CN114463570A - Vehicle detection method based on clustering algorithm - Google Patents

Abstract

The invention belongs to the field of target identification, and discloses a vehicle detection method based on a clustering algorithm, which comprises the following steps: preprocessing the image; the color quantization processing is carried out on the image, so that the color types of the pixel points are reduced; converting an image from an RGB color space to an LAB color space, extracting color characteristic vectors of image pixel points of the LAB color space, and forming a characteristic quantity value matrix by the color characteristic vectors; calculating the local density of each pixel point and the distance between the pixel point and the higher density point, screening out the pixel points with the local density larger than a threshold value and the distance between the pixel points with the higher density larger than the threshold value as cluster centers, and setting other pixel points to the cluster centers to cluster the pixel points; and generating a vehicle segmentation image according to the clustering result. The invention greatly reduces the calculation amount; the quantized color has more accurate quantized value through continuous iteration of the algorithm, and the subsequent detection effect is improved; the predefined cluster number does not need to be specified, and the calculation efficiency is high.

Description

A Vehicle Detection Method Based on Clustering Algorithm

technical field

The invention belongs to the technical field of target recognition, and in particular relates to a vehicle detection method based on a clustering algorithm.

Background technique

With the improvement of people's living standards, the number of vehicles is increasing, and the types, models, and frame structures of vehicles have different characteristics. Therefore, target detection and recognition technology using virtual reality technology and image processing technology, such as face Recognition, pedestrian feature detection, vehicle detection, etc., are increasingly applied to the traffic field in the process of the development of intelligent transportation and sensing technology, such as vehicle collision prediction and warning, vehicle departure from lanes and other emergencies; Through video capture in traffic, visual technology is used for analysis and tracking, and the congestion of people and vehicles can be detected, which can be used for convenient and efficient traffic management and reduce traffic accidents.

In vehicle detection technology, moving targets are detected through background images, textures and colors to improve detection accuracy. In the detection process, multi-target detection, tracking, identification etc. In vehicle detection, the background image is separated from video and image. The commonly used techniques are based on background difference, prior knowledge, optical flow method, machine learning, etc. Among them, the research and application of machine learning has become the current frontier field, and the method of target detection and more and more technology.

The clustering algorithm mainly discriminates the dissimilarity and similarity of the data in the generated clusters, and achieves the similarity measure in the object to the greatest extent possible. The existing clustering algorithm has low detection accuracy, large amount of calculation and low efficiency.

SUMMARY OF THE INVENTION

In view of this, the present invention proposes a vehicle detection method based on a clustering algorithm. In terms of the color quantization algorithm, the quantization value is more accurate through the continuous iteration of the algorithm, thereby improving the effect of subsequent detection. In vehicle detection, there is no need to specify a predefined number of clusters, and only a parameter with lower sensitivity is needed, thereby achieving the effect of univariate control and reducing the complexity of the algorithm. In addition, the computational complexity of the entire algorithm is reduced in various ways, thereby improving the efficiency of the entire algorithm.

Specifically, a vehicle detection method based on a clustering algorithm disclosed in the present invention includes the following steps:

Preprocessing the image, including filtering and denoising the image and resizing the image;

By quantizing the color of the image, the color types of the pixels are reduced;

Convert the image from the RGB color space to the LAB color space, extract the color feature vector of the image pixel in the LAB color space, and the color feature vector forms a feature value matrix;

Calculate the local density of each pixel and the distance from the higher density point, screen out the pixels whose local density is greater than the threshold and the distance from the higher density point is greater than the threshold as the cluster center, and place other pixels to the cluster center , to perform pixel clustering;

Generate vehicle segmentation images based on the clustering results.

Further, the filtering and noise reduction method of the image is a median filtering method, which is used to remove impulse noise and salt and pepper noise while retaining the edge details of the image; the adjustment of the image size is used to reduce the number of operations of the similarity measure and improve the running speed. .

