Expressway ETC toll evasion inspection method and system based on image recognition

Abstract

The invention discloses an expressway ETC toll evasion inspection method and system based on image recognition, and relates to the technical field of image recognition, the system calculates a comprehensive behavior index Sabn through the construction of a comprehensive analysis module and the utilization of the characteristics of passing time difference, track integrity, image fuzziness and the like, and calculates the toll evasion index Sabn of the expressway ETC toll evasion index Sabn; and secondary comparison evaluation is carried out according to a set inspection threshold value J1. The comprehensive analysis module not only can comprehensively analyze abnormal vehicles, but also can calculate an average value and a standard deviation according to behavior data of normal vehicles, dynamically adjust an inspection threshold value, and optimize identification precision of toll evasion vehicles. For vehicles whose abnormal behavior indexes exceed an inspection threshold value, the system can automatically generate manual auditing prompt information to remind workers to carry out further ETC inspection. Through intelligent analysis and flexibility of the system, the system can effectively adapt to and identify toll evasion vehicles under complex road conditions, different weather conditions and various toll evasion modes.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a method and system for investigating ETC toll evasion on expressways based on image recognition. Background Art

[0002] As an efficient means of non-stop toll collection, the ETC electronic toll collection system has been widely used on highways. The ETC system uses wireless communication between on-board equipment and road equipment to enable vehicles to pass through toll booths without stopping, thus achieving high-speed passage. In order to improve the detection capability of ETC toll evasion, image recognition and big data analysis technology have become an effective way to solve this problem.

[0003] At present, although the existing ETC toll evasion detection system can already realize vehicle license plate recognition and charge based on the license plate number, there are two major problems with the current system. First, most of the existing systems rely on a single license plate recognition technology, resulting in low license plate recognition accuracy under certain conditions, especially when the vehicle speed is fast, the ambient light is poor, or the vehicle intentionally blocks the license plate, the system is difficult to accurately identify, resulting in misjudgment or missed judgment. Secondly, the existing detection system does not fully utilize the comprehensive information of the vehicle's driving trajectory and behavior, and only conducts inspections through a single license plate recognition and time matching, which cannot accurately judge complex toll evasion behaviors. Especially when vehicles pass through toll stations at high speeds, or when there are toll evasion methods such as detours and skipping stations, traditional systems are prone to monitoring blind spots, making it difficult to achieve comprehensive and effective monitoring. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method and system for investigating ETC toll evasion on expressways based on image recognition, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: including a vehicle information collection module, a feature extraction module, a time consistency analysis module, an abnormality monitoring module and a comprehensive analysis module; The vehicle information collection module collects vehicle image sets in real time by setting up visual sensors at ETC inspection points, and transmits the collected vehicle image sets and vehicle flow data to the inspection server by building an inspection server; The feature extraction module receives the vehicle image set and the traffic flow data in real time from the inspection server, pre-processes the vehicle image set, obtains the standard image set, and then extracts features from the standard image set through image recognition technology to obtain the inspection feature vector set; The time consistency analysis module obtains the travel time difference value △T by performing difference calculation based on the audit feature vector set, and sets the travel time threshold T1, and performs a preliminary comparison and evaluation between the travel time threshold T1 and the travel time difference value △, and marks the vehicles with abnormal travel time based on the preliminary comparison and evaluation results; The abnormality monitoring module calculates and outputs the trajectory integrity index Itraj and the image blur anomaly index △Bscore based on the audit feature vector set, and analyzes the trajectory integrity and image blur of the vehicle; The comprehensive analysis module constructs a comprehensive analysis algorithm formula, inputs the acquired travel time difference value △T, trajectory integrity index Itraj and image blur anomaly index △Bscore into the comprehensive analysis algorithm formula, calculates and outputs the comprehensive behavior index Sabn, and sets the audit threshold J1, then performs a secondary comparative evaluation on the audit threshold J1 and the comprehensive behavior index Sabn, and generates manual review prompt information based on the secondary comparative evaluation results.

