Intelligent grating traffic control system

Abstract

The invention discloses an intelligent grating traffic control system, and relates to the field of intelligent traffic. The system comprises a camera, an image processing module, a classification processing module and a grating indication board, and the camera is used for shooting appearance images of vehicles to be classified; the image processing module is used for extracting a plurality of vehicle features of the appearance image, and the vehicle features comprise clear features and fuzzy features; the classification processing module is used for carrying out classification analysis on the to-be-classified vehicle according to the multiple vehicle features and determining the vehicle type of the to-be-classified vehicle; and the grating indication board is used for indicating the passing road of the to-be-classified vehicle according to the vehicle type. By implementing the technical scheme provided by the invention, the problem of low accuracy of classifying and guiding the passing vehicles in the current highway toll station is solved.

Description

Technical Field

[0001] The present application relates to the technical field of smart transportation, and specifically to an intelligent grating traffic control system. Background Art

[0002] In the past, highway toll stations basically did the classification and guidance of passing vehicles manually. However, with the rapid development of visual inspection technology and AI technology, this classification and guidance work has begun to be replaced by intelligent grating traffic systems, which has greatly saved labor costs.

[0003] However, with more and more vehicles passing through highway toll stations, the vehicle shooting environment is becoming more and more complex, and the quality of the captured images often deteriorates due to motion blur, low light or occlusion. This makes it difficult for today's intelligent grating traffic system to accurately determine the vehicle type, resulting in a significant reduction in classification accuracy. Summary of the invention

[0004] In order to solve the problem of low accuracy in the classification and guidance of passing vehicles at highway toll stations, the present application provides an intelligent grating traffic control system, which includes a camera, an image processing module, a classification processing module and a grating sign, wherein: The camera is used to capture the appearance image of the vehicle to be classified; The image processing module is used to extract multiple vehicle features from the appearance image, wherein the vehicle features include clear features and fuzzy features; The classification processing module is used to perform classification analysis on the vehicle to be classified according to the plurality of vehicle features, and determine the vehicle type of the vehicle to be classified; The grating sign is used to indicate the passage road for the vehicle to be classified according to the vehicle type.

[0005] Optionally, the classification analysis of the vehicle to be classified is performed based on the plurality of vehicle features to determine the vehicle type of the vehicle to be classified, specifically: calculating a ratio between the number of the sharp features and the number of the blurred features; Determine whether the ratio result is greater than or equal to a preset ratio; If the ratio result is greater than or equal to the preset ratio, the plurality of vehicle features are input into a vehicle classification model for classification to obtain the vehicle type of the vehicle to be classified.

[0006] Optionally, the determining whether the ratio result is greater than or equal to a preset ratio further includes: If the ratio result is less than the preset ratio, performing correlation analysis on the fuzzy feature to obtain a first correlation feature of the fuzzy feature; Performing a rationality evaluation on the first associated feature of the fuzzy feature and the clear feature to obtain a rationality score; If the rationality score is greater than or equal to a preset rationality score, the clear feature and the first associated feature are input into the vehicle classification model for classification, and the vehicle type of the vehicle to be classified.

[0007] Optionally, performing correlation analysis on the fuzzy feature to obtain a first correlation feature of the fuzzy feature specifically includes: Performing wavelet decomposition on the fuzzy features to obtain intermediate frequency subbands; Determining a first gain coefficient of the fuzzy feature according to a proportion of the number of the fuzzy features in the plurality of vehicle features; Based on the first gain coefficient, reconstructing the image of the intermediate frequency sub-band to obtain a fuzzy feature to be matched; The fuzzy feature to be matched is matched with a preset vehicle feature table to obtain a first associated feature of the fuzzy feature.

[0008] Optionally, matching the to-be-matched fuzzy feature with a preset vehicle feature table to obtain a first associated feature of the fuzzy feature specifically includes: Extracting a plurality of key pixel points and feature descriptors of the fuzzy features to be matched; Constructing a contour image of the fuzzy feature to be matched according to the plurality of key pixel points and feature descriptors; The contour image is matched with the preset vehicle feature table to obtain a first associated feature of the fuzzy feature.