Further, the steps of the color quantization processing include:

S1: randomly select K RGB components from the image, M _k = [R' _k , G' _k , B' _k ], where k is the count of K, and R' _k , G' _k , B' _k are selected respectively The R, G, B components corresponding to the K points of ;

S2: Calculate the color distance between each pixel and the selected K RGB components

Among them, k is the count of K RGB components, j is the count of pixels, and d _kj is the color distance between the jth pixel and the kth RGB component;

S3: classify the color distance into pixels, and the classification method is as follows:

Compare each pixel with the calculated K color distances of the K RGB components, and classify the pixel into the category with the smallest color distance corresponding to the K RGB components;

S4: According to the result of pixel classification, calculate the color average value of all pixel points in each category, and replace the value of M _k with the calculated average value;

S5: Calculate the color distance between the pixel values corresponding to each category and the calculated new K categories of pixel values, and determine whether the classification of the corresponding pixels in each category has changed; if the classification of all pixels in all categories If the classification of all the pixels in , has not changed, then go to step S6, otherwise jump to step S2;

S6: Continue according to the pixel values of the K categories obtained in steps S1 to S5 and the pixel points in each category, and replace the RGB components of the pixel points used in each category with the RGB components of the category pixel values in the category, Complete color quantization.

Further, to convert the image from the RGB color space to the LAB color space, the conversion steps are as follows:

Gamma-corrected color components Rg, Gg, and Bg are obtained by using the first Gamma correction method for the RGB components;

The gamma correction color component is carried out the conversion of XYZ color space, obtains XYZ color space color component X, Y, Z;

Performing the second Gamma correction on the color space color components X, Y, Z to obtain the corrected XYZ color space color components X1, Y1, Z1;

Convert the X1, Y1, Z1 to the color components 1, a, b of the LAB color space;

Extract the color information of the image pixels in the LAB color space, and generate the color feature vector of each pixel.

Further, the first Gamma correction method is as follows:

where x is one of the original color components of R, G, and B;

The second Gamma correction formula is as follows:

Where y is one of the color components X, Y, and Z in the XYZ color space;

The formula for the conversion of the Gamma-corrected color component to the XYZ color space is as follows:

[X,Y,Z]＝M*[Rg,Gg,Bg]

Where X, Y, Z are the color components of the XYZ color space,

The formulas for converting the X1, Y1, and Z1 into the color components l, a, and b of the LAB color space are as follows:

a=500(Xl-Yl)

b=500(Yl-Zl)

Further, the vehicle detection:

The eigenvalues of the color feature vector are calculated according to the following formula, and all the eigenvalues are formed into a eigenvalue matrix D:

where i represents the i-th pixel, j represents the j-th pixel, and d _ij represents the feature value between pixel i and pixel j, where l, a, and b represent the pixel in the LAB color space. , a, b three parameters;

The upper triangular matrix of the eigenvalue matrix D is arranged in ascending order, and the truncation distance d _c is calculated according to the median calculation method;

Calculate the local density ρ _i of each pixel point according to the following formula:

ρ _i =∑ _j χ(d _ij -d _c )

Among them, i is the number of the current pixel point, j represents the number of other pixel points except the current pixel point, d _ij is the feature value between the pixel point i and the pixel point j, and d _c is the cutoff distance;

The expression form of χ is:

Calculate the distance δ _i from the higher density point as follows:

δ _i =min(d _ij )

Wherein, d _ij is the feature value between the pixel point i and the pixel point j, i is the number of the current pixel point, j represents the number of other pixel points except the current pixel point and ρ _i <ρ _j ;

Screen the local density and the distance from the higher density point, and screen out the pixel point with the larger local density and the distance from the higher density point as the cluster center;

The pixel point of the determined cluster center is the center, and according to the local density formula, _all the pixel points whose distances from the pixel point of the cluster center are less than the cut-off distance dc are placed in the center of the cluster to complete the clustering of the pixel points;

According to the aspect ratio k of the actual vehicle in the image, the aspect ratio of the vehicle after pixel clustering is screened, and the clusters with a large error with the aspect ratio k of the vehicle are removed.

Further, the generating the vehicle segmentation image according to the clustering result includes: selecting the coordinates of the outermost pixel points of each cluster to form a rectangle according to the vehicle detection result, and generating the vehicle segmentation image.

The beneficial effects of the present invention are as follows:

The invention greatly reduces the amount of calculation and improves the efficiency under the condition of ensuring the detection accuracy through the steps of reducing the image size and color quantization;

In the color quantization algorithm proposed by the present invention, the quantized color makes its quantized value more accurate through the constant iteration of the algorithm, thereby improving the effect of subsequent detection.