[0006] Preferably, the vehicle information acquisition module includes a vehicle image acquisition unit and a data transmission unit; The vehicle image acquisition unit collects vehicle image sets in real time by setting up several inspection points at the high-speed ETC inspection place and installing several visual sensors at each inspection point; The visual sensor includes a camera; The data transmission unit transmits the collected vehicle image set to the inspection server by wirelessly connecting the visual sensor's own communication module to the inspection server through the 5G network.

[0007] Preferably, the feature extraction module includes a data processing unit and a feature extraction unit; The data processing unit receives the vehicle image set in real time from the audit server and automatically performs uniform processing on the vehicle image set through an image processing tool, wherein the uniform processing includes resolution uniformization, size uniformization, background interference removal and image feature enhancement processing to obtain a standard image set; The resolution unification and size unification are achieved by using image processing tools to unify the image resolutions of the vehicle image set into images of the same resolution, and at the same time scaling the size of the vehicle image set to the same size; The background interference removal is performed by using an image processing tool to crop the vehicle main area, remove the background part, and retain the vehicle main part; The image feature enhancement process is performed by performing contrast enhancement on the license plate characters in the vehicle image set; The feature extraction unit obtains the inspection feature vector set by extracting features from the standard image set, and then extracts character features from the license plates in the standard image set by using an optical character recognition algorithm (OCR), and labels each group of inspection feature vector sets according to the character features of each vehicle; The inspection feature vector set includes the license plate recognition interval time Tp, image matching degree Mg and image fuzziness Bg; The license plate recognition interval time Tp is to extract character features from the license plate in the standard image set based on the optical character recognition algorithm OCR, analyze the clarity according to the edge sharpness, color contrast and resolution of the character features, and generate the license plate recognition confidence Cp, and the range is [0,1]. The closer to 0, the more blurred the license plate is, and the lower the confidence is, which assists in judging whether the blurred license plate is used to evade tolls. The license plate recognition gap time Tp is recorded by the camera, the vehicle's license plate, timestamp and position, and compared with the continuous records of the standard image set of vehicles with the same license plate number, the recognition time interval of adjacent monitoring points is calculated to extract the license plate recognition gap time Tp. The specific algorithm formula is: , where t current Indicates the timestamp of the current audit point, t previous Indicates the timestamp of the last audit point; The image matching degree Mg is obtained by using a deep learning model to extract the global and local feature vectors of the standard image set, including color distribution, contour shape, and roof loading, and then comparing the global and local feature vectors taken at different inspection points, and using cosine similarity to calculate the extraction. The specific algorithm formula is: ; Wherein, F1 represents the global and local feature vectors of the first audit point, F2 represents the global and local feature vectors of the second audit point, and the image matching degree Mg ranges from [0,1]; The image blur Bg is obtained by performing edge detection on the standard image set and extracting the blur degree of the high standard image set, and the range is [0,1]. The closer to 1, the lower the image edge sharpness and the blurred details.

[0008] Preferably, the time consistency analysis module includes a time consistency detection unit and a consistency evaluation unit; The time consistency detection unit calculates and outputs the travel time difference value △T based on the license plate recognition gap time Tp in the inspection feature vector set of all vehicles, and analyzes the time deviation of the vehicle's time communication; The travel time difference value ΔT is calculated and outputted by the following algorithm formula: ; In the formula, N represents the total number of vehicles passing through the inspection point, N-1 represents the number of adjacent intervals, and Tp iIt represents the gap time of license plate recognition between the i-th inspection point, D represents the distance of the road section traveled by the vehicle, which is obtained by setting the distance of the inspection point, Vavg represents the average speed of the road section, which is obtained by ratioing the distance of the road section traveled by the vehicle and the gap time Tp of license plate recognition between the two inspection points.

[0009] Preferably, the consistency evaluation unit sets a pass time threshold T1 based on the normal pass time of the two inspection points, and then performs a preliminary comparative evaluation on the pass time threshold T1 and the pass time difference value △T, analyzes the abnormal time of vehicles passing through the inspection points, and marks the vehicles that pass abnormally. The specific evaluation content is as follows; When the travel time difference value △T> the travel time threshold T1, it means that the vehicle's travel time is abnormal, and the current vehicle is marked as an abnormal travel vehicle; When the travel time difference value △T≤ the travel time threshold T1, it means that the vehicle's travel time is normal, and the current vehicle is not marked at this time.

