JPH09226490A - Detector for crossing object - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently detect a comparatively small crossing object with a high precision. SOLUTION: On the basis of image data from a photographing device 1, distance distribution data for a front space is provided by means of a distance computing circuit 3. Using a plurality of images photographed at the different times, a moving quantity of a crossing person between images is computed by means of a moving quantity computing device 4. By means of a vertical edge number computing circuit 5, detection windows matching with the dimension of the crossing person as to the vertical area of the photographed image are arranged in the horizontal direction without any gap, and the number of vertical edges for each detection window is computed. When the number of vertical edges based on the vertical edge number computing circuit 5 is the specified threshold value or more, the detected window is detected as a proposed window by means of a microcomputer 7. By means of the microcomputer 7, an object heading to the center of a traveling path is temporarily detected as a crossing person on the basis of the center position of the traveling path and the moving direction of the object, while a detected window, in which the crossing person exists actually, is specified on the basis of the time coordinate process.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for detecting a crossing object, and more specifically, on a traveling road of a vehicle using image data captured by a CCD camera or the like, or distance data obtained by a scanning radar device. The present invention relates to a crossing object detection device for detecting a crossing object existing in a vehicle.

[0002]

2. Description of the Related Art As a conventional method, the distance distribution of an object existing in the front space is calculated by using stereoscopic vision of image processing or by scanning a laser radar, and a crossing person is calculated from the density of this distance distribution. It has detected obstacles including.

[0003]

However, in the method of detecting an obstacle from the density of the distance distribution as described above, when there is a pedestrian on the side of the road, the pedestrian crosses the road (for example, the white line). There was a problem that it was difficult to detect it before it appeared in the lane mark). That is, when a pedestrian is on the side of the road, the distance distribution between the pedestrian and the building behind it may be integrated. In such a case, the pedestrian cannot be detected at this stage, and can be detected as a single object only after the pedestrian comes to the road on which the vehicle travels. As a result, there is a risk that the detection of a crossing will be delayed.

As a method of detecting a moving object on a road, a method using an optical flow has been conventionally proposed. This is to detect a moving object in the image as a lump of moving area, and detect a jumping out of a vehicle, for example, at an intersection from the area. However, although the existing technology enables detection of a relatively large moving object such as an automobile, when detecting a relatively small object such as a person or a bicycle, the number of existing data in the entire image is small, so that detection is not possible. The problem arises that it is possible.

The present invention has been made in view of the above problems, and an object thereof is to detect a crossing object which can detect a relatively small crossing object accurately and efficiently. To provide.

[0006]

In order to achieve the above object, in the invention described in claim 1, the distance distribution measuring means comprises:
The distance distribution data of the object in the front space is acquired. More specifically, the distance distribution data is acquired by using a stereo image obtained by a CCD camera, or the distance distribution data is acquired by using light reflection by a scanning radar device. The movement amount calculation means acquires a plurality of image data of the front space at different times in the front space, and calculates the movement amount of the object. The pedestrian detection means sets an analysis region in which a predetermined crossing object exists in the vertical direction of the image with respect to the image data, provides a detection window corresponding to the size of the crossing object, and sets the detection window in the horizontal direction. When the direction of the movement amount obtained by the movement amount calculating means for the window in which the distance data of the front space exists in the detection window is toward the center of the traveling path of the host vehicle, the detection is performed. The crossing is detected as being present in the window.

That is, as shown in FIG. 15, for example, a region of a captured image of 30 to 60 m in front of the own vehicle is set as a crossing object existing region (analysis region), and an object (crossing person) moves in the analysis region. The amount is calculated. FIG.
As shown in, the detection window is set to have a size suitable for a pedestrian, for example (claim 4). As shown in FIGS. 18A, 18B, and 18C, the detection windows are preferably arranged so as to overlap with adjacent windows by a predetermined amount (for example, a half width) (claim 5). . Then, if the moving direction (direction of the moving amount) of the object in the detection window in which the distance data exists is toward the center of the traveling path of the host vehicle, the existence of the crossing object is detected. In this case, for example, as shown in FIG. 17, the center of the traveling road is detected from image information such as a lane mark indicating the traveling road of the host vehicle (claim 3).

