JP2003217081A - Vehicle detecting method and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a vehicle detecting method for shortening time required for comparing-checkup processing (matching processing). <P>SOLUTION: This device has (a) a step of storing template data obtained as a clustering result by clustering first characteristic quantity data for indicating a characteristic quantity of a contour line for indicating at least a part of a shape of a vehicle, (b) a step of forming a binarized image for expressing the contour line of the traveling vehicle by imaging and binarizing an image of the traveling vehicle, (c) a step of extracting second characteristic quantity data for indicating the characteristic quantity of the contour line of the binarized image, and (d) a step of determining that the vehicle is included in the binarized image on the basis of a comparing result by comparing the template data with the second characteristic quantity data. <P>COPYRIGHT: (C)2003,JPO

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 and apparatus capable of accurately recognizing a running vehicle and accurately detecting its position.

[0002]

2. Description of the Related Art It is necessary to identify a moving vehicle and identify its position, for example, in order to identify a vehicle that violates the speed limit or a vehicle loaded with dangerous goods whose entry into a long tunnel is prohibited. Is. That is, in order to identify these vehicles, the vehicle number on the license plate of these speeding vehicles and the letters on the plate such as "danger,""poison," and "play" posted on vehicles carrying dangerous goods are used. It is necessary to accurately detect the position that can be accurately read, but in order to do so, first the vehicle itself is accurately detected, and the position of the vehicle on the road is detected, and It is necessary to accurately predict the timing at which the plate that indicates that there is is readable.

In vehicle detection processing by moving image processing to which pattern recognition technology is applied, a large number of vehicle shape patterns (templates) to be detected are registered, and whether or not the vehicle is a vehicle is determined based on the degree of similarity with the registered pattern.

[0004]

In the moving image processing to which the pattern recognition technique described above is applied, if the number of patterns to be subjected to the matching processing is large, it takes a long time for the detection processing.
On the other hand, if the number of registered patterns is small, sufficient detection accuracy may not be obtained.

SUMMARY OF THE INVENTION An object of the present invention is to provide a vehicle detection method and apparatus that can reduce the time required for comparison / collation processing (matching processing). Another object of the present invention is to provide a vehicle detection method and device that can reduce the time required for comparison / collation processing without lowering detection accuracy.

Japanese Patent No. 2893814 discloses that
The following plate cutting device in the vehicle number automatic reading device is disclosed. That is, in a vehicle number automatic reading device for automatically identifying the plate number of a vehicle running or stopped on the road, a photographing means for photographing the road and a plate frame candidate determining means for judging a plate frame candidate from the photographed image. A character area candidate determination means for determining a character area candidate from a captured image; a portion surrounded by a plate frame estimated based on the plate frame candidate; and a character area estimated based on the character area candidate. And a plate position determining means for determining the plate position on the basis of the degree of overlap.

[0007]

According to the vehicle detection method of the present invention, (a) clustering is performed on first feature amount data indicating a feature amount of a contour line showing at least a part of the shape of a vehicle, and Storing template data obtained as a result of clustering; (b) capturing an image of a traveling vehicle and binarizing the image to generate a binarized image in which a contour line of the traveling vehicle is represented; c)
A step of extracting second characteristic amount data indicating a characteristic amount of the contour line of the binarized image; and (d) comparing the template data and the second characteristic amount data, and a result of the comparison. And determining that the vehicle is included in the binarized image.

In the vehicle detection method of the present invention, in (a), the center of each cluster obtained as a result of the clustering is stored as the template data.

In the vehicle detection method of the present invention, the first
The feature amount of the contour line shown in each of the second feature amount data and the second feature amount data is represented by a vector amount of the direction and size of the contour line.

In the vehicle detection method of the present invention, in (a), the average vector of each cluster obtained as a result of the clustering is stored as the template data.

In the vehicle detection method of the present invention, (a) stores the average vector, eigenvalue and eigenvector of each cluster obtained as a result of the clustering as the template data.

In the vehicle detection method of the present invention, in (a), either the minimax method or the k-mean method is used as the clustering method.

In the vehicle detection method of the present invention, as the template data, the first parameter indicating a feature amount of a contour line showing at least a part of the shape of the vehicle obtained by executing the vehicle detection method at a plurality of locations. The feature amount data is subjected to clustering, and the template data obtained as a result of the clustering is stored.

In the vehicle detection method of the present invention, the second
The feature amount data of indicates the feature amount of the contour line included in a partial region of the cut out binarized image.