[0010] Preferably, the anomaly monitoring module includes a trajectory integrity monitoring unit and an image fuzziness analysis unit; The track integrity monitoring unit extracts the inspection feature vector of the license plate characters of the marked abnormal passing vehicles, and calculates and outputs the track integrity index Itraj of the abnormal passing vehicles according to the image matching degree Mg in the inspection feature vector, so as to analyze the track integrity of the abnormal vehicles; The trajectory integrity index Itraj is calculated and output by the following algorithm formula; ; In the formula, Mg i It represents the image matching degree of the i-th inspection point. The significance of the formula is to evaluate the integrity of the vehicle trajectory by comparing the image features of the vehicle at each monitoring point. If the image matching degree of the vehicle is low, it may mean that it bypasses some monitoring points, resulting in an incomplete trajectory.

[0011] Preferably, the image blur analysis unit extracts the inspection feature vector of the license plate characters of the abnormally passing vehicles from the marked abnormally passing vehicles, and calculates and outputs the image blur abnormality index △Bscore according to the image blur Bg in the inspection feature vector to monitor the abnormal blur of the vehicle image; The image blur abnormality index △Bscore is calculated and output by the following algorithm formula; ; In the formula, Bg i represents the image blur of the i-th inspection point, The balance factor representing blur anomaly is used to adjust the influence relationship between image blur and actual vehicle speed. The significance of the formula is that if the image blur is high and the vehicle is traveling at a high speed, it may be that the toll evasion vehicle is traveling at a high speed through the toll station, resulting in poor image quality.

[0012] Preferably, the comprehensive analysis module includes a comprehensive analysis unit and a comprehensive evaluation unit; The comprehensive analysis unit constructs a comprehensive algorithm formula, extracts the travel time difference value △T, the trajectory integrity index Itraj and the image blur abnormality index △Bscore of the marked abnormal passing vehicle, inputs them into the comprehensive analysis algorithm formula, calculates and outputs the comprehensive behavior index Sabn, and comprehensively analyzes the abnormal situation of the vehicle; The comprehensive behavior index Sabn is calculated and output by the following algorithm formula; ; Where a1, a2, and a3 represent the adjustment coefficients of the travel time difference value △T, the trajectory integrity index Itraj, and the image blur anomaly index △Bscore, respectively.

[0013] Preferably, the comprehensive evaluation unit extracts the comprehensive behavior index Sabn of a large number of normal vehicles, calculates the average value and standard deviation of normal vehicles, and obtains the inspection threshold J1. The specific algorithm formula is: ,in, Represents the average comprehensive behavior index Sabn value of normal vehicles, represents the standard deviation of the comprehensive behavior index Sabn of normal vehicles, k represents the adjustment coefficient of the standard deviation of the comprehensive behavior index Sabn of normal vehicles; After the audit threshold J1 and the comprehensive behavior index Sabn are compared and evaluated twice, the abnormal situation of the vehicle is comprehensively analyzed, and manual review prompt information is generated based on the evaluation results. The specific evaluation content is as follows; When the comprehensive behavior index Sabn> the inspection threshold J1, it means that the vehicle is abnormal, that is, there is a risk of evading fees. At this time, a manual review prompt message is generated, and the inspection server prompts the staff to conduct ETC inspection on the abnormal vehicle at the next inspection point; When the comprehensive behavior index Sabn ≤ the audit threshold J1, it means that the vehicle is normal, that is, there is no risk of evasion. At this time, there is no need to generate manual review prompt information and continue monitoring.