The vehicle detection algorithm proposed by the present invention does not need to specify a predefined number of clusters, but only needs a parameter with lower sensitivity. Compared with other density-based methods, the present invention has higher computational efficiency.

Description of drawings

Fig. 1 is the flow chart of the vehicle detection method of the present invention;

Fig. 2 color quantization flow chart of the present invention;

Fig. 3 is a flow chart of cluster screening of the present invention.

Detailed ways

The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way, and any transformation or replacement based on the teachings of the present invention belongs to the protection scope of the present invention.

The technical scheme adopted in the present invention comprises the steps as follows:

The invention uses the clustering algorithm to segment the vehicle in the image and obtains the vehicle target detection image. Figure 1 is a flow chart of vehicle detection based on clustering algorithm. The specific description of each step is as follows:

1. Image preprocessing: The main tasks of image preprocessing in the present invention are image filtering and noise reduction and image size adjustment.

The method of image filtering is median filtering. The advantage of this method is that it can remove impulse noise and salt and pepper noise. In addition, it can retain the edge details of the image while removing noise. The adjustment of the image size is to reduce the number of operations of the similarity measurement value and improve the running speed. The specific scaling size is selected according to the actual hardware conditions. Generally, the ratio of reducing the original image by half is selected.

2. Color quantization: In step 1, in order to reduce the number of operations of the similarity measure and improve the running speed, the image size adjustment method is adopted, but for RGB images, the amount of operations is still very large, and continuing to reduce the resolution will lose the original image. a lot of information.

In order to solve this problem, the present invention proposes a color quantization method. By performing color quantization processing on the image, the color information of the original image is maximized while reducing the color types of the pixel points, and the operation efficiency is improved. Figure 2 is a flow chart of color quantization, and the specific steps are as follows:

2.1 K value selection: K RGB components are randomly selected from the image, and the general rule for selection is: as far as possible to disperse the selection in the image, and to represent the main colors in the image to the greatest extent.

Among them, K is the number of color quantization categories, and the value of K is proportional to the complexity of the color of the image scene. Usually, the value of K is greater than or equal to 3. Its representation is M _k =[R′ _k , G′ _k , B′ _k ], where k is the count of K, R′ _k , G′ _k , B′ _k are the R corresponding to the selected K points, respectively , G, B components.

2.2 Color distance calculation: Calculate the distance between each pixel and the K RGB components selected in step 2.1 according to formula 1.

Among them, k is the count of K RGB components, j is the count of pixels, and dkj is the distance between the jth pixel and the kth RGB component.

2.3 Pixel classification: The color distance calculated in step 2.2 is classified into pixels. The specific classification method is:

Compare each pixel in step 2.2 with the calculated K color distances of the K RGB components, and classify the pixel into the category with the smallest color distance corresponding to the K RGB components.

2.4 Category pixel value calculation: This step is to recalculate the color values of the K RGB components selected in step 2.1 to obtain a new category pixel value.

The specific calculation method is: according to the result of pixel classification in step 2.3, calculate the color average value of all pixels in each category according to formula 2, and replace M _k in step 2.1 with the calculated average value.

Among them, k is the count of K RGB components, R′ _k , G′ _k , B′ _k are the R, G, and B components corresponding to the selected K points, respectively, and n is the total number of pixels corresponding to the category, i is the number of the pixel in the category, Ri, Gi, Bi are the RGB components of the i-th pixel in the category, respectively.

2.5 Pixel category judgment: This step is an iterative step in the color quantization step, and the purpose of this step is to make the pixel classification more accurate and the category pixel value more reasonable.

The specific steps are: calculate the color distance between the pixel values corresponding to each category and the new K categories of pixel values calculated in step 2.4, and determine whether the classification of the corresponding pixels in each category has changed; if If the classification of all pixels in the classification of all pixels in all categories has not changed, proceed to the next step, otherwise skip to step 2.2.

2.6 Color replacement quantization: This step is the specific execution step of the color quantization algorithm. The pixel-wise RGB components are replaced with the RGB components in the pixel values of the category in this category, thereby completing color quantization.

