plurality of surfaces of the vehicle 10 with a smaller degree of shape conversion by the conversion unit 93, the image by shape conversion is used. The vehicle type can be discriminated by giving priority to the degree of similarity between the surface with less change in the reference image and the reference image. As the degree of shape conversion when the shape of the deformed surface is changed to a rectangle is larger, the amount of correction by image processing increases, so the reliability of similarity on the surface decreases. In the vehicle type identification device 103 of the present embodiment, the degree to which the similarity of each surface affects the vehicle type similarity can be adjusted by a weighting coefficient. For this reason, the vehicle type discriminating apparatus 103 can discriminate the vehicle type more accurately from the captured images taken at various angles by increasing the weighting coefficient of the similarity degree of the surface with a low degree of shape conversion. .

  Further, the side with the largest difference in shape between the bus and other vehicle types is the side. For this reason, in the vehicle type identification device 103 according to the present embodiment, the plurality of surfaces to be compared with the reference image include at least the side surface B ′ of the vehicle 10, and the weight coefficient for the side surface B ′ of the vehicle 10 among the plurality of surfaces is the side surface B. When it is larger than the surface other than ', the bus and other vehicle types can be distinguished with high accuracy.

  In addition, according to the vehicle type identification device 103 of the present embodiment, virtual planes R1 to R4 provided in a direction perpendicular to the traveling direction of the vehicle 10 are set on the road on which the vehicle 10 passes, and the virtual planes R1 to R1 are set. The circumscribed cuboid 800 is detected by translating each side constituting R4 and obtaining position coordinates in contact with the outline of the vehicle 10 in the captured image. Therefore, according to the vehicle type identification device 103 of the present embodiment, even when the resolution of the captured image is low and it is difficult to accurately extract the feature points of the vehicle 10, the circumscribed cuboid 800 of the vehicle 10 is detected from the captured image. Each face can be cut out. Further, by setting the virtual planes R1 to R4, the angle and direction of the vehicle 10 in the captured image can be estimated from the change in the shape of each virtual plane R1 to R4. The circumscribed cuboid 800 can be detected regardless of the angle.

  Further, according to the vehicle type identification device 103 of the present embodiment, the plurality of surfaces to be collated with the reference image include at least three surfaces of the vehicle 10, the front surface, the side surface, and the upper surface, thereby ensuring a certain level of accuracy in vehicle type identification. To do. The vehicle type discriminating apparatus 103 according to the present embodiment uses a weighting factor to determine the similarity for each surface obtained by comparing each of the front, side, and top surfaces of the vehicle 10 detected from the captured image with each of the reference images for each surface. By performing weighted addition, the vehicle type similarity is obtained, and the vehicle type having the highest vehicle type similarity is discriminated as the vehicle type of the imaged vehicle 10, so that the accuracy of matching can be improved. For this reason, even when the resolution of the captured image is low, the specific vehicle type in the vehicle type classification can be determined with higher accuracy.

  Furthermore, according to the vehicle type identification device 103 of the present embodiment, the weight of the vehicle 10 is estimated in order to classify the load of the vehicle 10 by collating the detected upper surface or side surface of the vehicle 10 with the reference image. Information can be provided. For this reason, it is possible to estimate the weight load on the road or the like by collecting the information on the cargo classified by the vehicle type discriminating apparatus 103 at the center. Further, by determining the type of cargo, the vehicle type of the vehicle 10 can be further divided and determined. Furthermore, if the type of the load that is likely to collapse is specified in advance, the vehicle 10 loaded with the load classified as such a load can be specified. Therefore, the information on the load classified by the vehicle type identification device 103 is obtained. By summing up at the center, it is possible to analyze the section of the vehicle 10 where there is a risk of cargo collapse, the time zone in which such a vehicle 10 travels a lot, and provide information for safety measures Can do.

  In the present embodiment, a plurality of surfaces of the vehicle 10 are detected from one captured image, but a configuration using a plurality of images may be employed. For example, the determination unit 95 selects the one with the smallest degree of conversion from each surface of the vehicle 10 detected from a plurality of captured images continuously captured by the imaging device 101 and uses it for matching with the reference image. You may employ | adopt the structure to do. When the captured image captured by the imaging apparatus 101 is a moving image, the detection unit 92 may detect each surface of the vehicle 10 from a plurality of frames in the moving image.