[0009] Optionally, the first associated feature of the fuzzy feature and the clear feature are evaluated for rationality to obtain a rationality score, specifically: Identifying a fuzzy type of the fuzzy feature; Extracting texture features from the first associated features according to the fuzzy type; The texture feature and the clear feature are subjected to similarity calculation to obtain the rationality score, wherein the similarity calculation includes structural similarity, brightness similarity and contrast similarity.

[0010] Optionally, after performing a rationality evaluation on the first associated feature of the fuzzy feature and the clear feature to obtain a rationality score, the method further includes: a. If the rationality score is less than the preset rationality score, then the score difference between the rationality score and the preset rationality score is calculated; b. Based on the score difference, amplify and adjust the first gain coefficient to obtain a second gain coefficient; c. Using the second gain coefficient, performing image enhancement, correlation feature matching, and rationality evaluation on the fuzzy feature to obtain a second correlation feature and a second rationality score; d. determining whether the second rationality score is greater than or equal to the preset rationality score; If yes, inputting the second associated feature and the clear feature into a vehicle classification model for classification to obtain the vehicle type of the vehicle to be classified; If not, repeat steps a to d until the rationality score calculated by the rationality evaluation is greater than or equal to the preset rationality score.

[0011] Optionally, the image processing module further includes a cache unit and an image definition recognition unit. The cache unit is used to store the first appearance image of the vehicle to be classified, and select, according to the memory proportions of the plurality of first appearance images, a plurality of first appearance images whose memory proportions are greater than or equal to a preset proportion from the plurality of first appearance images as the appearance images to be processed; The image definition recognition unit is used to perform definition recognition on the multiple appearance images to be processed, and select the image with the highest definition from the multiple appearance images to be processed as the appearance image.

[0012] Optionally, the performing clarity recognition on the plurality of appearance images to be processed specifically includes: A lightweight semantic segmentation network is used to extract multiple determination areas corresponding to each of the multiple appearance images to be processed, wherein the multiple determination areas are preset areas on the vehicle with low reflective interference; Performing wavelet decomposition on the multiple determination areas corresponding to the multiple appearance images to be processed, to obtain the gradient energies of the multiple appearance images to be processed in the low-frequency sub-band and the gradient energies of the high-frequency sub-band; The ratio of the gradient energy of the low-frequency sub-band to the gradient energy of the high-frequency sub-band of each of the plurality of appearance images to be processed is calculated to obtain the clarity of the plurality of appearance images to be processed.

[0013] Optionally, the extracting a plurality of vehicle features from the appearance image further includes: Segmenting the appearance image into a plurality of feature images of the same size; Identify the clear pixel points of the vehicle and the total pixel points of the vehicle corresponding to each of the plurality of feature images; Determining the clarity of the plurality of feature images according to the ratio of the vehicle clear pixel points in the plurality of feature images to the total pixel points of the vehicle; If the clarity of the first feature image is greater than or equal to a preset clarity threshold, determining that the first feature image is a clear feature, the first feature image is any one of the plurality of feature images; If the clarity of the second feature image is less than a preset clarity threshold, it is determined that the second feature image is a blurred feature, and the second feature image is any one of the multiple feature images.

[0014] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. When a passing vehicle needs to pass through a highway toll station, the present application uses a camera to capture the appearance image of the vehicle to be classified, and then the image processing module extracts clear features and fuzzy features from the appearance image. The classification processing module then determines whether the clarity of the appearance image can be recognized by the vehicle classification model based on the proportion of the clear features in the entire appearance image; if the proportion is high, it means that the clarity of the appearance image meets the requirements and can be directly input into the vehicle classification model to determine the vehicle type; if the proportion is low, the fuzzy features need to be amplified for correct recognition, but directly enhancing the fuzzy features not only requires high hardware resources, but may also introduce too much noise, thereby causing image distortion; therefore, the present application matches the fuzzy features with the preset vehicle feature table to obtain the corresponding associated features, and then recombine the associated features with the original clear features into a new complete image, and then performs a rationality analysis on the new image to obtain a rationality score. If the rationality score is high enough, it means that the new image can represent the original image, thereby indirectly improving the clarity of the original image to improve the accuracy of vehicle type recognition.