According to the above construction, it is possible to detect a pedestrian crossing the side of the road, and it is possible to predict that the pedestrian is about to cross the running path of the vehicle or jumps out on the running path. Becomes In addition, by detecting a crossing person in units of a detection window, it is possible to avoid detection omission even for a relatively small crossing object, and improve detection accuracy. As a result, highly accurate recognition processing can be performed in front of the host vehicle.

According to the second aspect of the invention, in particular, the vertical edge number calculating means calculates the number of vertical edges in the detection window for each of the detection windows. The crossing candidate window detecting means detects the detection window as a crossing candidate window when the number of vertical edges calculated by the vertical edge calculating means is equal to or more than a predetermined threshold value. further,
The crossing object detection means is such that among the plurality of detected candidate windows, the direction of the movement amount obtained by the movement amount calculation means with respect to the candidate window in which the distance data of the front space exists is the traveling path of the host vehicle. When moving toward the center, the crossing object is detected as being present in the detection window.

That is, paying attention to the fact that the density of the number of vertical edges is high in the image of a pedestrian (crossing object), and the window in which the pedestrian exists is detected based on the number of vertical edges, so that the detection process is efficient. Can be carried out in a specific manner. At this time, the detected window is set as a “candidate window”, and then the actual crossing person is determined from the “candidate window” using the distance data and the movement amount data, so that a more reliable crossing person can be obtained. It becomes possible to detect.

According to the sixth aspect of the present invention, the crossing object detection means uses the detection result of the crossing object in one cycle as a temporary detection result and performs the temporal association processing of the temporarily detected crossing object. The presence of the crossing object is determined when the correspondence is performed for a predetermined cycle or more. In this case, the reliability of crossing detection can be further improved.

According to a seventh aspect of the present invention, the crossing object detection means excludes the yawing component of the own vehicle from the movement amount data obtained by the movement amount calculating means, and the crossing is detected based on the result of the elimination of the yawing component of the own vehicle. Identify the thing. in this case,
The precise movement of the pedestrian can be known, and highly accurate detection results can be obtained.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an outline of a crossing detection system according to this embodiment. In FIG. 1, an imaging device 1 is a device that captures stereoscopic image data with a pair of left and right cameras,
The left and right cameras are configured such that their optical axes are parallel to each other and their base lengths are parallel to the road surface. At this time, the imaging device 1 obtains a stereo image as shown in FIG.

The basic structure of the optical system of the image pickup apparatus 1 is shown in FIG. That is, the imaging device 1 has two lenses 11 and 12 facing an object (traverser) in order to form two viewpoints.
Is provided, and a CCD as an image pickup device whose optical axes match behind the two lenses 11 and 12 is provided.
(Charge Coupled Device) 13 and 14 are provided.

Here, "d" in the figure is the lenses 11, 12
Between the object and the object [m], “f” is the focal length [mm] of the lenses 11 and 12, and “a” and “b” are points from the object to the CCDs 13 and 14 via the lenses 11 and 12, respectively. Distance from optical axis [mm], “m” is camera base line length [c
m]. In such a case, the distance d [m] can be calculated by d = fm · (a + b). That is, since the base line length m and the focal length f are known, the distance d to the object can be calculated if (a + b) is obtained. It should be noted that (a + b) is obtained as the best matching shift amount when comparing the luminance values of the objects in the left and right images while shifting them little by little.

Further, when (a + b) in the above equation is replaced with the parallax s represented by the number of dots and the cell size c [mm / dot] of the CCDs 13 and 14, the following equation (1) is obtained.