The vehicle detection device of the present invention performs clustering on the first feature amount data indicating the feature amount of the contour line showing at least a part of the shape of the vehicle, and the template data obtained as a result of the clustering. A template data generation and storage unit that stores a binary image, a binarized image generation unit that captures an image of a traveling vehicle and binarizes the image, and generates a binarized image in which a contour line of the traveling vehicle is represented; A feature amount extraction unit that extracts second feature amount data indicating the feature amount of the contour of the image, the template data, and the second feature amount data are compared, and based on the result of the comparison, The binarized image includes a determination unit that determines that the vehicle is included.

[0016]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of a vehicle detection method of the present invention will be described with reference to the accompanying drawings.

FIG. 1 is a schematic block diagram of an apparatus showing this embodiment, FIGS. 2 and 3 are flow charts of a vehicle detection method of this embodiment, and FIG. 4 is a pattern feature amount on a screen imaged by an image pickup apparatus. It is a figure for demonstrating the rectangular area used for calculation.

In FIG. 1, reference numeral 1 is a road, 2 is a passing vehicle, 3 is a gantry for installing a light source 5 such as a stroboscope for illuminating the vehicle on the road, 4 is a vehicle detection image pickup device, and 6
Is a position on the road 1 that is optimal for capturing an image of the license plate of the vehicle 2.

Reference numeral 7 is an analog / digital conversion circuit for analog / digital converting the signal from the image pickup device 4, reference numeral 8 is a data smoothing circuit for removing noise,
Reference numeral 9 is an edge enhancement circuit that enhances the edges of the image by subjecting the input image to differential processing using a Sobel differential filter or the like, and reference numeral 10 is a binarization / storage circuit that binarizes and stores the enhanced image.

Further, reference numeral 11 shows whether the contour line of the binarized image has a component in which direction, that is, a horizontal component, a vertical component, a 45 ° component, and the like. Is a first pattern feature amount extraction circuit for extracting a pattern feature amount expressed by a vector amount of the direction and size of the contour line.

Further, reference numeral 12 compares the vehicle shape template of the front shape of the vehicle stored in the vehicle shape template storage device 14 with the pattern feature quantity extracted by the first pattern feature quantity extraction circuit 11 to compare both of them. This is a comparison / collation circuit that calculates the degree of similarity.

Reference numeral 13 is a threshold value for checking whether the similarity of the pattern feature quantity calculated by the comparison / collation circuit 12 is equal to or more than a certain threshold value.

Further, reference numeral 14 expresses a contour line direction pattern feature amount in the front shape of the vehicle to be detected by a vector amount of the direction and size of the contour line, stores a certain number of vehicle types as a template, and stores the other types. It is a vehicle shape template storage device that statistically processes and deforms a pattern by a known method to process it so that pattern matching can be performed.

Reference numeral 71 is a vehicle shape pattern database in which image data of a plurality of vehicle shapes are stored.

Reference numeral 72 indicates, for each of a plurality of vehicle shape images stored in the vehicle shape pattern database 71, in which direction and to what extent the contour line of the image has components, that is, in the horizontal and vertical directions. Ingredient, 45
This is a second pattern feature quantity extraction unit that looks at how much each direction component is present and extracts the pattern feature quantity expressed by the vector quantity of the direction and size of the contour line.

Reference numeral 73 is a clustering unit for performing clustering, which is a statistical method for classifying the plurality of pattern feature amounts given from the second pattern feature amount extraction unit 72 into a plurality of groups (clusters). The clustering unit 73 uses the minimax method (maximum distance algorithm) and the k-means method (km) as its clustering method.
Any of known methods such as eans) can be used. The clustering method is described, for example, in "Pattern Information Processing" (Seiji Nakagawa, Maruzen), pp. 36, 69-73.

As a result of the clustering by the clustering unit 73, the center (center of gravity) of each cluster is obtained. Vehicle shapes are classified into a plurality of clusters, and each vehicle shape is represented by the center of gravity of the cluster. Here, the obtained center of each cluster is stored in the vehicle shape template storage device 14 as a template (matching pattern data).

Here, the representative value of each class obtained by the clustering unit 73 and registered in the vehicle shape template storage device 14 as a template is an average vector of pattern feature amounts (vehicle shape feature pattern vector) of each class. You can Alternatively, the vehicle shape template storage device 14 obtained by the clustering unit 73
The representative value of each class registered as a template in
The average vector + eigenvalue + eigenvector often used in the pattern recognition technique can be used.

Here, the reason why the grouping method is performed by the second pattern feature quantity extraction unit 72 and the clustering unit 73 as described above is as follows.