[0015] 2. Before matching the fuzzy features with the preset vehicle feature table, in order to improve the matching accuracy, the present application performs adaptive gain on the fuzzy features, determines the corresponding gain coefficient according to the proportion of the fuzzy features in the entire original image, and then performs image enhancement on the fuzzy features based on the gain coefficient. In addition, before image enhancement, the present application performs wavelet decomposition on the fuzzy features to determine the distribution of key information therein. It should be noted that the key information in the fuzzy features is mainly concentrated in the intermediate frequency sub-band, and the noise has little effect on it. Although the high frequency sub-band contains key information, its corresponding key information is detail information, and the noise has the greatest impact on the detail information, resulting in the effect of its gain is not ideal. Therefore, the present application gains the intermediate frequency sub-band and then reconstructs the gained intermediate frequency sub-band to obtain the gained fuzzy features, thereby improving the matching accuracy with the preset vehicle feature table. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a structural schematic diagram of an intelligent grating traffic control system provided in an embodiment of the present application.

[0017] Figure 2 It is a structural diagram of an image processing module provided in an embodiment of the present application.

[0018] Explanation of the reference numerals: 1. Camera; 2. Image processing module; 3. Classification processing module; 31. Cache unit; 32. Image clarity recognition unit; 4. Grating sign. DETAILED DESCRIPTION

[0019] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0020] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0021] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0022] With the rapid development of China's economy, the number of car buyers has increased significantly, resulting in a significant increase in the traffic volume of highway toll stations. In order to avoid traffic congestion and improve traffic efficiency, vehicles about to pass through are generally classified and guided 2 kilometers away from the highway toll station, so that vehicles of the same type are evenly distributed in the corresponding multiple gate lanes to avoid the situation where one gate lane has a heavy workload and another gate lane has a high idle rate. In the past, highway toll stations basically did manual work to classify and guide passing vehicles, but with the rapid development of visual inspection technology and AI technology, this classification and guidance work has begun to be replaced by intelligent grating traffic systems, which photograph passing vehicles and then input the photographed images into the AI ​​model for vehicle type recognition. Finally, based on the recognition results, the intelligent grating sign at the classification intersection is controlled to instruct passing vehicles to go to the gate lane with low congestion. This not only improves traffic efficiency, but also greatly saves labor costs.

[0023] However, with more and more vehicles passing through highway toll stations, the vehicle shooting environment is becoming more and more complex, and the quality of the captured images often deteriorates due to motion blur, low light or occlusion. This makes it difficult for today's intelligent grating traffic system to accurately determine the vehicle type, resulting in a significant reduction in classification accuracy.

[0024] In order to solve the above problems, the present application provides an intelligent grating traffic control system, such as Figure 1 As shown, the system includes a camera 1, an image processing module 2, a classification processing module 3 and a grating sign 4, wherein Camera 1 is used to capture the appearance image of the vehicle to be classified. Specifically, when the vehicle to be classified enters the preset shooting range, the camera 1 will be triggered to shoot immediately. The triggering method can be infrared signal triggering or road pressure sensor triggering, so as to avoid incomplete appearance images caused by premature shooting or untimely shooting. Camera 1 will capture multiple continuous frame images during the shooting process to meet the requirements of different models for the best shooting angle, and also ensure that the vehicle accounts for more than 90% of the image; then, the multiple continuous frame images captured are transmitted to the image processing module 2.