3. Color space conversion: RGB color space needs to be converted into LAB color space in the present invention, and the concrete conversion steps are as follows:

3.1 Gamma correction: Gamma correction is performed on R, G, and B according to formula 3 to obtain the corrected Rg=f(R), Gg=f(G), Bg=f(B), where R, G, and B are the original Color components, Rg, Gg, Bg are Gamma correction color components.

3.2 XYZ color space conversion: convert the XYZ color space according to formula 4.

[X,Y,Z]＝M*[Rg,Gg,Bg] (4)

Among them, Rg, Gg, Bg are Gamma correction color components, X, Y, Z are XYZ color space color components,

3.3 XYZ linear normalization: perform Gamma correction on X, Y, and Z according to formula 5 to obtain the corrected Xl=g(X), Yl=g(Y), Zl=g(Z), where X, Y, Z is the color component of the XYZ color space, and X1, Y1, and Z1 are the color components of the values after linear normalization of X, Y, and Z.

3.4 LAB color space conversion: carry out the LAB color space according to formula 6, where l, a, b are the color components of the LAB color space, X1, Yl, Zl are the linearly normalized X, Y, Z in the XYZ color space. value.

4. Extract feature vector: extract the color information of the image pixels in the LAB color space, and generate the color feature vector of each pixel.

The feature vector of each pixel is: Li=[l,a,b], where l, a, b represent the three parameters l, a, b of the pixel in the LAB color space, and i is the number of the image pixel .

5. Vehicle detection:

5.1 Calculate the feature value: Calculate the feature value of the adjustment vector obtained in step 4 according to formula 7, and put all the feature value into the feature value matrix D.

where i represents the i-th pixel, j represents the j-th pixel, and d _ij represents the feature value between pixel i and pixel j, where l, a, and b represent the pixel in the LAB color space. , a, b three parameters.

The representation form of the eigenvalue matrix D is:

5.2 Determining the truncation distance: The specific method for determining the truncation distance dc in this step is: arrange the upper triangular matrix of the feature value matrix D calculated in step 5.1 in ascending order, and calculate the _truncation distance dc according to the median calculation method.

5.3 Calculate the local density: The local density ρ _i in this step represents the number of similar pixels around the pixel. The larger the value, the more the number of similar pixels around the pixel. The distance between the pixels is less than the number of pixels with the cut-off distance _dc .

The specific local density method is: calculating the local density of each pixel point according to formula 8.

ρ _i =∑ _j χ(d _ij -d _c ) (8)

Among them, i is the number of the current pixel, j represents the number of other pixels except the current pixel, d _ij is the feature value between pixel i and pixel j, and d _c is the cutoff distance. The expression form of χ is:

5.4 Calculate the distance from the higher density point: The purpose of the distance δ _i calculated from the higher density point in this step is to find all the pixels whose local density is greater than the ith pixel, which is the same as the ith pixel. the minimum distance between them.

The specific calculation method is as follows: according to formula 9, the distance δ _i from the higher density point is calculated.

δ _i =min(d _ij ) (9)

Among them, d _ij is the feature value between pixel i and pixel j, i is the number of the current pixel, j represents the number of other pixels except the current pixel, and ρ _i <ρ _j .

5.5 Determine the cluster center: This step is to determine the cluster center of the pixel points in the image, that is, to initially find the center of the vehicle in the image. The specific determination method is: according to the local density calculated in step 5.3 and step 5.4 and the distance from the higher density point, screen out the pixel points with the larger local density and the larger distance from the higher density point as the cluster center.

5.6 Pixel point clustering: Perform pixel point clustering according to the cluster center determined in step 5.5. The specific clustering method is as follows: the pixel point of the determined cluster center is the center, and according to formula 8, all the pixel points whose distance from the pixel point of the cluster center is less than the cut-off distance dc are placed in the center of the cluster, so as to complete the pixel point. clustering. The clustering results of this step represent the preliminary detection results of vehicles in the image.

5.7 Cluster screening: This step is to screen the vehicle centers and vehicle detection results initially detected in steps 5.5 and 5.6.

The specific screening method is: according to the aspect ratio k of the actual vehicle in the image, screen the aspect ratio of the vehicle after the pixel clustering in step 5.6, and remove the cluster with a large error with the aspect ratio k of the vehicle.

