JP3505362B2 - Vehicle detection method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vehicle detection method for distinguishing a vehicle from a shadow based on the size of the length of edge components in a thinned edge image.

[0002]

2. Description of the Related Art As a conventional vehicle detection method, a vehicle detection target area, for example, a method of detecting a vehicle from a brightness difference by comparing the road surface brightness of a parking lot or a driving lane with the vehicle brightness, or a captured image There is known a method of detecting a vehicle from the magnitude of the edge brightness of an edge image obtained by processing the.

[0003]

However, in any of the above-mentioned conventional methods, it is difficult to accurately identify and extract a dark vehicle such as black and a shadow of another vehicle. For this reason, many false detections occurred in the evening or at dawn on a sunny day where a long shadow that easily reaches the adjacent parking area or driving lane is likely to appear.

The present invention has been made in order to solve the above problems, paying attention to the complexity of the edge shape of a vehicle, and accurately distinguishing the vehicle and the shadow based on the size of the edge, An object of the present invention is to provide a vehicle detection method that surely extracts only a vehicle.

[0005]

In order to achieve the above object, a vehicle detection method according to claim 1 of the present invention is provided with edge extraction processing means for extracting edges from a photographed image of a vehicle detection target area. Thinning means for converting the line width of each edge in the edge image into a thinned edge image, and the length of the edge component included in the thinned edge image by the thinning processing means. Determination processing means for identifying the vehicle and the shadow based on the above and extracting only the vehicle.

Further, the vehicle detection method according to claim 2,
Edge extraction processing means for extracting edges from a captured image of a vehicle detection target area to obtain a multivalued edge image, and binarization processing for binarizing the obtained multivalued edge image to convert it into a binary edge image. Means, thinning processing means for subjecting the binarized image to thinning processing to obtain an image thinned to a one-pixel width or a prescribed pixel width, and an edge image thinned by the thinning processing means Determination processing means for discriminating between the vehicle and the shadow based on the magnitude of the length of the edge component being present and extracting only the vehicle.

Further, the vehicle detection method according to claim 3,
The vehicle detection method according to claim 1 or 2, wherein the determination processing unit measures an edge component length for each vehicle detection area in the thinned edge image,
The vehicle is extracted by comparing the edge length obtained by normalizing the length of the obtained edge component with the area of the detection area with a predetermined threshold value.

A vehicle detection method according to a fourth aspect is the vehicle detection method according to the second aspect, wherein an edge within the detection area is detected for each vehicle detection area from the multivalued edge image before thinning. It is characterized in that the average luminance value of the pixels forming the above is obtained, and the vehicle is extracted using both the average luminance value of the edge and the normalized edge length.

Further, the vehicle detection method according to claim 5 is
The vehicle detection method according to any one of claims 1 to 4, wherein the vehicle detection target region is a parking lot, and the vehicle detection area is an individual parking area in the parking lot. It is a thing.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION Before describing specific embodiments of the present invention, the detection principle of the present invention will be described first.

Usually, a vehicle is made up of a large number of parts having complicated shapes such as a body, a window, a headlight, a backlight, a side mirror, a tire, a bumper, etc., so that an extremely complicated edge group is formed. Therefore, when the captured image is converted into the edge image, the edge component of the vehicle, that is, the number of pixels of the edge becomes a very large value. On the other hand, the shadow due to sunlight or the like almost always has an extremely simple shape because the contour line of the object only appears as an edge. Therefore, the edge component of the shadow has an extremely shorter length than that of the vehicle. In view of this, the present invention focuses on the difference in the length of the edge component of the vehicle and the shadow and compares the length of the edge component with a threshold value to accurately extract only the vehicle.

That is, an image obtained by taking a bird's-eye view of a vehicle detection target area such as a parking lot is subjected to edge extraction processing and binarization processing to convert it into a binary edge image, and then to this binary edge image thinning processing. Is performed to convert the edge image into a thin line having a one-pixel width or a specified pixel width.