[0025] After acquiring multiple continuous frame images, the image processing module 2 selects an image with higher definition as the appearance image for subsequent vehicle type identification, such as Figure 2As shown, the image processing module 2 also includes a cache unit 31 and an image clarity recognition unit 32, wherein the cache unit 31 is used to store the first appearance image of the vehicle to be classified, the first appearance image is a continuous frame image, and some images in the multiple first appearance images may be affected by factors such as motion and lighting, resulting in a serious decrease in clarity. For these first appearance images with severely decreased clarity, more computing resources will be occupied, thereby resulting in reduced data processing efficiency; therefore, the present application needs to screen out images with higher clarity from the multiple first appearance images, so as to improve the effective utilization of computing resources. Specifically: the cache module reads the memory proportions of multiple first appearance images, and then selects multiple first appearance images with a memory proportion greater than or equal to a preset proportion from the multiple first appearance images as the appearance images to be processed. It should be explained that the image proportion reflects the richness of details in the image. For a clear image, it contains more details and accurate pixel information. Each pixel requires more bytes to store its color, brightness and other information, so its memory proportion will be higher; and since a blurred image has lost some details visually, its pixel information is relatively less rich. When stored, the specific color and brightness value of each pixel is not recorded as accurately as a clear image. For example, for some areas with complex color transitions, they can be approximated by fewer colors after blurring, thereby reducing the storage space required for each pixel, and the overall memory proportion is lower than that of a clear image.

[0026] However, the memory usage only considers the richness of details in the image. If the area with rich details in the image is an invalid area, the value of the image is not high. Therefore, in order to select the image with the highest analysis value from multiple appearance images to be processed, the image clarity recognition unit 32 performs clarity recognition on multiple appearance images to be processed. This process uses a lightweight semantic segmentation network to extract multiple judgment areas corresponding to each of the multiple appearance images to be processed. The multiple judgment areas are preset areas on the vehicle with low reflective interference. The preset areas can be understood as pre-identified high-value areas (headlights, wheels, license plates, etc.). If the clarity in the judgment area is high enough, it can be determined as a high-value image, which can be used as a reference image for subsequent identification of vehicle types.

[0027] When performing clarity recognition on multiple judgment images, the present application performs wavelet decomposition on multiple judgment areas to determine the noise distribution of the multiple judgment areas. After wavelet decomposition, the judgment areas will obtain the gradient energy of the low-frequency subband and the gradient energy of the high-frequency subband. The gradient energy of the low-frequency subband mainly reflects the contour information of the image, and the gradient energy of the high-frequency subband mainly reflects the detail information of the image. The noise has a smaller impact on the contour information of the image and a greater impact on the detail information. Therefore, for a blurred image, the gradient energy of its low-frequency subband will not change much, while the gradient energy of the high-frequency subband decreases due to the loss of detail information; therefore, the present application calculates the ratio of the gradient energy of the low-frequency subband of each of the multiple appearance images to be processed to the gradient energy of the high-frequency subband, so as to determine the detail loss amount of the multiple appearance images to be processed, and this detail loss amount also represents the clarity of the appearance image to be processed. At this time, the appearance image to be processed with the largest ratio result is selected as the benchmark appearance image for subsequent identification of the vehicle type.

[0028] After the image processing module 2 has determined the appearance image of the vehicle to be classified, it divides the appearance image into multiple feature images of the same size, extracts the vehicle features of the multiple feature images and analyzes their clarity, then compares the clarity of the multiple vehicle features with a preset clarity threshold, divides them into fuzzy features and clear features, and annotates the images. The annotated multiple vehicle features are sent to the classification processing module 3, the fuzzy features are fuzzy feature images, and the clear features are clear feature images. When analyzing the clarity of multiple feature images, since some feature images are not entirely composed of vehicles, the clarity of the feature images is determined based on the ratio of vehicle clear pixels to total vehicle pixels in the feature image. For example, if the feature image is composed of 1,000 pixels, of which 800 are vehicle total pixels and 600 are vehicle clear pixels, then its clarity is 600 / 800=0.75. In addition, for the case where the feature image does not contain a vehicle, since it does not contain relevant features of the vehicle, its clarity does not affect the judgment of the vehicle type itself. Therefore, this application sets its clarity to 1 (high clarity) to reduce its occupation of excessive subsequent image processing resources.