[0017]

[Equation 1] On the other hand, in FIG. 1, the image information obtained by the imaging device 1 is stored in the right camera image storage memory M1 and the left camera image storage memory M2. Data stored in both memories M1 and M2 are input to the distance calculation circuit 2. In the distance calculation circuit 2, an image is divided into meshes with vertical 8 dots and horizontal 8 dots as a unit area, and the parallax s [dot] is calculated by correlation calculation using the right image data as standard data and the left image data as reference data. To be done. Then, by substituting the calculated parallax s into the equation (1), the distance distribution in the image can be obtained. This distance distribution data is stored in the distance data storage memory M3.

Further, the analysis data calculation circuit 3 takes in the image data of the right camera stored in the right camera image storage memory M1 and sums the data of 8 dots in the vertical direction of the unit area (this is the analysis data). Is calculated).
Then, the latest analysis data calculated by the analysis data calculation circuit 3 is stored in the cycle analysis data storage memory M4 as "cycle analysis data". At this time,
Since the analysis data is stored as the total sum of data for 8 dots, the data capacity is reduced to 1/8. The analysis data calculation circuit 3 executes the above-described analysis data calculation processing in a 66 ms cycle. The analysis data stored in the cycle analysis data storage memory M4 is transferred to the previous cycle analysis data storage memory M5 as "previous cycle analysis data" every 66 ms. Therefore, the memory M4 and M5 store the analysis data of the cycle before and after the interval of 66 ms.

Here, the analysis area by the analysis data calculation circuit 3 is set on the screen as shown in FIG. 4, for example. That is, in the present embodiment, the front 30 of the vehicle is
The purpose is to detect a pedestrian in the range of m to 60 m, and more specifically, the analysis area is set with reference to the pedestrian standing on the ground 50 m ahead. On the screen of FIG. 4, a vertical region having a width of 60 dots extending in the horizontal direction is set as the analysis region.

Further, the movement amount calculation circuit 4 takes in the analysis data stored in the main cycle analysis data storage memory M4 and the previous cycle analysis data storage memory M5, and shifts the image for each unit area between both analysis data. Calculate the amount of crossing movement. Here, in order to reduce the calculation amount,
Search only in the lateral direction (horizontal running line direction). That is,
The maximum range in which the pedestrian will move is examined on the desk, and the search range for correlation calculation is set. Within this search range, the sum of the absolute values of the differences (correlation value S (z) ) Is calculated. Specifically, the correlation value S (z) for the unit area is calculated using the following equation (2).

[0021]

[Equation 2] However, "D0" is the present cycle analysis data, "D1" is the previous cycle analysis data, "x0" is the upper left coordinate of the image (search start point of the correlation calculation), and "w" is the width of the search range in the image [dots]. ], And “z” is a shift amount [dot]. Note that FIG. 5 shows a state in which the pedestrian has moved horizontally by a predetermined amount between the previous cycle and the main cycle.

With respect to the minimum value of the correlation value S (z) obtained as described above, the shift amount at that time corresponds to the moving amount of the pedestrian. At this time, it corresponds to the case where the crossing collides with the crossing when the moving amount of the crossing is “0”, that is, when the shift amount is “0” on the image. This is because a collision occurs when the traveling direction of the host vehicle and the angle of the pedestrian do not change, and when the distance to the pedestrian becomes 0 m. FIG. 6 shows the relationship between the correlation value S (z) and the shift amount when the pedestrian moves horizontally to the right. In such a case, the shift amount = k corresponds to the shift amount. At this time, in order to improve accuracy, an interpolation calculation may be performed using the minimum correlation value and two correlation values adjacent thereto, and the movement amount may be calculated with 1/10 dot accuracy.

In this way, the movement amount calculation circuit 4 uses 6
The amount of movement of the pedestrian is calculated from the correlation calculation of two temporally consecutive images captured every 6 ms. Then, this movement amount is stored in the movement amount data storage memory M6.