6. Generate a vehicle segmentation image: This step is to segment the vehicle from the image or frame it in the image. The specific method is: according to the vehicle detection result in step 5, select the coordinates of the outermost pixel points of each cluster. A rectangle, thereby completing the segmentation of the vehicle, generates a vehicle segmentation image.

The beneficial effects of the present invention are as follows:

The invention greatly reduces the amount of calculation and improves the efficiency under the condition of ensuring the detection accuracy through the steps of reducing the image size and color quantization;

In the color quantization algorithm proposed by the present invention, the quantized color makes its quantized value more accurate through the constant iteration of the algorithm, thereby improving the effect of subsequent detection.

The vehicle detection algorithm proposed by the present invention does not need to specify a predefined number of clusters, but only needs a parameter with lower sensitivity. Compared with other density-based methods, the present invention has higher computational efficiency.

As used herein, the word "preferred" means serving as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, use of the word "preferred" is intended to present concepts in a specific manner. The term "or" as used in this application is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless specified otherwise or clear from context, "X employs A or B" is meant to naturally include either of the permutations. That is, "X employs A or B" is satisfied in any of the preceding examples if X employs A; X employs B; or X employs both A and B.

Furthermore, although the present disclosure has been shown and described with respect to one implementation or implementation, equivalent variations and modifications will occur to those skilled in the art based on a reading and understanding of this specification and drawings. The present disclosure includes all such modifications and variations and is limited only by the scope of the appended claims. In particular with respect to the various functions performed by the above-described components (eg, elements, etc.), the terms used to describe such components are intended to correspond to any component that performs the specified function of the component (eg, which is functionally equivalent) (unless otherwise indicated), even if not structurally equivalent to the disclosed structures that perform the functions of the exemplary implementations of the present disclosure shown herein. Furthermore, although a particular feature of the present disclosure has been disclosed with respect to only one of several implementations, such feature may be combined with one or other features of other implementations as may be desirable and advantageous for a given or particular application combination. Moreover, to the extent that the terms "comprising," "having," "containing," or variations thereof are used in the detailed description or the claims, such terms are intended to include in a manner similar to the term "comprising."

Each functional unit in this embodiment of the present invention may be integrated into one processing module, or each unit may exist physically alone, or multiple or more units may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware, and can also be implemented in the form of software function modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium. The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, and the like. The above-mentioned apparatuses or systems may execute the storage methods in the corresponding method embodiments.

To sum up, the above-mentioned embodiment is an embodiment of the present invention, but the embodiment of the present invention is not limited by the embodiment, and any other changes that deviate from the spirit and principle of the present invention, Modifications, substitutions, combinations, and simplifications should all be equivalent substitutions, which are all included within the protection scope of the present invention.

Claims (7)

1. a vehicle detection method based on clustering algorithm, is characterized in that, comprises the following steps: Preprocessing the image, including filtering and denoising the image and resizing the image; By quantizing the color of the image, the color types of the pixels are reduced; Convert the image from the RGB color space to the LAB color space, extract the color feature vector of the image pixel in the LAB color space, and the color feature vector forms a feature value matrix; Calculate the local density of each pixel and the distance from the higher density point, screen out the pixels whose local density is greater than the threshold and the distance from the higher density point is greater than the threshold as the cluster center, and place other pixels to the cluster center , to perform pixel clustering; Generate vehicle segmentation images based on the clustering results. 2. The vehicle detection method based on a clustering algorithm according to claim 1, wherein the filtering noise reduction method of the image is a median filtering method for removing impulse noise and salt and pepper noise while retaining the edge of the image Details; resizing of the image is used to reduce the number of operations for the similarity measure and increase the speed. 3. The vehicle detection method based on a clustering algorithm according to claim 1, wherein the step of the color quantization processing comprises: S1: randomly select K RGB components from the image, M _k = [R' _k , G' _k , B' _k ], where k is the count of K, and R' _k , G' _k , B' _k are selected respectively The R, G, B components corresponding to the K points of ; S2: Calculate the color distance between each pixel and the selected K RGB components