[0029] The classification processing module 3 classifies and analyzes the vehicles to be classified according to the image annotation of each feature image by the image processing module 2, so as to determine the vehicle type of the vehicle to be classified. Specifically, firstly, the ratio of the number of clear features to the number of fuzzy features in multiple vehicle features (feature images) is calculated to judge the overall clarity of the appearance image, and then it is judged whether the ratio result is greater than or equal to the preset ratio. The preset ratio here is set by the vehicle classification model. If the accuracy of the vehicle classification model is high, the preset ratio can be relatively set low. If the accuracy of the vehicle classification model is low, the preset ratio needs to be relatively set high. The vehicle classification model is a neural network model with image recognition function; if the ratio result is greater than or equal to the preset ratio, it means that the clarity of the appearance image can be recognized by the vehicle classification model, so at this time, the appearance image can be directly input into the vehicle classification model for image classification and recognition, so as to obtain the vehicle type of the vehicle to be classified. If the ratio result is less than the preset ratio, it means that the clarity of the appearance image is low, and the recognition accuracy of the vehicle classification model is low; at this time, the fuzzy feature needs to be gain processed to improve the clarity of the appearance image.

[0030] However, directly gaining the fuzzy features not only requires high hardware resources, but also in the process of gaining, in order to obtain a better gain effect, more noise may be introduced or some irrelevant features may be generated, resulting in poor correlation between the fuzzy features and the gained image; in order to solve this problem, the present application performs correlation analysis on the fuzzy features, matches the fuzzy features with a preset vehicle feature table, and obtains a first associated feature corresponding to the fuzzy features. The first associated feature is an associated image with a high degree of fuzzy feature similarity. The preset vehicle feature table contains clear images of various models. In addition, the preset vehicle feature table will be updated in real time, and new or changed models will be added thereto to ensure that it can be fully covered. When matching the fuzzy feature with the preset vehicle feature table, part of the key information in the fuzzy feature may be blurred, resulting in inaccurate expression of key pixels, which in turn causes poor matching results. Therefore, in order to improve the matching effect, the present application uses SIFT (Scale Invariant Feature Transform) to extract multiple key pixels and feature descriptors from the fuzzy feature. The feature descriptor is a data representation that describes specific features (edges, corners, textures) in an image. SIFT can accurately extract pixels at key positions when the image is blurred, but cannot change the properties of the pixels themselves. Therefore, there may be blurred points in the multiple key pixels extracted from the fuzzy feature. At this time, the present application constructs the multiple key pixels into a contour image, and then uses the feature descriptor to supplement the details of the contour image. Finally, the supplemented contour image is matched with the preset vehicle feature table to obtain the first associated feature. This step uses contour similarity instead of pixel similarity. In addition, the feature descriptor is used to supplement the detailed information of the contour to make the contour more accurate, and describes the fuzzy feature from the image as a whole, instead of focusing too much on the detailed information, thereby reducing the problem of poor matching effect caused by blurred pixels.