The vertical edge number calculation circuit 5 takes in the image data of the right camera stored in the right camera image storage memory M1, and for each detection window partitioned into a predetermined size, the number of vertical edges in the window. To count. Here, the detection window is set so as to correspond to the size of the pedestrian in the analysis area of FIG. In the present embodiment, the detection window is set in an area of 60 dots in the vertical direction × 16 dots in the horizontal direction, as shown in FIG. 7, on the basis of the size of the pedestrian in front of 50 m.

Further, as shown in FIG. 18, the detection windows are arranged such that 8 dots, which is half the width of the detection window, overlaps with the adjacent detection window. In this case, since the number of pixels in the horizontal direction of the image is 512 dots,
127 detection windows are set.

The method of calculating the number of vertical edges will be described below. That is, the image data of this cycle is read from the upper left of the analysis area, and for the point x whose magnitude of change in luminance in the scanning line direction is equal to or greater than a predetermined threshold value, the vertical edge number data storage memory corresponding to that x position. "1" is added to the position of M7. Do this for all analysis areas. Next, regarding the width of the detection window, the sum of the data at the corresponding x positions is calculated from the vertical edge number data storage memory M7, and the number of vertical edges of the detection window is obtained.

Further, the traveling road detection circuit 6 takes in the image data of the right camera stored in the right camera image storage memory M1 and detects the traveling road of the host vehicle. More specifically, the left and right white line lane marks on the road are detected from the brightness information in the image, and the inside of the lane mark is detected as the traveling road.

The microcomputer 7 is a well-known CP.
U, a memory, an input / output circuit, etc., the distance data storage memory M3, the movement amount data storage memory M
6, the vertical edge number data storage memory M7, and the stored data in the traveling road data storage data M8 are read, and the crossing detection processing described later is executed. Then, the microcomputer 7 activates the alarm buzzer 8 according to the detection result of a crossing on the road. A yaw rate sensor 9 is connected to the microcomputer 7, and the microcomputer 7 detects the yawing component of the host vehicle from the output result of the sensor 9.

In the present embodiment, the microcomputer 7 constitutes the crossing candidate window detecting means and the crossing detecting means, and the yaw rate sensor 9 constitutes the yawing detecting means.

Next, a series of crossing detection procedures in this embodiment will be described focusing on the arithmetic processing by the microcomputer 7. 8. FIG. 8 is a flowchart showing a crossing detection routine, which is executed by the microcomputer 7 every 66 ms.

When the routine starts, in step 101, the number of vertical edges is read out from the vertical edge number data storage memory M7 for each detection window, and a window whose number is equal to or more than a set threshold value is a candidate window for a crossing person. (FIG. 9).

Subsequently, in step 102, the distance data is read from the distance data storage memory M3, and if the distance data corresponding to the crossing candidate window exists, the movement amount data corresponding to the area is stored.
Call from 6 and average. Further, the amount of movement of the object (crossing person) in the candidate window is calculated based on the averaged data. At this time, the distance to the object is also held at the same time.

Further, in step 103, the yaw amount of the own vehicle measured in advance is subtracted from the calculated movement value, and this value is set as the absolute object movement amount on the image. That is, since the detected value of the yaw rate sensor 9 is an angle change (deg / sec), this is converted into the amount of fluctuation on the image. When the angle of view is 20 degrees and the horizontal resolution of the image is 512 dots, the value obtained by dividing the resolution by the angle of view 2
The product of 5.6 (dots / deg) and the yaw rate value is the yawing amount on the image. By subtracting this from the previously calculated amount of movement of the object, the yawing component of the host vehicle is excluded from the amount of movement of the object.

On the other hand, since the amount of movement of the object in the horizontal direction calculated as described above is the amount of movement in units of the number of dots on the image, in the following steps 104 and 105, the amount of movement of the object in the real space is calculated. Horizontal movement speed v [km /
h]. That is, in step 104, the moving distance L [m] in the real space is calculated from the horizontal moving amount g [dot] on the image at that time using the following expression (3).