Among them, k is the count of K RGB components, j is the count of pixels, and d _kj is the color distance between the jth pixel and the kth RGB component; S3: classify the color distance into pixels, and the classification method is as follows: Compare each pixel with the calculated K color distances of the K RGB components, and classify the pixel into the category with the smallest color distance corresponding to the K RGB components; S4: According to the result of pixel classification, calculate the color average value of all pixel points in each category, and replace the value of M _k with the calculated average value; S5: Calculate the color distance between the pixel values corresponding to each category and the calculated new K categories of pixel values, and determine whether the classification of the corresponding pixels in each category has changed; if the classification of all pixels in all categories If the classification of all the pixels in , has not changed, then go to step S6, otherwise jump to step S2; S6: Continue according to the pixel values of the K categories obtained in steps S1 to S5 and the pixel points in each category, and replace the RGB components of the pixel points used in each category with the RGB components of the category pixel values in the category, Complete color quantization. 4. the vehicle detection method based on clustering algorithm according to claim 1, is characterized in that, described image is converted into LAB color space from RGB color space, and conversion step is as follows: Gamma-corrected color components Rg, Gg, and Bg are obtained by using the first Gamma correction method for the RGB components; The gamma correction color component is carried out the conversion of XYZ color space, obtains XYZ color space color component X, Y, Z; Performing the second Gamma correction on the color space color components X, Y, Z to obtain the corrected XYZ color space color components X1, Y1, Z1; Convert the X1, Y1, Z1 to the color components 1, a, b of the LAB color space; Extract the color information of the image pixels in the LAB color space, and generate the color feature vector of each pixel. 5. The vehicle detection method based on a clustering algorithm according to claim 4, wherein the first Gamma correction method is as follows:

where x is one of the original color components of R, G, and B; The second Gamma correction formula is as follows:

Where y is one of the color components X, Y, and Z in the XYZ color space; The formula for the conversion of the Gamma-corrected color component to the XYZ color space is as follows: [X,Y,Z]＝M*[Rg,Gg,Bg] Where X, Y, Z are the color components of the XYZ color space,

The formulas for converting the X1, Y1, and Z1 into the color components l, a, and b of the LAB color space are as follows:

a=500(Xl-Yl) b=500(Yl-Zl) 6 . The vehicle detection method based on a clustering algorithm according to claim 1 , wherein the calculation of the local density of each pixel point and the distance from the higher density point is performed, and the local density greater than the threshold and the distance from the higher density point are screened out. 7 . Pixels whose distance between high-density points is greater than the threshold are taken as the cluster center, and other pixels are placed in the center of the cluster, and the clustering of pixel points includes: The eigenvalues of the color feature vector are calculated according to the following formula, and all the eigenvalues are formed into a eigenvalue matrix D:

where i represents the i-th pixel, j represents the j-th pixel, and d _ij represents the feature value between pixel i and pixel j, where l, a, and b represent the pixel in the LAB color space. , a, b three parameters; The upper triangular matrix of the eigenvalue matrix D is arranged in ascending order, and the truncation distance d _c is calculated according to the median calculation method; Calculate the local density ρ _i of each pixel point according to the following formula: ρ _i =∑ _j χ(d _ij -d _c ) Among them, i is the number of the current pixel point, j represents the number of other pixel points except the current pixel point, d _ij is the feature value between the pixel point i and the pixel point j, and d _c is the cutoff distance; The expression form of χ is:

Calculate the distance δ _i from the higher density point as follows: δ _i =min(d _ij ) Wherein, d _ij is the feature value between the pixel point i and the pixel point j, i is the number of the current pixel point, j represents the number of other pixel points except the current pixel point and ρ _i <ρ _j ; Screen the local density and the distance from the higher density point, and screen out the pixel point with the larger local density and the distance from the higher density point as the cluster center; The pixel point of the determined cluster center is the center, and according to the local density formula, _all the pixel points whose distances from the pixel point of the cluster center are less than the cut-off distance dc are placed in the center of the cluster to complete the clustering of the pixel points; According to the aspect ratio k of the actual vehicle in the image, the aspect ratio of the vehicle after pixel clustering is screened, and the clusters with a large error with the aspect ratio k of the vehicle are removed. 7. The vehicle detection method based on a clustering algorithm according to claim 1, wherein the generating a vehicle segmentation image according to the clustering result comprises: selecting the coordinates of the outermost pixel points of each cluster according to the vehicle detection result Form a rectangle to generate vehicle segmentation images.

Images