In the moving image processing to which the pattern recognition technique is applied, it is effective to suppress the number of patterns to be subjected to the matching processing (hereinafter, matching pattern) in order to reduce the detection processing time. In order to reduce the number of matching patterns, a plurality of vehicle shapes are grouped by similar shapes, and an average vehicle shape (template) of the plurality of vehicle shapes classified into each group is calculated,
Only the average vehicle shape is registered as a matching pattern.

However, for example, when a group in which large vehicles, ordinary vehicles, and light vehicles coexist is formed, even if an average vehicle shape is obtained in the group, matching pattern data that is far from the shape of the existing vehicle is obtained. However, the shape may be meaningless, which may cause a false detection. If the number of matching pattern data is increased to avoid the erroneous detection, the number of matching increases and the calculation load increases.

Therefore, accurate grouping is performed so that meaningful matching pattern data (matching pattern data that enables more accurate vehicle detection with a smaller number of matching times) is obtained as an average value (template) within the group. Is required to do.

For example, when character recognition is performed by pattern recognition technology, the number of characters to be recognized is at most a small number such as "0" to "9" or "A" to "Z", and each character is to be recognized. It is possible to clearly distinguish it from other characters by the feature. Therefore, when registering the matching pattern data in the database, refer to the image data whose shape is shown and classify each image data into a class (group).
There is no problem even if a person performs the work of classifying into.

On the other hand, when the pattern recognition of the vehicle shape is performed, when a person refers to the image data showing each vehicle shape, the person classifies each vehicle shape data into each class (group). , The result is significantly different from the group in which the computer that performs the matching process determines that the shapes are “similar”. As a result, the average vehicle shape (template) of the group classified by the person becomes a value that does not have much meaning in the pattern matching performed by the computer. In the case of vehicle shape, for example, even if it is the same ordinary car, the vehicle width, vehicle height, etc. may differ slightly depending on each vehicle type,
Calculators tend to emphasize slight differences in vehicle type and vehicle width when determining the similarity of shapes, whereas
One of the reasons is that people tend not to attach much importance.

From the above, the pattern feature amount (first pattern) to be compared and collated in the comparison / collation circuit 12 is not grouped with similar items based on the vehicle shape image. With the pattern feature amount extracted by the feature amount extraction circuit 11 as a reference, similar items are grouped so as to be classified and the center thereof is used as a template, whereby meaningful matching pattern data can be obtained.

In the comparison / collation circuit 12, the template stored in the vehicle shape template storage device 14 and the first template are stored.
The pattern feature amount extraction circuit 11 calculates the similarity of the pattern feature amounts. The comparison / matching circuit 12 performs template matching as a class identification method. The vehicle shape template storage device 14 is provided with a basic (representative) pattern (center) of each class as a template. The comparison / collation circuit 12 calculates the similarity between the data to be identified (the pattern feature amount extracted by the first pattern feature amount extraction circuit 11) and the template of each class.

In the present embodiment, since the matching pattern data registered in the vehicle shape template storage device 14 is set as described above, the number of matching pattern data to be registered is small and the time required for the matching process is short. . In addition, the vehicle shape template storage device 1
Since meaningful matching pattern data obtained as a result of the above clustering is registered in 4,
Vehicle detection can be performed without lowering detection accuracy.

Further, reference numeral 16 indicates that when a vehicle is detected, the vehicle is license plate, "danger", "poison",
A signal is sent to the light source light emission control device 17 that controls the light source 5 such as a strobe to emit light when the optimum timing (position) for reading a plate such as "play" is reached, and the signal is obtained. It is a character recognition control circuit that performs control to receive image pickup data from the binarization / storage device 10.

In the present embodiment, for example, a plurality of image pickup devices 4 and a light source 5 such as a strobe are installed on a gantry 3 provided across the road 1, and the plurality of image pickup devices 4 cover the entire width of the road 1. A vehicle traveling at any position on the road 1 can be captured by capturing an image.

When the vehicle 2 is detected by the vehicle detection device of the present embodiment, the vehicle 2 picks up an image pickup position 6 on the road 1 (the image pickup position 6 in the figure is a temporary frame shown for convenience of explanation. ), The light source 5 such as a strobe is turned on to recognize the characters on the license plate of the vehicle 2 or on the plates such as "danger,""poison," and "drama" posted on vehicles carrying dangerous goods. Recognize the character.