[0014] A method for checking ETC toll evasion on expressways based on image recognition, comprising the following steps: S1. By setting up visual sensors at ETC inspection points, vehicle image sets are collected in real time, and then the collected vehicle image sets and traffic flow data are transmitted to the inspection server by building an inspection server; S2, receiving the vehicle image set and the traffic flow data in real time in the inspection server, preprocessing the vehicle image set to obtain a standard image set, and then extracting features from the standard image set by image recognition technology to obtain an inspection feature vector set; S3. Based on the audit feature vector set, a difference calculation is performed to obtain a travel time difference value △T, and a travel time threshold T1 is set at the same time. The travel time threshold T1 is preliminarily compared and evaluated with the travel time difference value △, and based on the preliminary comparison and evaluation results, vehicles with abnormal travel time are marked; S4. Based on the audit feature vector set, the trajectory integrity index Itraj and the image blur abnormality index △Bscore are calculated and output respectively to analyze the trajectory integrity and image blur of the vehicle; S5. By constructing a comprehensive analysis algorithm formula, the obtained travel time difference value △T, trajectory integrity index Itraj and image blur anomaly index △Bscore are input into the comprehensive analysis algorithm formula, and the comprehensive behavior index Sabn is calculated and output, and the audit threshold J1 is set. Then, the audit threshold J1 and the comprehensive behavior index Sabn are subjected to a secondary comparative evaluation, and based on the secondary comparative evaluation result, a manual review prompt information is generated.

[0015] The present invention provides a method and system for checking ETC toll evasion on expressways based on image recognition. It has the following beneficial effects: (1) The system achieves high-precision monitoring of ETC toll evasion on highways by setting up visual sensors at ETC inspection points and combining the real-time collection and analysis of vehicle image sets and traffic flow data. The vehicle information collection module can obtain vehicle image sets in real time at different inspection points through cameras, and then transmit and store these images and traffic flow data through the constructed inspection server. The feature extraction module performs unified processing on the image set in the inspection server, including image resolution unification, size unification, background interference removal and image feature enhancement processing, and then extracts feature vectors such as license plate recognition confidence, license plate recognition gap time, image matching degree and image blur. By combining these multi-dimensional inspection feature vectors, the present invention can accurately evaluate the travel time, trajectory integrity and image quality of each vehicle, effectively identify toll-evading vehicles, and significantly improve the accuracy and real-time performance of ETC toll evasion monitoring.

[0016] (2) The system has comprehensively improved its ability to identify toll evasion through the time consistency analysis module and the anomaly monitoring module. The time consistency analysis module calculates the difference in vehicle travel time △T, uses the license plate recognition gap time Tp, and combines the average speed of the road section and the distance traveled by the vehicle to evaluate the difference between the travel time of each vehicle and the normal travel time. If the travel time difference value exceeds the set travel time threshold T1, the system automatically marks the vehicle as an abnormal travel vehicle and conducts further analysis. At the same time, the anomaly monitoring module analyzes the trajectory integrity and image blur, outputs the trajectory integrity index Itraj and the image blur anomaly index △Bscore respectively, and conducts in-depth anomaly analysis of abnormal travel vehicles. This multi-dimensional anomaly monitoring method effectively enhances the early warning and prevention and control capabilities of toll evasion, ensures that toll evasion vehicles can be discovered and manually reviewed in a timely manner, and greatly improves the accuracy and prevention and control effect of toll evasion monitoring.