It should be noted that the reason for converting a photographed image into an edge image thinned to a one-pixel width or a specified pixel width instead of a simple edge image is that when the thinned one-pixel width or the specified pixel width is used, Since the width of the line is constant, if the number of pixels forming an edge is counted, that value gives the length of the edge component as it is. On the other hand, if the captured image is simply converted to an edge image, the edge width of each part may become wider or narrower depending on the captured shadow and the state of the vehicle, and the number of pixels that make up the edge may be reduced. This is because the length of the edge component cannot be measured even if it is counted.

Next, using the edge image subjected to the thinning process, the length of the edge component included in the detection area is measured for each predetermined vehicle detection area in the image, and the length of the edge component is calculated. Edge length normalized by the area of the detection area, that is, (edge length) / (detection area area)
To calculate.

Then, by comparing the edge length normalized by the area of the detection area with a preset threshold value, the vehicle is extracted based on its size.

As described above, in the method of the present invention, the photographed image of the vehicle detection target area is converted into a thinned edge image, and the length of the edge component included in the thinned edge image is reduced. Since the vehicle and shadow are identified based on the vehicle and only the vehicle is extracted, it is possible to accurately identify the vehicle and shadow, and in the morning on a sunny day where a long shadow that easily reaches the adjacent parking area or driving lane Even in the evening, the shadow portion will not be erroneously detected as a vehicle.

If necessary, the average brightness value of the pixels forming the edge in the detection area is calculated for each detection area from the multi-valued edge image before the thinning process, and the average brightness value and the regular brightness value are obtained. The vehicle may be extracted using both the converted edge length. As a result, vehicle detection with higher accuracy can be realized.

Next, an embodiment of the method of the present invention constructed on the basis of the above-described detection principle will be specifically described with reference to the drawings.

FIG. 1 shows an embodiment in which the above-described vehicle detection method according to the present invention is applied to a parked vehicle detection device in a parking lot. Reference numeral 1 denotes an ITV camera constituting a photographing means, The ITV camera 1 takes a bird's-eye view of a parking lot, which is a vehicle detection target area, from a predetermined height position. Reference numeral 2 denotes an AD conversion unit for converting a captured image of a parking lot captured by the ITV camera 1 from an analog image signal to a digital image signal, and 3 is a captured image memory including a RAM for storing the AD converted captured image. is there.

An edge extraction processing unit 4 performs edge extraction processing on the photographed image of the parking lot stored in the photographed image memory 3 to extract only the edge component in the photographed image to obtain a multi-valued edge. It is a circuit that takes out as an image. 5
Is an edge image memory including a RAM for storing the multi-valued edge image extracted by the edge extraction processing unit 4.

Reference numeral 6 denotes a binarization processing unit which performs binarization processing on the multi-valued edge image stored in the edge image memory 5 and converts it into a binary edge image composed of black and white, for example. Circuit. Reference numeral 7 is a binary image memory including a RAM for storing the converted binary edge image.

A thinning processing unit 8 is a circuit for thinning the line width of each edge in the binary edge image stored in the binary image memory 7 to, for example, one pixel width. Reference numeral 9 denotes a thinned image memory including a RAM for storing the thinned edge image.

A determination processing unit 10 obtains the length of the edge component included in the parking area for each parking area from the thinned edge image stored in the thinned image memory 9. Using the obtained length of the edge component, the edge length normalized by the parking area area, (edge length) / (parking area area) is calculated, and the multi-value stored in the edge image memory 5 is calculated. From each edge area, the average brightness value of the pixels forming the edge in the detection area is calculated for each detection area, and the vehicle exists in each parking area using both the average brightness value and the normalized edge length. It is a circuit for determining whether or not.

The parked vehicle detection device having the above-described configuration may be configured as hardware by a dedicated processing circuit, or may be configured as software by using a processing program and a personal computer. Is.

Next, the operation of the embodiment having the above configuration will be described with reference to the flowchart of FIG. 2 and the processed image examples of FIGS.