[0031] After obtaining the first associated feature, since the fuzzy feature lacks detailed information, the first associated feature does not necessarily match the original clear feature. Therefore, further rationality evaluation is needed to judge the correlation between the first associated feature and the original clear feature. If the correlation is high, it means that the first associated feature can well characterize the original fuzzy feature. Specifically: first identify the blur type of the blur feature. In the scenario of highway toll stations, the blur types mainly focus on motion blur, low-light blur and local occlusion blur. Each blur type has different texture features in the image, among which motion blur is manifested as directional texture features, low-light blur is manifested as non-uniform texture features, and local occlusion blur is manifested as discontinuous texture features. Then, according to the blur type of the blur feature, extract the corresponding texture features in the first associated features. At this time, the extracted texture features reflect the essence of the blur feature. For example, if the blur type of the blur feature is motion blur, extract its directional texture features from the first associated features. At this time, if the directional texture features of the first associated features have a high correlation with the clear features, it means that the directional texture features of the first associated features can well characterize the directional texture features in the blur feature, further indicating that the first associated features are more reasonable than the original clear features. When judging rationality, the present application calculates the similarity between the texture feature of the first associated feature and the clear feature to obtain a rationality score, and the calculation dimensions include structure, brightness and contrast, wherein the structural similarity represents the similarity of directional texture features, the brightness similarity represents the similarity of non-uniform texture features, and the contrast similarity represents the similarity of non-continuous texture features. Since the fuzzy type of the fuzzy feature may not only include one fuzzy type, the present application sets the weights of multiple calculation dimensions according to the influence of each calculation dimension in the fuzzy feature, and then performs weighted summation of the calculated similarities of the multiple dimensions according to the weights of the multiple calculation dimensions to obtain the final rationality score, wherein the influence of each calculation dimension in the fuzzy feature can be adopted by the probability distribution of each fuzzy type in the fuzzy feature output by the lightweight CNN. At this time, these probability distributions reflect the proportion of various fuzzy types in the fuzzy feature, that is, the influence of each calculation dimension in the fuzzy feature.

[0032] When the rationality score of the first associated feature and the original clear feature is greater than or equal to the preset rationality score, the first associated feature can replace the original fuzzy feature, and then the first associated feature and the original clear feature are input into the vehicle classification model for identification, so as to obtain a more accurate vehicle type.

[0033] In a possible implementation, before matching the fuzzy features with a preset vehicle feature table, in order to improve the accuracy of the match, the present application needs to perform image gain on the fuzzy features in advance to improve their clarity; specifically: first, the fuzzy features are decomposed by wavelet, and low-frequency sub-bands, medium-frequency sub-bands and high-frequency sub-bands are obtained. Among them, the low-frequency sub-band contains the outline of the image, the medium-frequency sub-band contains the medium-scale details of the image, and the high-frequency sub-band contains the fine details of the image. It should be noted that since blurring will significantly attenuate high-frequency information, resulting in a significant decrease in the energy of the high-frequency sub-band, the high-frequency sub-band is not suitable as a gain object. Although the low-frequency sub-band is least affected by blurring, it cannot provide the detail information required for matching. Therefore, the low-frequency sub-band is also not suitable as a gain object; the medium-frequency sub-band not only contains the key structural information of the vehicle, but also this information will be partially retained after blurring. Therefore, after performing wavelet decomposition on the fuzzy features, the present application extracts the intermediate frequency subband as the subsequent gain object; in the gain process, in order to avoid distortion of the fuzzy features after gain due to excessive gain, the gain coefficient needs to be adaptively adjusted. The present application determines the first gain coefficient of the fuzzy features according to the proportion of the number of fuzzy features in multiple vehicle features. If the proportion of fuzzy features is high, it means that the image is blurred as a whole and requires strong enhancement to restore key information. If the proportion of fuzzy features is low, it means that the image is clear as a whole and only requires slight enhancement to avoid noise amplification. Then, after gaining the intermediate frequency subband using the first gain coefficient, image reconstruction is performed to obtain the fuzzy features to be matched for matching. Finally, the fuzzy features to be matched are matched with the preset vehicle feature table to obtain the first associated features of the fuzzy features, thereby improving the matching accuracy of the fuzzy features. The present application adopts the wavelet domain limited enhancement technology for the gain of the intermediate frequency subband.