[0017] (3) The system builds a comprehensive analysis module, uses features such as travel time difference, trajectory integrity and image blur, calculates the comprehensive behavior index Sabn, and performs a secondary comparative evaluation based on the set inspection threshold J1, thereby improving the intelligence and flexibility of the inspection system. The comprehensive analysis module can not only conduct a comprehensive analysis of abnormal vehicles, but also calculate the average and standard deviation based on the behavior data of normal vehicles, dynamically adjust the inspection threshold, and optimize the identification accuracy of toll-evading vehicles. For vehicles whose abnormal behavior index exceeds the inspection threshold, the system will automatically generate a manual review prompt message to remind staff to conduct further ETC inspections. The intelligent analysis and flexibility of the system enable it to effectively adapt to and identify toll-evading vehicles under complex road conditions, different weather conditions and multiple toll-evading methods, and has strong application adaptability and long-term stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the process of an expressway ETC toll evasion inspection system based on image recognition according to the present invention; Figure 2 The present invention is a schematic diagram of the steps of a method for investigating ETC toll evasion on expressways based on image recognition. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Example 1 See also Figure 1The present invention provides a highway ETC toll evasion inspection system based on image recognition. To achieve the above purpose, the present invention is implemented through the following technical solutions: including a vehicle information collection module, a feature extraction module, a time consistency analysis module, an abnormality monitoring module and a comprehensive analysis module; The vehicle information collection module collects vehicle image sets in real time by setting up visual sensors at ETC inspection points, and then transmits the collected vehicle image sets and traffic flow data to the inspection server by building an inspection server; The feature extraction module receives the vehicle image set and traffic flow data in real time from the inspection server, pre-processes the vehicle image set, obtains the standard image set, and then extracts features from the standard image set through image recognition technology to obtain the inspection feature vector set; The time consistency analysis module obtains the travel time difference value △T by performing difference calculation based on the audit feature vector set, and sets the travel time threshold T1. It also performs a preliminary comparison and evaluation between the travel time threshold T1 and the travel time difference value △, and marks vehicles with abnormal travel time based on the preliminary comparison and evaluation results. The abnormality monitoring module calculates and outputs the trajectory integrity index Itraj and the image blur anomaly index △Bscore based on the audit feature vector set, and analyzes the vehicle's trajectory integrity and image blur; The comprehensive analysis module constructs a comprehensive analysis algorithm formula, inputs the obtained travel time difference value △T, trajectory integrity index Itraj and image blur anomaly index △Bscore into the comprehensive analysis algorithm formula, calculates and outputs the comprehensive behavior index Sabn, and sets the audit threshold J1. The audit threshold J1 is then compared and evaluated with the comprehensive behavior index Sabn for the second time, and based on the results of the second comparison and evaluation, manual review prompt information is generated.

[0021] In this embodiment, the vehicle information collection module of the system collects vehicle image sets in real time at the ETC inspection point through visual sensors and transmits them to the inspection server; the feature extraction module preprocesses the vehicle images and extracts the inspection feature vector set to ensure data standardization and clarity; the time consistency analysis module calculates the travel time difference value △T based on the extracted feature vector, and accurately identifies vehicles with abnormal travel time by comparing it with the preset time threshold T1; the abnormal monitoring module further evaluates the vehicle trajectory integrity and image fuzziness, and detects the evasion behavior of abnormal vehicles by calculating the trajectory integrity index Itraj and the image fuzziness abnormality index △Bscore; the comprehensive analysis module calculates the comprehensive behavior index Sabn by inputting the above data, and compares and evaluates it twice with the inspection threshold J1, and finally generates manual review prompt information to further check potential evasion vehicles. This system significantly improves the accuracy and real-time performance of ETC evasion monitoring through the integration and analysis of multi-dimensional data. In complex traffic environments, it can effectively identify and warn evasion behaviors, greatly improving the monitoring capability and prevention level of the highway ETC system, and achieving the goals of optimizing toll management, improving traffic efficiency and reducing evasion losses.

[0022] Example 2 Specifically: the vehicle information acquisition module includes a vehicle image acquisition unit and a data transmission unit; The vehicle image acquisition unit collects vehicle image sets in real time by setting up several inspection points at the high-speed ETC inspection place and installing several visual sensors at each inspection point; Visual sensors include cameras; The data transmission unit wirelessly connects the visual sensor's built-in communication module to the inspection server through the 5G network, and transmits the collected vehicle image set to the inspection server.

[0023] In this embodiment, the system sets up multiple inspection stations at high-speed ETC inspection points, and each inspection station is equipped with several visual sensors, such as high-definition cameras, through which these sensors collect image sets of passing vehicles in real time. The collected image data is transmitted to the inspection server wirelessly using the 5G network through the communication module of the visual sensor itself. The data transmission unit ensures the efficiency and stability of data transmission, and ensures that the image data can be transmitted to the server in a timely manner under high traffic conditions. This implementation method can realize the real-time collection of a large amount of vehicle image information, and ensure the timeliness and stability of data transmission through the 5G high-speed network, which greatly improves the efficiency of the inspection process. Through the operation of this module, it can provide accurate and real-time basic data for subsequent image processing, feature extraction and evasion behavior analysis, and improve the system's monitoring ability and response speed in highway ETC evasion inspection. The improvement of the specific effect is reflected in: realizing real-time dynamic monitoring, reducing data transmission delays, improving the system's early identification and early warning capabilities for evasion behaviors, and ensuring efficient inspections in high-traffic environments.