First, the ITV camera 1 takes a bird's-eye view of the entire parking lot to be processed from a predetermined height position (step S1 in FIG. 2). This captured image is stored in the captured image memory 3 after being converted from an analog image signal to a digital image signal in the AD conversion unit 2. An example of this photographed image is shown in FIG. In FIG. 3, each of the parking areas is enclosed by a rectangular white frame.

In FIG. 3, the parking lot is photographed in the morning, and the shadow of the parked vehicle is 2 because the altitude of the sun is low.
You can see that it extends to the next parking area. In the conventional method, it is erroneously detected that the vehicle is parked in the parking area where the vehicle does not exist due to the long shadow.

Next, the edge extraction processing section 4 reads out the photographed image of the parking lot from the photographed image memory 3 and performs edge extraction processing (for example, Laplacian filter, first derivative filter, Sobel filter, Roberts filter, pre-filter). Vit filter, color difference processing, etc.) are applied to extract only the edge components in the image to obtain a multivalued edge extracted image (step S2). This multivalued edge image is stored in the edge image memory 5. An example of this multi-valued edge image is shown in FIG. As shown in FIG. 4, by performing edge extraction processing, the body, window, headlights,
Backlights, side mirrors, tires, bumpers, etc.
The outlines of many parts having complex shapes are extracted as edges, and the shadow of the vehicle elongated by the sunrise or sunset is also extracted as edge components.

Next, the binarization processing unit 4 reads the multivalued edge image from the edge image memory 5 and performs the binarization process to convert the multivalued edge image into, for example, a black and white binary edge image. (Step S3). This binary edge image is stored in the binary image memory 7. This binarization process is a pre-process for the subsequent edge thinning process. An example of this binary edge image is shown in FIG. As shown in FIG. 5, the multi-value edge image (FIG. 4) is a binary image in which the edge portion is black.

Next, the thinning processing unit 8 reads the binary edge image from the binary image memory 7 and performs thinning processing so that the line width of each edge in the binary image has a predetermined width. , For example, a thin line having a width of 1 pixel is converted (step S4). The thinned edge image is stored in the thinned image memory 9. This thinning processing is necessary in order to accurately measure the line length of the edge without being influenced by the line width of the edge by counting only the black pixels forming the edge in the next determination processing unit 10. Processing.

In the thinning process, the pixels forming the connected component are sequentially erased from the erasable pixels of the contour portion by sequential processing, and the pixels that are equidistant from the contour portion are extracted. A processing method called a “waveform propagation method” or a known processing method called a “mask pattern method” that erases unnecessary pixels by sequentially applying a mask pattern indicating erasable pixels to an image is adopted. It can be realized by

An example of this thinned image is shown in FIG. As shown in FIG. 6, many complicated edges appear in the vehicle part, while only one smooth edge appears in the shadow part, and most of the empty parts appear in the empty part. It can be seen that no edge appears. Therefore, by utilizing the feature of the appearance of the edge, the vehicle and the shadow can be discriminated as described later, and only the vehicle can be accurately extracted.

Next, the determination processing unit 10 performs the following evaluation amount calculation processing (step S5), and uses this evaluation amount to determine whether or not a vehicle exists in each parking area. Processing is performed (step S6).

(1) Evaluation amount calculation processing (step S5) (a) The measurement determination processing unit 10 for measuring the length of the edge component included in each parking area is made into a thin line stored in the thinned image memory 9. The edge image (see FIG. 6) is read out, and for each parking area in the thinned edge image, the total length of the edge components included in the parking area is measured. To measure the length of this edge component, the number of black pixels (see FIG. 6) indicating the edge component may be counted. This is because when the line width of the edge is converted to 1 pixel, the total number of black pixels in each parking area gives the total length of the edge components existing in the parking area itself.

(B) Normalization of Length of Edge Component by Parking Area Area Next, the determination processing unit 10 normalizes the length of the edge component obtained in (a) above by the area of each parking area. Convert to edge length. The normalized edge length may be calculated by the following formula. When measuring the length of the edge component by adding the distance between the centers of the pixels (edge length) / (parking area area)

(C) Calculation of Average Luminance Value of Edge of Each Parking Area Further, in the case of the above-mentioned embodiment, the determination processing unit 10 causes the edge image memory 5 to perform more reliable vehicle determination.
The multi-valued edge image (see FIG. 4) is read out from the multi-valued edge image and the average brightness value of the pixels forming the edge in the parking area is calculated for each parking area from the multi-valued edge image.