[0034] In a possible implementation, if the rationality score of the first associated feature and the original clear feature is less than the preset rationality score, it means that the first associated feature obtained by matching the fuzzy feature to be matched cannot well represent the original fuzzy feature. At this time, in order to improve the matching accuracy of the fuzzy feature, the present application needs to further perform image gain based on the original gain. Specifically: calculate the score difference between the current rationality score and the preset rationality score. The score difference at this time indicates the degree to which the original gain image needs further gain; then, based on the score difference, amplify and adjust the first gain coefficient to obtain the second gain coefficient, specifically, calculate the ratio of the score difference to the preset rationality score, then add 1 to the ratio result, and multiply it by the first gain coefficient to obtain the amplified second gain coefficient. This step is to further improve the clarity of the fuzzy feature on the one hand, and avoid image distortion caused by excessive gain on the other hand; then, use the second gain coefficient to re-enhance the fuzzy feature, and then pre-process the enhanced image and match it with the preset vehicle feature table to obtain the second associated feature, wherein the pre-processing is to use SIFT to obtain the image according to the second The key pixels and feature descriptors are extracted from the fuzzy features after gain coefficient gain, and then a contour image is constructed based on the key pixels and feature descriptors. When matching, the contour image is matched with the preset vehicle feature table to ensure the matching accuracy; then the texture features of the second associated features obtained are calculated for similarity with the clear features to obtain a second rationality score, wherein the similarity calculation includes structural similarity, brightness similarity and contrast similarity; at this time, it is judged whether the second rationality score is greater than or equal to the preset rationality score. If so, the second associated features and the clear features are input into the vehicle classification model for classification to obtain the vehicle type of the vehicle to be classified. If not, the above steps are repeated until the rationality score obtained by the rationality evaluation calculation is greater than or equal to the preset rationality score.

[0035] Finally, the grating sign 4 allocates the corresponding lane for the vehicle to be classified according to the vehicle type to be classified obtained by the classification processing module 3, and indicates the route of the vehicle to be classified through the grating indicator light to complete the classification guidance work of the vehicle to be classified.

[0036] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0037] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. An intelligent grating traffic control system, characterized in that: The system comprises a camera (1), an image processing module (2), a classification processing module (3) and a grating sign (4), wherein: The camera (1) is used to capture an appearance image of the vehicle to be classified; The image processing module (2) is used to extract a plurality of vehicle features from the appearance image, wherein the vehicle features include clear features and fuzzy features; The classification processing module (3) is used to perform classification analysis on the vehicle to be classified according to the plurality of vehicle characteristics, and determine the vehicle type of the vehicle to be classified; The grating sign (4) is used to indicate the passage road for the vehicle to be classified according to the vehicle type.

2. The system according to claim 1, characterized in that The classification analysis of the vehicle to be classified is performed based on the plurality of vehicle features to determine the vehicle type of the vehicle to be classified, specifically: calculating a ratio between the number of the sharp features and the number of the blurred features; Determine whether the ratio result is greater than or equal to a preset ratio; If the ratio result is greater than or equal to the preset ratio, the plurality of vehicle features are input into a vehicle classification model for classification to obtain the vehicle type of the vehicle to be classified.

3. The system according to claim 2, characterized in that The determining whether the ratio result is greater than or equal to a preset ratio specifically includes: If the ratio result is less than the preset ratio, performing correlation analysis on the fuzzy feature to obtain a first correlation feature of the fuzzy feature; Performing a rationality evaluation on the first associated feature of the fuzzy feature and the clear feature to obtain a rationality score; If the rationality score is greater than or equal to a preset rationality score, the clear feature and the first associated feature are input into the vehicle classification model for classification, and the vehicle type of the vehicle to be classified.

4. The system according to claim 3, characterized in that The performing association analysis on the fuzzy feature to obtain a first association feature of the fuzzy feature specifically includes: Performing wavelet decomposition on the fuzzy features to obtain intermediate frequency subbands; Determining a first gain coefficient of the fuzzy feature according to a proportion of the number of the fuzzy features in the plurality of vehicle features; Based on the first gain coefficient, reconstructing the image of the intermediate frequency sub-band to obtain a fuzzy feature to be matched; The fuzzy feature to be matched is matched with a preset vehicle feature table to obtain a first associated feature of the fuzzy feature.