The image pickup device 4 and the light source 5 such as a strobe
Was installed on the gantry 3, but it is obvious that this may be installed on a pillar standing on the roadside.

The image pickup device 4 detects the vehicle 2 on the road 1 by continuously picking up the image of the road 1, and the detection is indicated by the bold line in the picked-up image 30 (one frame) shown in FIG. A rectangular area 32 for pattern feature calculation, which is large enough to cover the front surface of the vehicle represented by a plurality of squares in the detection target area 31, is cut out, and the image data therein is stored in the vehicle shape template storage device 14. This is done by comparing with the front pattern template of the vehicle. Image data captured by the image capturing device 4 and the vehicle shape template storage device 1
The pattern matching with the front surface pattern template of No. 4 can be performed by a known method.

The cutting out of the pattern feature amount calculation rectangular area 32 is performed by cutting the pattern feature amount calculation rectangular area 32 in the vehicle detection target area 31 in the lateral (X) direction 33 and the vertical (y) direction.
The vehicle is detected in each of the cut out regions by moving one unit in the direction 34, for example, to the position indicated by (n, m).

Then, this operation is repeated for each frame of the image pickup screen, and when the vehicle 2 is captured at the position of the pattern feature amount calculation rectangular area 32 shown in FIG. 5, vehicle detection is performed at that position. Is determined as.

This determination method will be described with reference to the flow charts shown in FIGS. 2 and 3 and the block diagram of the vehicle detection device of this embodiment shown in FIG. FIG. 2 and FIG. 3 have a relationship of being connected at the two-dot chain line.

First, in step S21, the pattern feature quantity calculation rectangular area 32 is set to the origin position (n = 0, m = 0) in the vehicle detection target area 31, so that x = 0 and y = 0.

Next, as shown in step S22, one frame 30 of the image pickup screen from the image pickup device 4 is read, analog / digital conversion is performed by the analog / digital conversion circuit 7, and then smoothing is performed as shown in step S23. The noise component is removed by the smoothing circuit 8 by a known smoothing method.

Next, as shown in step S24, the image within the vehicle detection target area 31 of the image pickup screen (one frame) 30 is edge-enhanced by the edge-enhancing circuit 9 using a known Sobel differential filter or the like. Further, as shown in step S25, the edge-enhanced image is binarized by the binarization / storage circuit 10 and stored. The edge emphasizing circuit 9 extracts only the contour of the vehicle by performing two-dimensional differentiation on the image in which the vehicle or the like is shown. The binarization / storage circuit 10 further binarizes the contour of the vehicle.

In the next step S26, the rectangular area 32 for calculating the pattern feature quantity is cut out at the origin position of the vehicle detection target area 31 of FIG. 4 in which the vehicle candidate is detected. next,
In step S27, the first pattern feature quantity extraction circuit 1
The feature amount of the pattern in the pattern feature amount calculation rectangular area 32 is extracted by 1. This is because, in each direction, each of the contour lines of the pattern in which the edge in the pattern feature amount calculation rectangular area 32 is emphasized has a component,
That is, the pattern feature amount is calculated by observing the vector amount of the direction and size of the contour line. For example, how large are each of the horizontal direction component, the vertical direction component, and the 45 ° direction component of the contour line? Is calculated and used as pattern feature amount data.

After the pattern feature amount data in the first rectangular region 32 for pattern feature amount calculation is extracted in step S27, as shown in step S28, the comparison / comparison is performed.
The collation circuit 12 compares and collates the extracted pattern feature amount data with the vehicle shape template of the front shape of the vehicle stored in the vehicle shape template storage device 14 to calculate and store the degree of similarity. Since this similarity calculation is a value represented by a vector quantity indicating in which direction and in what extent the pattern feature amount data is present as described above, the front surface of the vehicle stored in the vehicle shape template storage device 14 is calculated. It can be easily calculated by comparing with a value indicating the vector amount of the contour line in the vehicle shape template of the shape.
Moreover, since the pattern feature amount data is a value indicating the vector amount of the contour line, the similarity can be calculated regardless of the size of the captured image.

In this way, when the similarity between the pattern feature amount data in the first pattern feature amount calculation rectangular area 32 and the vehicle shape template is calculated and stored, the pattern feature amount of the next pattern feature amount calculation rectangular area 32 is stored. In order to calculate, the pattern feature amount calculation rectangular area 32 is moved by x + in one step in the x direction in step S29.
1 is calculated, and it is determined in step S30 whether the value of x does not exceed the final position coordinate n of the vehicle detection target area 31. In this case, since x is 2, the processing is step S26.
Returning to step S27, the process from step S27 to step S29 described above is repeated. If the value of x exceeds n in step S30, the process goes to step S31.