  For example, in the continuously captured images 501 to 504 shown in FIGS. 5A to 5D, the degree of deformation of the front and side surfaces of the vehicle 10 is the smallest, and the degree of deformation of the upper surface is the smallest. It is assumed that the captured image 504. In this case, the degree of conversion of the front and side shapes by the conversion unit 93 is the smallest in the captured image 503, and the degree of conversion in the shape of the upper surface is the smallest in the captured image 504. The discriminating unit 95 weights and adds the similarities obtained by comparing the front C ′ and the side B ′ cut out from the captured image 503 and the top surface A ′ cut out from the captured image 504 with the reference image, respectively, using Equation (1). However, a configuration for discriminating the vehicle may be employed.

  Further, a configuration may be adopted in which the optimal positions for imaging the front, side, and top surfaces of the vehicle 10 are set in advance as the imaging positions of the respective surfaces from the imaging range of the imaging device 101. Specifically, an imaging position where an image can be captured without further deformation of the front surface of the vehicle 10, an imaging position where an image can be captured without further deformation of a side surface, and an imaging position where an image can be captured without further deformation of the upper surface are set in advance. When adopting this configuration, the determination unit 95 detects the upper surface A ′ of the vehicle 10 detected from the captured image that optimally captures the upper surface of the vehicle 10 and the side surface of the vehicle 10 detected from the captured image that optimally captures the side surface. B ′, the front C ′ of the vehicle 10 detected from the captured image obtained by optimally capturing the front is collated with the reference image of each surface. The discriminating unit 95 employs a configuration in which the similarity obtained by collating each surface acquired from different captured images with the reference image is weighted and added according to equation (1) to discriminate the vehicle. Also good.

(Modification)
In the above-described embodiment, the weighting coefficient is adjusted according to the degree of conversion of each surface of the vehicle 10 and the use of the vehicle type identification system 1, but a configuration in which the weighting coefficient is adjusted according to the illuminance condition may be adopted.

  For example, the vehicle type identification system 1 may employ a configuration that further includes an illuminance sensor (not shown) that measures the illuminance of the imaging range of the imaging device 101.

When the image pickup apparatus 101 is a CCTV camera, it is assumed that the image quality of the picked-up image is deteriorated or the resolution is lowered due to a disturbance such as the illuminance condition of the road. Differences between captured images depending on illuminance conditions will be specifically described with reference to the drawings.
FIG. 15A is a diagram illustrating an example of a captured image 1501 in a normal state according to the present modification. As shown in FIG. 15A, when the illuminance condition is within an appropriate range, as described in the above embodiment, the circumscribed cuboid 800 of the vehicle 10 is detected, and the upper surface A, the side surface B, and the front surface C are cut out. Can do.

  FIG. 15B is a diagram illustrating an example of a captured image 1502 when the illuminance according to this variation is low. When the illuminance on the road or the like in the imaging range is low, it may be difficult to accurately detect the entire view of the vehicle 10 from the captured image 1502. In the example shown in FIG. 15B, the front and side surfaces of the vehicle 10 in the captured image 1502 are relatively clear, but the top surface of the vehicle 10 is unclear because of low illuminance. In such a case, it is difficult to accurately detect the outline 305 at the top of the vehicle 10 in the captured image 1502. For this reason, the accuracy of detection of the upper surface A of the circumscribed rectangular parallelepiped 800 of the vehicle 10 is lowered. From the position of the illumination device around the imaging device 101, the imaging angle of the imaging device 101, and the like, a surface where the outline 305 becomes unclear when the illuminance is low can be estimated in advance.

  The communication unit 91 in the present modification obtains the illuminance value on the road in the imaging range from an illuminance sensor (not shown). Then, when the illuminance acquired by the communication unit 91 is low, the adjustment unit 94 reduces the weighting coefficient for the surface where the outline 305 is unclear compared to other surfaces. In the example illustrated in FIG. 15B, the adjustment unit 94 decreases the weighting coefficient for the upper surface A compared to the side surface B and the front surface C when the illuminance acquired by the communication unit 91 is low. The discriminating unit 95 uses the weighting coefficient adjusted by the adjusting unit 94 as the vehicle type of the vehicle 10 as the vehicle type of the vehicle 10 with the highest vehicle type similarity obtained by weighted addition of the similarity of each surface according to equation (1). Determine. Thereby, the vehicle type discriminating apparatus 103 can reduce the influence degree of the surface in the vehicle type discrimination when a part of the surface of the vehicle 10 on the captured image becomes unclear due to the illuminance. It is possible to prevent a decrease in the accuracy of discrimination.