[0035]

(Equation 3) However, in the above formula (3), “c” is the size of the cell of the CCD (mm / dot), “f” is the focal length of the lens [mm], and “d1” is the object before moving and the own vehicle. And [d2] are the distance [m] between the object and the vehicle after the object has moved.

Further, in step 105, the following (4)
Using the formula, the moving distance L [m] in the real space within the 66 ms cycle is converted into the moving speed v [km / h]. This moving speed v [km / h] is held in the memory in the microcomputer 7 as the moving amount of the object.

[0037]

(Equation 4) Next, in step 106, the moving amount (moving speed) of the object
The pedestrian is tentatively detected based on the relationship between and the center of the road. That is,
The stored data in the traveling road data storage memory M8 is retrieved and the center position of the traveling road corresponding to the distance of the object is calculated. Then, based on the center position of the traveling path and the moving direction of the object, the object heading toward the center of the traveling path is provisionally detected as a crosser. At this time, as shown in FIG. 10, the pedestrian is identified as an object heading toward the center of the road in the image, and the utility pole or standing tree is identified as an object heading outward from the center of the road in the image. This temporary detection result is stored in the memory in the microcomputer.

Then, in step 107, it is determined whether or not a crossing person actually exists by performing temporal correspondence processing (hereinafter referred to as tracking processing) on the temporary detection result. The specific contents of the tracking process will be described below.

That is, when the temporary detection results for four cycles are obtained, they are averaged to create tracking data for the tracking process. In this case, if the sum of the weights (previously set) of the processing cycles of the existing data is greater than or equal to the predetermined threshold value, the weights of those data are added to calculate the average value, and the calculated value is used as the tracking data. To do. The details are shown in FIG. When weighting each cycle, “4” in this cycle, “3” in 1 cycle ago, “2” in 2 cycles ago, “1” in 3 cycles ago, and is tracked if the total weight is “6” or more. Calculate the data. as a result,
If the tracking data exists, it will be detected that a crossing person exists in the corresponding detection window. In such a tracking process, even if the data only exists in a scattered manner due to a detection error, it is not necessary to consider the absence of the data due to the detection error, and the crossing person is detected based on the continuity.

Next, a specific tracking process on the tracking data map will be described by dividing it into cases of the tracking processes A to D. The contents of the tracking processes A to D are shown in FIG.
FIG. 13 shows a series of tracking examples illustrated in (d) and reflecting the above tracking processes A to D. The numbers on the map are tracking data (movement amount), and the moving speed of the crossing in the real space [km
/ H]. The negative moving speed represents a speed value of moving outward from the center of the traveling path.

Tracking process A: When the tracking data of the previous cycle and the tracking data of this cycle are in the same detection window or the adjacent detection window, their moving speeds are compared and the difference is within the allowable range (here, 2 km). / H), the data can be traced. That is, FIG.
In (a), the tracking data window is moved to the adjacent window, but this is treated as a movable window.

Tracking process B: Two or three tracking data may correspond to one movable window. In this case, as shown in FIG. 12B, a window having a large number of times of tracking is selected and processed. However, when the number of times of tracking is the same, priority is given to data on the right side in the area of the image on the left side of the center of the traveling road and data on the left side of the area of the right side image (see B ′ in FIG. 13).

Tracking process C: As shown in FIG.
When two or three movable windows correspond to one tracking data, the same number of tracking data are produced by the corresponding number. However, an identification label is attached to each tracking data.

Tracking process D: As shown in FIG.
When there is no movable window, the tracking process is interrupted at that point. The tracking data is stored in the memory in the microcomputer 7, and at the same time, the corresponding distance data [m] and the horizontal position zx [m] from the center of the image are also stored. The storage amount is up to 10 cycles.

The horizontal position zx [m] is calculated from the horizontal position DX [dot] of the object from the center of the image and the distance data as follows. That is, the parallax s [dot] of the object is calculated from the distance from the equation (2). This means that in the distance data, the magnitude [dot] of the parallax s corresponds to the camera base line m [cm]. Means Therefore, the horizontal position zx [m] is calculated from the following equation (5).