(2) Parking / empty vehicle determination processing (step S
6) The determination processing unit 10 uses the normalized edge length obtained as described above and the average brightness value of the edge to compare the presence or absence of a vehicle by comparing each with a preset threshold value. Make a decision. That is, the normalized edge length and the average luminance value of the edges include the following (I) to (II
Since there is a property such as I), for each of the normalized edge length and the average brightness value of the edge, if an optimal threshold value is set in advance, from the size of the edge length and the average brightness value, It is possible to accurately determine whether or not there is a vehicle in the parking area.

(I) When the vehicle exists in the parking area, the vehicle is composed of parts having a complicated shape, so that the length of the edge component is long, the value of the normalized edge length is large, and the edge component is large. Since it itself is brighter than the shaded area, the average brightness of the edges also becomes large. (II) When there is a shadow in the parking area, only the outline of the shadow appears as an edge, so the value of the normalized edge length does not become so large, and the average brightness of the edge is not so large. (III) When there are no vehicles or shadows in the parking area, there are no objects including shadows, so the normalized edge length value is almost 0 and the average edge brightness is also almost zero. It approaches 0.

FIG. 7 shows the parking / vacant judgment result of the thinned edge image (FIG. 6) obtained as described above. The horizontal axis is the average luminance value of the edges, and the vertical axis is the normalized edge length = (edge length) / (parking area area). In the case of the example in FIG. 7, 8% is set as the threshold value for determining the normalized edge length on the vertical axis, and 68/2 is set as the threshold value for determining the average luminance value of the edges on the horizontal axis.
56 steps (when 8 bits = 256 steps are displayed)
Is adopted.

As is apparent from FIG. 7, a higher determination accuracy can be obtained by performing the vehicle determination in the vertical axis direction (normalized edge length) than in the horizontal axis direction (average luminance value of edges). Further, when the vehicle is determined using both the normalized edge length on the vertical axis and the average brightness value of the edges on the horizontal axis, the vehicle exists in the area (A) that satisfies both thresholds. Since it is determined as a parking area for parking, higher determination accuracy can be obtained.

FIG. 8 shows an example of the judgment processing image in which the judgment result of FIG. 7 is displayed in an overlapping manner on the original photographed image (FIG. 3). In FIG. 8, a portion surrounded by a white frame is a parking area determined to have a vehicle, and a portion surrounded by a black frame is a parking area determined not having a vehicle. As is clear from FIG. 8, the correct determination result is shown in most of the parking areas, and it can be seen that no erroneous determination is made even if the shadow of the vehicle extends long in the parking area.

Although the above-described embodiment has been described with respect to the case where the present invention is applied to the parked vehicle detection device in the parking lot, the present invention is not limited to the detection of the parked vehicle. Set the vehicle detection area on the line,
The present invention can be applied to various fields such as distinguishing a vehicle traveling in the detection area from a shadow.

[0043]

As described above, according to the first aspect of the present invention.
According to the vehicle detection method described, the captured image of the vehicle detection target area is converted into a thinned edge image, and the vehicle and the shadow are shaded based on the magnitude of the length of the edge component included in the thinned edge image. Since the vehicle is identified and only the vehicle is extracted, the vehicle can be accurately detected even in the evening or at dawn on a sunny day where a long shadow that tends to reach the adjacent parking area or the driving lane is likely to appear.

According to the vehicle detection method of the second aspect of the present invention, the captured image of the vehicle detection target area is converted into a multi-valued edge image and then binarized to be converted into a binary edge image. Since the binary edge image is subjected to a thinning process to obtain an edge image thinned to a one-pixel width or a specified pixel width, the thinned edge image necessary for realizing the method of the present invention is obtained. Can be obtained easily and surely.