Then, in step S31, x is set to 0, and in step S32, y + 1 is calculated in order to move the rectangular area 32 for calculating the pattern feature amount in the y direction by one box, and the value of y is calculated in step S33. It is determined whether the vehicle candidate does not exceed the final position coordinate m of the vehicle detection target area 31 in which the vehicle candidate is detected. In this case, y is 2, so the process returns to step S26, and step S27 described above is executed.
The processing from step S32 is repeated from step S3
When the value of y exceeds m in 3, the process goes to step S34.

Then, in this step S34, the pattern feature amount calculation rectangular area 32 having the highest similarity between the pattern feature amount data stored in the comparison / collation circuit 12 and the vehicle shape template is selected, and the step is performed. This similarity is compared with the threshold 13 in S35.
As a result, when the similarity is higher than the threshold 13, it is determined that the vehicle 2 exists in the pattern feature amount calculation rectangular area 32, a vehicle detection flag is set in step S36, and the vehicle detection result is character recognition control in step S37. It is sent to the circuit 16. When the data for one frame is processed in this manner, the process returns to the first step S21, and the vehicle detection operation described above is repeated.

On the other hand, the character recognition control circuit 16 to which the vehicle detection result is sent receives the pattern feature amount calculation rectangular area in which the vehicle 2 exists in the vehicle detection target area 31 stored in the binarization / storage circuit 10. 32, the size of the vehicle 2 is determined to determine whether it is the optimum timing (position) for reading a license plate or a plate such as “danger,” “poison,” or “play”. If it is determined that the characters are too small in the reading of the plate at this position, the size of the vehicle 2 and the position where the vehicle 2 is detected in the next frame or in the next frame, and the size of the vehicle 2 in the previous frame. The speed of the vehicle 2 is calculated according to the change from the position or size of the vehicle 2, and the license plate or "danger",
The optimum timing (position) for reading a plate such as "poison" or "drama" is calculated, and a signal is sent to the light source emission control circuit 17.

Therefore, the light source light emission control device 17 is provided in the vehicle 2
The light source 5 such as a strobe light is emitted at the optimum timing (position) to read the license plate of "No", "dangerous", "poison", "drama", etc. Smoothing circuit 8 through 7
After being smoothed by, the character plate is sent to the character recognition control circuit 16 and the character recognition circuit (not shown) recognizes the license plate of the vehicle 2 and characters such as “danger”, “poison”, and “play”.

As described above, the center of each cluster (average vector or average vector and eigenvalue and eigenvector) obtained by clustering pattern feature quantities is registered in the vehicle shape template storage device 14. Therefore, the number of matching pattern data registered in the vehicle shape template storage device 14 can be suppressed to a small number, and the registered matching pattern data can be meaningful data.

In the above embodiment, before the vehicle detection device of this embodiment is installed, the vehicle shape data generated by the computer by the CG technology is clustered using the vehicle CAD data, and the center of each cluster obtained by this is clustered. It is stored in the vehicle shape template storage device 14 as a template. After the installation of the vehicle detection device of the present embodiment, the shape pattern of the actual vehicle photographed by the imaging device 4 is added to the CG vehicle shape pattern before the installation of the device, and clustering is performed on both of them. The center of each resulting cluster can be automatically registered in the vehicle shape template storage device 14 at any time.

Further, the vehicle detection device of the present embodiment is installed at a large number of points, the devices installed at the large number of points are mutually connected by a network, and the vehicle shape patterns obtained by the respective devices are mutually connected. The detection accuracy is improved by automatically registering the result of clustering by reflecting the information in each device.

Further, according to this embodiment, the vehicle 2 can be operated by the single image pickup device 4 without separately providing a vehicle position detecting device using a laser, an ultrasonic wave, a loop coil or the like.
Timing (position) for recognizing information, reading license plates and plates such as "danger,""poison," and "play"
And calculation of image data for recognizing these characters can be performed. Moreover, the recognition of the vehicle is
By using the contour direction pattern feature amount, it can be performed even if the vehicle is far away, and the shadow of the vehicle itself and the shadow of the building are not mistaken for the vehicle, and the license plate of the vehicle and the " A vehicle detection method and device optimal for recognizing characters such as "danger", "poison", and "play" can be provided.