[0024] Example 3 Specifically: the feature extraction module includes a data processing unit and a feature extraction unit; The data processing unit receives the vehicle image set in real time from the audit server and automatically performs uniform processing on the vehicle image set through image processing tools. The uniform processing includes resolution uniformization, size uniformization, background interference removal and image feature enhancement processing to obtain a standard image set. Resolution unification and size unification: By using image processing tools, the image resolution of the vehicle image set is unified to the same resolution image, and the size of the vehicle image set is scaled to the same size; Background interference removal: Use image processing tools to crop the main area of ​​the vehicle, remove the background part, and retain the main part of the vehicle; Image feature enhancement processing is performed by contrast enhancement of license plate characters in a vehicle image set; The feature extraction unit obtains the inspection feature vector set by extracting features from the standard image set, and then extracts character features from the license plate in the standard image set through the optical character recognition algorithm OCR, and labels each group of inspection feature vector sets according to the character features of each vehicle; The inspection feature vector set includes the license plate recognition gap time Tp, image matching degree Mg and image fuzziness Bg; The license plate recognition interval time Tp is based on the optical character recognition algorithm OCR to extract the character features of the license plate in the standard image set, and the clarity is analyzed according to the edge sharpness, color contrast and resolution of the character features to generate the license plate recognition confidence Cp, and the range is [0,1]. The closer to 0, the more blurred the license plate is, and the lower the confidence is, which assists in judging whether the blurred license plate is used to evade tolls. The license plate recognition gap time Tp is recorded by the camera, including the vehicle's license plate, timestamp and location. The camera compares the continuous records of the standard image set of vehicles with the same license plate number, calculates the recognition time interval of adjacent monitoring points, and extracts the license plate recognition gap time Tp. The specific algorithm formula is: , where t current Indicates the timestamp of the current audit point, t previous Indicates the timestamp of the last audit point; Image matching Mg uses a deep learning model to extract global and local feature vectors of the standard image set, including color distribution, contour shape, and roof loading, and then compares the global and local feature vectors taken at different inspection points, using cosine similarity calculation and extraction. The specific algorithm formula is: ; Wherein, F1 represents the global and local feature vectors of the first audit point, F2 represents the global and local feature vectors of the second audit point, and the image matching degree Mg ranges from [0,1]; The image blur Bg extracts the blur degree of the high-standard image set by performing edge detection on the standard image set, and the range is [0,1]. The closer it is to 1, the lower the image edge sharpness and the blurred details.

[0025] In this embodiment, the data processing unit of the system receives the image data transmitted from the vehicle information acquisition module in real time, and uses the image processing tool to perform a series of unified processing on the vehicle image, thereby obtaining a standardized image set. These processing ensures that image data of different sources and different qualities can be subsequently analyzed in a consistent format, reducing image noise and interference and improving image quality. The feature extraction unit then performs an in-depth analysis of the standard image set, extracts the license plate character features using the optical character recognition (OCR) algorithm, and generates an inspection feature vector set. Among them, the license plate recognition confidence and the license plate recognition gap time provide a preliminary analysis of the license plate clarity and the vehicle driving path, and the image matching degree compares the similarity of vehicle images taken at different inspection points through a deep learning algorithm to evaluate whether the vehicle's passage trajectory is complete. The image blur further evaluates the quality of the image and helps determine whether there are abnormal situations such as high-speed vehicle passing and blurred license plates. The purpose of this embodiment is to comprehensively improve the monitoring and judgment capabilities of vehicle evasion behavior through efficient image processing and feature extraction. The specific effects are reflected in the following aspects: First, the image unified processing ensures that the images collected under different conditions are consistent, thereby improving the accuracy of subsequent analysis.