According to the vehicle detection method of claim 3 of the present invention, the length of the edge component is measured for each vehicle detection area in the thinned edge image, and the obtained edge component is measured. Is normalized by the detection area area, and the vehicle is extracted by comparing the normalized edge length with a predetermined threshold value. Therefore, stable vehicle detection can be realized.

According to the vehicle detection method of the fourth aspect of the present invention, the average of the pixels forming the edge in the detection area is averaged for each vehicle detection area from the multivalued edge image before thinning. Since the brightness value is obtained and the vehicle is extracted using both the average brightness value of the edge and the normalized edge length, more accurate vehicle detection can be realized.

According to the vehicle detection method of the fifth aspect of the present invention, the vehicle detection target area is a parking lot,
Moreover, since the vehicle detection area is an individual parking area in the parking lot, the vehicle parked in the parking area can be accurately detected without being mistaken for a shadow, and the vehicle management of the parking lot can be performed more accurately. It can be done efficiently.

[Brief description of drawings]

FIG. 1 is a block circuit diagram showing an embodiment in which the method of the present invention is applied to a parked vehicle detection device in a parking lot.

FIG. 2 is a flowchart of a processing operation of the above embodiment.

FIG. 3 is a diagram showing an example of a captured image of a parking lot taken from above with the ITV camera shown in FIG. 1;

FIG. 4 is a diagram showing an example of a multivalued edge image obtained by an edge extraction processing unit in FIG.

5 is a diagram showing an example of a binary edge image obtained by a binarization processing unit in FIG.

6 is a diagram showing an example of a thinned edge image obtained by a thinning processing unit in FIG.

FIG. 7 is a graph showing an example of parking / empty vehicle determination processing in a determination processing unit in FIG. 1.

FIG. 8 is a diagram showing an example of a determination processing image in which the determination result of FIG. 7 is superimposed and displayed on the captured image of FIG.

[Explanation of symbols]

1 ITV camera (shooting means) 2 AD converter 3 Captured image memory 4 Edge extraction processing unit 5 Edge image memory 6 Binarization processing unit 7 Binary image memory 8 Thinning processing section 9 Thin line image memory 10 Judgment processing unit

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yuuo Horiba 50-2 Shinbayashi, Higashisakai-cho, Kariya city, Aichi prefecture (56) References JP-A-9-83996 (JP, A) JP-A-6-331335 (JP, A) JP 9-62842 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G08G 1/00-9/02 G01C 21/00-21/24 G06T 1/00

Claims (5)

(57) [Claims] 1. An edge extraction processing means for extracting edges from a captured image of a vehicle detection target area, and a thinning processing means for converting a line width of each edge in the obtained edge image into a thinned edge image. A vehicle detection method comprising: a determination processing unit that distinguishes between a vehicle and a shadow based on the magnitude of the length of an edge component included in the edge image thinned by the thinning processing unit and extracts only the vehicle. 2. An edge extraction processing means for extracting edges from a captured image of a vehicle detection target area to obtain a multivalued edge image, and binarizing the obtained multivalued edge image to convert it into a binary edge image. Binarizing processing means, thinning processing means for performing thinning processing on the binarized image to obtain an image thinned to one pixel width or a prescribed pixel width, and thinning processing performed by the thinning processing means. And a determination processing unit that identifies a vehicle and a shadow based on the size of the length of the edge component included in the edge image and extracts only the vehicle. 3. The determination processing means measures the length of an edge component for each vehicle detection area in the thinned edge image, and the obtained length of the edge component is the area of the detection area. The vehicle detection method according to claim 1 or 2, wherein the vehicle is extracted by comparing the normalized edge length with a predetermined threshold value. 4. An average brightness value of pixels forming an edge in each vehicle detection area is calculated for each vehicle detection area from the multi-valued edge image before thinning, and the average brightness value of the edge and the normalization are performed. The vehicle detection method according to claim 2, wherein the vehicle is extracted using both of the edge lengths. 5. The vehicle detection target area is a parking lot,
The vehicle detection method according to any one of claims 1 to 4, wherein the vehicle detection area is an individual parking area in the parking lot.