In the above description, the analog / digital conversion circuit 7, the smoothing circuit 8, the edge emphasizing circuit 9, the binarization / storage circuit 10, are used to process the signal from the image pickup device 4.
First pattern feature quantity extraction circuit 11, comparison / collation circuit 1
2, the threshold 13, the vehicle shape template storage device 14, the character recognition control circuit 16, the light source emission control device 17, the vehicle shape pattern database 71, the second pattern feature amount extraction unit 72, the clustering unit 73, and the like have been described. However, it goes without saying that circuits other than the analog / digital conversion circuit 7 may be configured using a computer and its storage device.

[0061]

According to the vehicle detection method of the present invention, comparison and
The time required for the matching process is short.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of an apparatus showing an embodiment of a vehicle detection method of the present invention.

FIG. 2 is a flow chart of a vehicle detection method of the present embodiment.

FIG. 3 is a flow chart of a vehicle detection method of the present embodiment.

FIG. 4 is a diagram for explaining a rectangular area used for pattern feature amount calculation on a screen imaged by an imaging device in the present embodiment.

FIG. 5 is an explanatory diagram when a vehicle is detected in the present embodiment.

[Explanation of symbols]

1 road 2 passing vehicles 3 gantry 4 Vehicle detection imaging device 5 light sources 6 Imaging position on the road 7 Analog / digital conversion circuit 8 Smoothing circuit 9 Edge enhancement circuit 10 Binarization / memory circuit 11 First Pattern Feature Extraction Circuit 12 Comparison / collation circuit 13 threshold 14 Vehicle shape template storage device 16 character recognition control circuit 17 Light source emission control device 71 Vehicle shape pattern database 72 Second pattern feature amount extraction unit 73 Clustering Department

Continued front page    (72) Inventor Jun Kuroda             2-1-1 Niihama, Arai-cho, Takasago, Hyogo Prefecture             Takasago Laboratory, Mitsubishi Heavy Industries, Ltd. F term (reference) 5B057 AA16 BA02 CA08 CA12 CA16                       CB08 CB12 CB16 CC01 CE02                       CE03 CE05 CE09 CE12 DA07                       DA12 DB02 DB09 DC16 DC34                 5H180 AA01 CC04 DD10 EE07                 5L096 BA04 CA02 FA06 FA32 JA09                       JA11 MA07

Claims (9)

[Claims] (A) Clustering is performed on first feature amount data indicating a feature amount of a contour line showing at least a part of a vehicle shape, and template data obtained as a result of the clustering is stored. Steps,
(B) capturing an image of the traveling vehicle and binarizing the image to generate a binarized image in which the contour of the traveling vehicle is represented; and (c) the feature amount of the contour of the binarized image. Extracting the second characteristic amount data shown, (d)
A vehicle detection method comprising: comparing the template data with the second feature amount data, and determining that a vehicle is included in the binarized image based on a result of the comparison. 2. The vehicle detection method according to claim 1, wherein (a) stores the center of each cluster obtained as a result of the clustering as the template data. 3. The vehicle detection method according to claim 1 or 2, wherein the feature quantity of the contour line indicated in each of the first and second feature quantity data is the direction and size of the contour line. Vehicle detection method represented by vector quantity. 4. The vehicle detection method according to claim 3, wherein (a) stores an average vector of each cluster obtained as a result of the clustering as the template data. 5. The vehicle detection method according to claim 3, wherein (a) stores an average vector, an eigenvalue, and an eigenvector of each cluster obtained as a result of the clustering as the template data. 6. The vehicle detection method according to claim 1, wherein (a) uses a minimax method or a k-mean method as the clustering method. Method. 7. The vehicle detection method according to claim 1, wherein the template data is at least a part of a shape of a vehicle obtained by executing the vehicle detection method at a plurality of locations. The vehicle detection method for performing clustering on the first feature amount data indicating the feature amount of the contour line indicating, and storing the template data obtained as a result of the clustering. 8. The vehicle detection method according to claim 1, wherein the second feature amount data is the contour included in a partial region of the cut out binarized image. A vehicle detection method that indicates the feature amount of a line. 9. Template data generation for performing clustering on first feature amount data indicating a feature amount of a contour line showing at least a part of a vehicle shape, and storing template data obtained as a result of the clustering. A storage unit; a binarized image generation unit that captures an image of the traveling vehicle and binarizes the image to generate a binarized image representing the contour line of the traveling vehicle; A feature amount extraction unit that extracts second feature amount data indicating a feature amount, the template data, and the second feature amount data are compared, and based on the result of the comparison, the binary image is obtained. A vehicle detection device including a determination unit that determines that a vehicle is included.