[0046]

(Equation 5) If the number of traces of the traced data exceeds 10 cycles, the average moving speed AMV [km / km / km of the crossing person is calculated from the stored data for a total of 10 cycles including the past 9 cycles and this cycle
h] and average horizontal position AX [m] are calculated. The average horizontal position AX [m] is calculated from the sum of the horizontal distance from the center of the image to the crossing (the horizontal position zx) and the horizontal distance from the center of the image to the white lane mark on the road. This relationship is shown in FIG. The distance DL [m] and the road width W [m] of the own vehicle to the travel road at the same distance as the crossing are calculated based on the distance data and the travel road detection data.

Using the above data, step 10 in FIG.
At 8, it is determined whether or not the current traveling state of the host vehicle is dangerous for the crossing detected as described above. In the present embodiment, as an example thereof, the time tm1 [s] until the crossing person reaches the running path, the time tm2 [s] when the crossing person finishes passing through the running path, and the point where the own vehicle is at the same distance as the crossing person are set. Danger determination is performed based on the reached time tc [s].
That is, if tm1 ≦ tc ≦ tm2, it is determined that there is a risk of collision between the host vehicle and the pedestrian, and the microcomputer 7 activates the alarm buzzer 8 in step 109 to warn the driver of the detection of the pedestrian. To do. Also, tc <tm
If 1 or tc> tm2, there is no risk of collision between the own vehicle and the pedestrian, that is, it is safe, and the microcomputer 7 ends this routine as it is.

Here, the time tm1 [s] until the crossing reaches the traveling road is calculated from the following (6) based on the average horizontal position AX [m] and the average moving speed AMV [km / h]. It is supposed to be done.

Tm1 = AX · 3.6 / AMV (6) The time tm2 [s] at which the vehicle finishes passing the traveling road is the time tm1.
It is calculated from the equation (7) in which the time required for passing the traveling road width W [m] is added to [s].

Tm2 = W.3.6 / AMV + tm1 (7) Further, the time tc at which the vehicle reaches the point at the same distance as the pedestrian.
[S] is the road DL [m] and the vehicle speed ACV
It is calculated from the following equation (8) based on [km / h].

Tc = DL.3.6 / ACV (8) According to the embodiment described in detail above, the following effects can be obtained. (A) In short, in the present embodiment, a detection window having a size commensurate with a pedestrian in a vertical region (analysis region) of a captured image is arranged in the horizontal direction without a gap, and the distance data of the front space is detected. If the moving direction of the object in the window (direction of the moving amount data) is toward the center of the road, the presence of a pedestrian on the road is detected. According to the above configuration, it is possible to detect a pedestrian existing on the side of the road, and it is possible to predict that the pedestrian is about to cross the traveling path of the vehicle or jumps out on the traveling path. Further, by detecting a crossing person in units of a detection window, it is possible to avoid omission of detection even for a relatively small object such as a crossing person, and it is possible to improve detection accuracy. As a result, highly safe imaging processing can be realized for the host vehicle.

(B) The traveling road data (including the detection result of the traveling road center) of the own vehicle is extracted from the detection information of the white lane mark, and the presence / absence of a crossing (determination of the degree of danger) is performed from the traveling road data. I did it. With such a configuration, it is possible to realize a more secure image capturing process.

(C) The detection windows are arranged so as to overlap with the adjacent windows by a predetermined amount (for example, half the width).
As a result, as compared with the case where the detection windows are arranged side by side, the detection window corresponding to the pedestrian can be appropriately obtained, and the movement of the pedestrian can be calculated accurately.

(D) The number of vertical edges is calculated for each detection window, and when the number of vertical edges is equal to or larger than a predetermined threshold value, the detection window is detected as a candidate window. Then, for those candidate windows, the crossing on the road is detected from the movement amount data. In other words, focusing on the fact that the density of the number of vertical edges is high in the image of the pedestrian, and detecting the window in which the pedestrian exists based on the number of vertical edges, the detection process can be efficiently performed. At this time, the detected window is set as a “candidate window”, and then the actual crossing person is discriminated from the “candidate window” using the distance distribution data and the movement amount data. Crossing can be detected.