  FIG. 15C is a diagram illustrating an example of a captured image 1503 when the illuminance according to this modification is very low. When the illuminance is very low at night or the like, it is assumed that the image quality is greatly deteriorated as compared with the captured image 1501 at normal time. For example, as in a captured image 1503 illustrated in FIG. 15C, light sources for the upper surface and the side surface of the vehicle 10 may be insufficient, and it may be difficult to detect a surface other than the front surface C in the circumscribed cuboid 800 of the vehicle 10. In such a case, the adjustment unit 94 increases the weighting coefficient for the front surface C and decreases the weighting coefficient for the upper surface A and the side surface B.

Further, a configuration may be adopted in which a reference image used for collation when illuminance is low is stored in advance in the dictionary DB 962 separately from the normal reference image.
FIG. 16 is a diagram illustrating an example of collation between a plurality of reference images and captured images according to the resolution according to the present modification. As shown in FIG. 16, a configuration may be adopted in which a plurality of reference images are stored in advance in the dictionary DB 962 for each surface, and collation with an appropriate reference image is performed according to illuminance.

  As described above, according to the vehicle type identification device 103 of the present modified example, by adjusting the weighting factor according to the illuminance condition, it is possible to prevent a decrease in the accuracy of the vehicle type identification due to a change in illuminance on the imaging target road or the like. For this reason, according to the vehicle type identification device 103 of the present modification, the accuracy of the vehicle type identification can be stably ensured even when the illuminance condition changes.

  In this modification, the illuminance is measured by an illuminance sensor (not shown), but a configuration in which the illuminance is estimated based on a time zone may be employed. For example, the illuminance condition may be set assuming that the illuminance is high during the daytime and the illuminance is low during the nighttime. Further, when the imaging apparatus 101 is a CCTV camera, there may be a night mode in which a black and white captured image is captured at night. Such a black and white image is colored and collated with a reference image with high accuracy. A configuration may be adopted.

  Further, instead of measuring the illuminance by an illuminance sensor (not shown), a configuration may be adopted in which the illuminance is estimated and the weighting coefficient is adjusted from the brightness of the captured image. Further, not only the illuminance but also a configuration in which the adjustment unit 94 adjusts the value of the weighting coefficient may be adopted when the resolution of the captured image is lowered due to disturbance such as rain or fog.

  In addition, the vehicle type identification device 103 may further include a function of classifying the paint on the vehicle 10. Further, the vehicle type discriminating apparatus 103 may adopt a configuration in which the discriminated vehicle type of the vehicle 10, the classification of painting, the captured image, etc. are transmitted to the center. The center groups vehicles 10 with the same paint, collects captured images for the purpose of absorbing differences in imaging results of the imaging device 101 due to differences in lighting and weather conditions, classifies information such as vehicle type and cargo, and evaluates patterns Can be saved as Moreover, you may employ | adopt the structure which improves the precision of discrimination | determination of the vehicle type of the vehicle type discrimination | determination apparatus 103 further using the information which the center collected.

  As described above, according to the above-described embodiment and modification, it is possible to provide various services by specifying a specific vehicle type of the vehicle 10 using existing facilities.

  The vehicle type identification device 103 according to the embodiment and the modification includes a control device such as a CPU, a storage device such as a ROM (Read Only Memory) and a RAM, an external storage device such as an HDD and a CD drive device, and a display such as a display device. The apparatus includes an input device such as a keyboard and a mouse, and has a hardware configuration using a normal computer.

  The vehicle type identification program executed by the vehicle type identification device 103 of the embodiment and the modification is a file in an installable format or an executable format, and is a CD-ROM, flexible disk (FD), CD-R, DVD (Digital Versatile Disk). And the like recorded on a computer-readable recording medium.

Further, the vehicle type identification program executed by the vehicle type identification device 103 according to the embodiment and the modification may be provided by being stored on a computer connected to a network such as the Internet and downloaded via the network. good. In addition, the vehicle type identification program executed by the vehicle type identification device 103 according to the embodiment and the modification may be provided or distributed via a network such as the Internet.
Moreover, you may comprise so that the vehicle type discrimination | determination program of embodiment and a modification may be provided by previously incorporating in ROM etc.