[0026] Example 4 Specifically: the time consistency analysis module includes a time consistency detection unit and a consistency evaluation unit; The time consistency detection unit calculates and outputs the travel time difference value △T based on the license plate recognition gap time Tp in the inspection feature vector set of all vehicles, and analyzes the time deviation of the vehicle's time communication; The travel time difference value △T is calculated and output by the following algorithm formula; ; In the formula, N represents the total number of vehicles passing through the inspection point, N-1 represents the number of adjacent intervals, and Tp i It represents the gap time of license plate recognition between the i-th inspection point, D represents the distance of the road section traveled by the vehicle, which is obtained by setting the distance of the inspection point, Vavg represents the average speed of the road section, which is obtained by ratioing the distance of the road section traveled by the vehicle and the gap time Tp of license plate recognition between the two inspection points.

[0027] The consistency assessment unit sets the pass time threshold T1 based on the normal pass time of the two inspection points, and then makes a preliminary comparison and assessment between the pass time threshold T1 and the pass time difference value △T, analyzes the abnormal time of vehicles passing through the inspection points, and marks the vehicles that pass abnormally. The specific assessment contents are as follows; When the travel time difference value △T> the travel time threshold T1, it means that the vehicle's travel time is abnormal, and the current vehicle is marked as an abnormal travel vehicle; When the travel time difference value △T≤ the travel time threshold T1, it means that the vehicle's travel time is normal, and the current vehicle is not marked at this time.

[0028] In this embodiment, the system uses the time consistency detection unit and the consistency evaluation unit to work together to accurately analyze the vehicle's travel time at the ETC inspection point in order to identify possible evasion or other abnormal travel situations. The specific implementation method is: the time consistency detection unit analyzes the license plate recognition gap time Tp in the vehicle's inspection feature vector set to calculate the travel time difference value △T. The difference value is calculated based on the license plate recognition gap time of the vehicle passing through adjacent inspection points and the ratio of the road section distance and the average vehicle speed. The core purpose of this process is to analyze whether there is a travel time anomaly by comparing the difference between the actual travel time of the vehicle and the expected travel time. The consistency evaluation unit then sets a travel time threshold T1 and performs a preliminary comparison and evaluation with the calculated travel time difference value △T. If the travel time difference value is greater than the travel time threshold △T>T1, the system will mark the vehicle as an abnormal travel vehicle; conversely, if the travel time difference value is less than or equal to the travel time threshold △T≤T1, it is considered to be normal travel. Through the implementation of this module, the system can accurately analyze the time consistency of vehicles passing through the inspection point and promptly detect vehicles with abnormal travel time. Its main purpose is: on the one hand, by calculating the travel time difference value, it can effectively identify whether the vehicle has evaded tolls such as passing at high speeds and bypassing toll stations; on the other hand, by setting a reasonable travel time threshold, it ensures that the system accurately identifies normal vehicles and avoids misjudgment.

[0029] Example 5 Specifically: the anomaly monitoring module includes a trajectory integrity monitoring unit and an image fuzziness analysis unit; The track integrity monitoring unit extracts the inspection feature vector of the license plate characters of the marked abnormal passing vehicles, and calculates and outputs the track integrity index Itraj of the abnormal passing vehicles based on the image matching degree Mg in the inspection feature vector, and analyzes the track integrity of the abnormal vehicles; The trajectory integrity index Itraj is calculated and output by the following algorithm formula; ; In the formula, Mg i It represents the image matching degree of the i-th inspection point. The significance of the formula is to evaluate the integrity of the vehicle trajectory by comparing the image features of the vehicle at each monitoring point. If the image matching degree of the vehicle is low, it may mean that it bypasses some monitoring points, resulting in an incomplete trajectory.

[0030] The image blur analysis unit extracts the inspection feature vector of the license plate characters of the abnormally passing vehicles from the marked abnormally passing vehicles, and calculates and outputs the image blur abnormality index △Bscore based on the image blur Bg in the inspection feature vector to monitor the abnormal blur of the vehicle image; The image blur abnormality index △Bscore is calculated and output by the following algorithm formula; ; In the formula, Bg i represents the image blur of the i-th inspection point, The balance factor representing blur anomaly is used to adjust the influence relationship between image blur and actual vehicle speed. The significance of the formula is that if the image blur is high and the vehicle is traveling at a high speed, it may be that the toll evasion vehicle is traveling at a high speed through the toll station, resulting in poor image quality.