(E) Using the detection result of the crossing object in one cycle as the temporary detection result, the temporally associating process (tracking process) of the temporarily detected crossing object is performed, and the correspondence of a predetermined cycle or more is made. I decided to determine the existence of a crossing. Therefore, the reliability of crossing detection can be further enhanced. In particular, in the present embodiment, a weighted average value for four cycles is calculated as tracking data, and the tracking data is used to perform a crossing detection process. Therefore, detection errors due to various disturbances, nonuniformity of walking speed, and the like can be suppressed. In addition, by performing tracking processing from the first discovery of a crossing and observing its continuity, even if a sudden detection error occurs, the existence of a crossing is accurately estimated and The adverse effect can be reduced.

(E) The yawing component of the own vehicle is excluded from the movement amount data, and the crossing object in the image is specified based on the result of the elimination of the yawing component of the own vehicle. In this case, the precise movement of the crossing person can be known, and a highly accurate detection result can be obtained.

The present invention can be embodied in the following modes other than the above-described modes. (1) In the above embodiment, in order to detect a pedestrian on the road, a detection window adapted to the size of the pedestrian is set (vertical 60 dots × horizontal 16 dots). However, this may be changed arbitrarily. For example, a detection window may be set that is applied to the size of a bicycle that crosses the road.

(2) In the above embodiment, the adjacent detection windows are arranged so as to overlap each other by half the width, but this may be changed. For example, the width 1 of the adjacent detection windows
/ 3 may be arranged in an overlapping manner, or 2/3 of the width may be arranged in an overlapping manner. If there are few overlapping portions, the calculation load can be reduced, and if there are many overlapping portions, detection accuracy improves.

(3) The process of specifying the window in which the pedestrian actually exists may be changed from the temporary detection result of the detection window. For example, when there is a window detected 6 times or more at the time when the temporary detection result for 9 cycles is obtained,
Identify that there is a pedestrian in the window.

(4) In the above embodiment, the distance calculation circuit 2, the analysis data calculation circuit 3, the movement amount calculation circuit 4, the vertical edge number calculation circuit 5 and the traveling road detection circuit 6 are configured by hardware, but All or part of them may be configured by software processing by the microcomputer 7.

(5) In the above embodiment, the stereo image analysis (stereoscopic view) is used to acquire the front distance distribution data. However, the method of acquiring the distance distribution data is not limited to this, and for example, a scanning radar device may be used. This is known as a means for calculating the distance from the round trip time until the light is reflected back to the object and returned, and by scanning (scanning) this light within a predetermined angle in front, the distance distribution data is obtained. Can be obtained.

More specifically, the range of the light scanning angle is set as the angle of view of the image pickup device, and the scanning angle for radiating the light is set and used as described below, which can be used for implementing the present invention. Here, the angle of view of the imaging device is 20 degrees, and the number of horizontal pixels is 51.
When the number of dots is 2, the image is displayed in the horizontal direction 8
A case will be described in which the distance distribution is calculated by dividing the dots into a mesh of 8 dots in the vertical direction. In this case, since the area of 8 dots in the horizontal direction corresponds to 0.313 degrees, the scanning angle is sequentially changed in units of this angle (0.313 degrees) within the range of the angle of view. Then, the obtained distance data is transferred to the memory and called from the CPU when necessary.
In this way, the distance distribution data corresponding to the detection window can be acquired.

[Brief description of drawings]

FIG. 1 is a block configuration diagram showing an outline of a crossing detection system according to an embodiment of the invention.

FIG. 2 is a diagram showing a stereo image by left and right cameras.

FIG. 3 is a schematic diagram showing a basic configuration of an imaging device.