5. The method according to claim 4, characterized in that The step of matching the to-be-matched fuzzy feature with a preset vehicle feature table to obtain a first associated feature of the fuzzy feature specifically includes: Extracting a plurality of key pixel points and feature descriptors of the fuzzy features to be matched; Constructing a contour image of the fuzzy feature to be matched according to the plurality of key pixel points and feature descriptors; The contour image is matched with the preset vehicle feature table to obtain a first associated feature of the fuzzy feature.

6. The system according to claim 3, characterized in that The rationality evaluation of the first associated feature of the fuzzy feature and the clear feature to obtain a rationality score is specifically: Identifying a fuzzy type of the fuzzy feature; Extracting texture features from the first associated features according to the fuzzy type; The texture feature and the clear feature are similarly calculated to obtain the rationality score, and the similarity calculation includes structural similarity, brightness similarity and contrast similarity.

7. The system according to claim 4, characterized in that After evaluating the rationality of the first associated feature of the fuzzy feature and the clear feature to obtain a rationality score, the method further includes: a. If the rationality score is less than the preset rationality score, then the score difference between the rationality score and the preset rationality score is calculated; b. Based on the score difference, amplify and adjust the first gain coefficient to obtain a second gain coefficient; c. Using the second gain coefficient, performing image enhancement, correlation feature matching, and rationality evaluation on the fuzzy feature to obtain a second correlation feature and a second rationality score; d. determining whether the second rationality score is greater than or equal to the preset rationality score; If yes, inputting the second associated feature and the clear feature into a vehicle classification model for classification to obtain the vehicle type of the vehicle to be classified; If not, repeat steps a to d until the rationality score calculated by the rationality evaluation is greater than or equal to the preset rationality score.

8. The system according to claim 1, characterized in that The image processing module (2) further comprises a cache unit (31) and an image definition recognition unit (31). The cache unit (31) is used to store the first appearance image of the vehicle to be classified, and select, according to the memory proportions of the plurality of first appearance images, a plurality of first appearance images whose memory proportions are greater than or equal to a preset proportion from the plurality of first appearance images as the appearance images to be processed; The image definition recognition unit (31) is used to perform definition recognition on the multiple appearance images to be processed, and select the image with the highest definition from the multiple appearance images to be processed as the appearance image.

9. The system according to claim 7, characterized in that The performing clarity recognition on the plurality of appearance images to be processed specifically includes: A lightweight semantic segmentation network is used to extract multiple determination areas corresponding to each of the multiple appearance images to be processed, wherein the multiple determination areas are preset areas on the vehicle with low reflective interference; Performing wavelet decomposition on a plurality of determination regions corresponding to each of the plurality of appearance images to be processed, to obtain gradient energies of the plurality of appearance images to be processed in low-frequency sub-bands and gradient energies of the plurality of high-frequency sub-bands; The ratio of the gradient energy of the low-frequency sub-band to the gradient energy of the high-frequency sub-band of each of the plurality of appearance images to be processed is calculated to obtain the clarity of the plurality of appearance images to be processed.

10. The system according to claim 1, characterized in that The extracting of multiple vehicle features from the appearance image specifically includes: Segmenting the appearance image into a plurality of feature images of the same size; Identify the clear pixel points of the vehicle and the total pixel points of the vehicle corresponding to each of the plurality of feature images; Determining the clarity of the plurality of feature images according to the ratio of the vehicle clear pixel points in the plurality of feature images to the total pixel points of the vehicle; If the clarity of the first feature image is greater than or equal to a preset clarity threshold, determining that the first feature image is a clear feature, the first feature image is any one of the plurality of feature images; If the clarity of the second feature image is less than a preset clarity threshold, it is determined that the second feature image is a blurred feature, and the second feature image is any one of the multiple feature images.

Images