  The vehicle type identification program executed by the vehicle type identification device 103 according to the embodiment and the modified example has a module configuration including the above-described units (input unit, communication unit, detection unit, conversion unit, adjustment unit, determination unit). As actual hardware, a CPU (processor) reads out and executes a vehicle type identification program from the storage medium, so that the respective units are loaded on the main storage device, and an input unit, a communication unit, a detection unit, a conversion unit, and an adjustment unit The discriminating unit is generated on the main storage device.

  Although embodiments and modifications of the present invention have been described, these embodiments and the like are presented as examples and are not intended to limit the scope of the invention. These novel embodiments and the like can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

1 vehicle type discrimination system 10, 10a to 10c vehicle 90 input unit 91 communication unit 92 detection unit 93 conversion unit 94 adjustment unit 95 discrimination unit 96 storage unit 101 imaging device 102 ETC radio wave receiver 103 vehicle type discrimination device 301 to 304, 501 to 504 , 1501 to 1503 Captured image 305 Outer frame 800 circumscribed cuboid 961 Vehicle type classification database (vehicle type classification DB)
962 Dictionary database (Dictionary DB)
1001 Center 1002a to 1002b Service area R1 to R4 Virtual plane

Claims (9)

A storage unit that associates and stores a specific vehicle type in the vehicle type classification of the vehicle and a reference image for each surface serving as a determination criterion for each of the plurality of surfaces of the vehicle;
A detection unit that detects the plurality of surfaces from a captured image obtained by imaging the vehicle by the imaging device;
Each of the plurality of surfaces detected by the detection unit is compared with a reference image for each surface to obtain a similarity indicating the similarity to the reference image for each surface, and the similarity for each surface is determined. Determining a vehicle type similarity that is a similarity for each vehicle type by multiplying the weight coefficient and performing weighted addition, and determining the vehicle type of the imaged vehicle based on the vehicle type similarity;
A vehicle type identification device. The detection unit detects a plurality of surfaces of the vehicle by detecting a circumscribed cuboid of the vehicle from the captured image and cutting out a plurality of surfaces of the detected circumscribed cuboid.
The vehicle type identification device according to claim 1. A conversion unit that converts the shape of the plurality of surfaces detected by the detection unit from the captured image into a rectangular shape;
The vehicle type identification device according to claim 2. The weighting factor is a larger value as the surface of the plurality of surfaces has a smaller degree of conversion of the shape by the conversion unit.
The vehicle type identification device according to claim 3. The plurality of surfaces include at least a side surface of the vehicle,
The weight coefficient for the side surface of the vehicle among the plurality of surfaces is larger than the surface other than the side surface,
The vehicle type discrimination device according to any one of claims 1 to 3. The detection unit sets a virtual plane provided in a direction intersecting the traveling direction of the vehicle on a road on which the vehicle passes, translates each side constituting the virtual plane, Detecting the circumscribed cuboid by obtaining position coordinates in contact with the outline of the vehicle in the captured image;
The vehicle type discrimination device according to any one of claims 2 to 4. The plurality of surfaces include at least a front surface, a side surface, and an upper surface of the vehicle,
The discriminating unit weights the similarity for each surface obtained by comparing each of the front, side, and top surfaces of the vehicle detected from the captured image with each of the reference images for each surface by the weighting factor. Obtaining the vehicle type similarity by performing addition, and determining the vehicle type having the largest vehicle type similarity as the vehicle type of the imaged vehicle;
The vehicle type discrimination device according to any one of claims 1 to 6. The determination unit further classifies the load of the vehicle by collating the upper surface or side surface of the vehicle detected by the detection unit with the reference image.
The vehicle type discrimination device according to any one of claims 1 to 7. A vehicle type identification method executed by a vehicle type identification device,
The vehicle type identification device includes a storage unit that stores a specific vehicle type in a vehicle type classification of a vehicle and images of a plurality of surfaces of the vehicle in association with each other,
A detection step in which the imaging device detects the plurality of surfaces from a captured image obtained by imaging the vehicle;
The plurality of detected surfaces and the image are collated, and a vehicle type similarity that is a similarity for each vehicle type is obtained by performing weighted addition using a weighting factor to the similarity for each surface that is a collation result, and the vehicle type A determination step of determining the vehicle type of the imaged vehicle based on the similarity;
Discriminating method including car.