FIG. 4 is a diagram in which an analysis area is set in an image.

FIG. 5 is a diagram showing a state in which a pedestrian moves horizontally rightward by a predetermined amount.

FIG. 6 is a graph showing a relationship between a correlation value and a movement amount.

FIG. 7 is a diagram showing the size of a detection window.

FIG. 8 is a flowchart showing a pedestrian detection routine.

FIG. 9 is a diagram showing a candidate window for a pedestrian in the image.

FIG. 10 is a diagram showing a temporary detection result of a crossing person.

FIG. 11 is a diagram for explaining tracking processing.

FIG. 12 is a diagram for individually explaining an example of tracking processing.

FIG. 13 is a diagram for explaining a series of tracking processes.

FIG. 14 is a diagram showing a positional relationship between the own vehicle and a pedestrian.

FIG. 15 is a diagram for explaining an analysis area in an image.

FIG. 16 is a diagram for explaining a detection window.

FIG. 17 is a diagram showing a state in which a pedestrian is detected in the detection window.

FIG. 18 is a diagram showing an arrangement state of detection windows.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Imaging device, 2 ... Distance calculation circuit as distance distribution measurement means, 4 ... Movement amount calculation circuit as movement amount calculation means, 5
... vertical edge number calculating circuit as vertical edge number calculating means, 6 ... running path detecting circuit as running path data detecting means, 7 ... microcomputer as crossing object candidate window detecting means and crossing object detecting means, 9 ... yawing detection A yaw rate sensor as a means.

Claims (7)

[Claims] 1. A distance distribution measuring means for acquiring distance distribution data of an object in the front space, and a moving amount calculation for calculating a moving amount of the object by acquiring a plurality of image data at different times in the front space by an imaging device. And a detection window corresponding to the size of the crossing object, for setting an analysis region in which a predetermined crossing object exists in the vertical direction of the image with respect to the image data. The windows are arranged in a horizontal direction without any gap, and the direction of the movement amount obtained by the movement amount calculating means for the window in which the distance data of the front space exists in the detection window is toward the center of the traveling path of the host vehicle. At this time, the crossing object detection device is provided with crossing object detection means for detecting the crossing object as if the crossing object exists in the detection window. 2. The crossing detection apparatus according to claim 1, wherein a vertical edge number calculation means for calculating the number of vertical edges in the detection window for each of the detection windows, and a vertical edge number by the vertical edge number calculation means. Is a predetermined threshold value or more, and a crossing object candidate window detection means for detecting the detection window as a candidate window for crossing objects, the crossing object detection means, among the plurality of candidate windows detected Then, when the direction of the movement amount obtained by the movement amount calculation means is toward the center of the traveling path of the own vehicle for the candidate window in which the distance data of the front space exists, it is assumed that a crossing object exists in the detection window. A crossing object detection device characterized by detecting the crossing object. 3. The crossing object detecting device according to claim 1, further comprising a running road data detecting unit that detects a lane mark or the like indicating a running road of the vehicle from the image data. Is a device for detecting a crossing object for detecting the center of a traveling road based on the detection result of the traveling road data detecting means. 4. The crossing detection device according to claim 1, wherein the detection window is sectioned and set to match the size of a crossing. 5. The crossing object detection device according to claim 1, wherein the detection windows are arranged so as to overlap adjacent windows by a predetermined amount. 6. The crossing object detection means performs a time-based process of associating the temporarily detected crossing object with the detection result of the crossing object in one cycle as a tentative detection result, and associates it for a predetermined cycle or more. The crossing detection device according to claim 1, wherein the presence of the crossing is determined when the crossing is performed. 7. A yawing detection means for measuring a yawing component of traveling of the own vehicle, wherein the crossing object detection means removes the yawing component of the own vehicle from the movement amount data obtained by the movement amount calculation means. The crossing detection device according to any one of claims 1 to 6, wherein the crossing is specified based on a result obtained by removing the yawing component of the host vehicle.