[0031] In this embodiment, the trajectory integrity monitoring unit of the system calculates the trajectory integrity index Itraj of the abnormal vehicle by extracting the inspection feature vector of the abnormal passing vehicle, especially the image matching degree Mg. This index is based on the image matching degree of different inspection points to judge the integrity of the vehicle trajectory. If the image matching degree of the vehicle is low, it means that there are large differences in the image features of the vehicle between different inspection points. It may be that some monitoring points are bypassed or the vehicle is not traveling along the normal route, resulting in an incomplete trajectory. This analysis can effectively identify abnormal vehicles that attempt to evade monitoring. The image blur analysis unit calculates the image blur abnormality index △Bscore based on the inspection feature vector of the vehicle, especially the image blur Bg. This abnormality index combines the image blur and the balance factor to analyze the image quality problem of the abnormal vehicle. The image blur abnormality index takes into account the relationship between the vehicle's driving speed and the image clarity. If the image blur is high and the vehicle is driving at a fast speed, it may indicate that the vehicle passes through a toll station during high-speed driving, resulting in a decrease in image quality, which is usually related to toll evasion. Through the collaboration of these two sub-units, the system can conduct in-depth analysis of abnormal vehicles from two dimensions: trajectory integrity monitoring calculates the trajectory integrity index to evaluate whether the vehicle's driving path between different inspection points is abnormal or detours. Image blur analysis calculates the image blur anomaly index to identify image quality problems and then determine whether there are violations such as highway toll evasion.

[0032] Example 6 Specifically: The comprehensive analysis module includes a comprehensive analysis unit and a comprehensive evaluation unit; The comprehensive analysis unit constructs a comprehensive algorithm formula, extracts the travel time difference value △T, the trajectory integrity index Itraj and the image blur abnormality index △Bscore of the marked abnormal passing vehicles, inputs them into the comprehensive analysis algorithm formula, calculates and outputs the comprehensive behavior index Sabn, and comprehensively analyzes the abnormal situation of the vehicle; The comprehensive behavior index Sabn is calculated and output by the following algorithm formula; ; Where a1, a2, and a3 represent the adjustment coefficients of the travel time difference value △T, the trajectory integrity index Itraj, and the image blur anomaly index △Bscore, respectively.

[0033] The comprehensive evaluation unit extracts the comprehensive behavior index Sabn of a large number of normal vehicles, calculates the average value and standard deviation of normal vehicles, and obtains the inspection threshold J1. The specific algorithm formula is: ,in, Represents the average comprehensive behavior index Sabn value of normal vehicles, represents the standard deviation of the comprehensive behavior index Sabn of normal vehicles, k represents the adjustment coefficient of the standard deviation of the comprehensive behavior index Sabn of normal vehicles; After the audit threshold J1 and the comprehensive behavior index Sabn are compared and evaluated for the second time, the abnormal situation of the vehicle is comprehensively analyzed, and manual review prompt information is generated based on the evaluation results. The specific evaluation content is as follows; When the comprehensive behavior index Sabn> the inspection threshold J1, it means that the vehicle is abnormal, that is, there is a risk of evading fees. At this time, a manual review prompt message is generated, and the inspection server prompts the staff to conduct ETC inspection on the abnormal vehicle at the next inspection point; When the comprehensive behavior index Sabn ≤ the audit threshold J1, it means that the vehicle is normal, that is, there is no risk of evasion. At this time, there is no need to generate manual review prompt information and continue monitoring.

[0034] In this embodiment, the system analyzes the parameters of multiple dimensions to evaluate the abnormal behavior of the vehicle, and generates manual review prompt information based on the comprehensive behavior index. The specific implementation method is as follows: the comprehensive analysis unit constructs a comprehensive algorithm formula, inputs the travel time difference value △T, the trajectory integrity index Itraj and the image blur abnormality index △Bscore of the marked abnormal passing vehicle into the comprehensive analysis formula, and calculates and outputs the comprehensive behavior index Sabn. The comprehensive behavior index comprehensively considers various key indicators and can effectively evaluate whether the vehicle has abnormal behavior, such as evasion. In the comprehensive algorithm formula, the adjustment coefficient is used to balance the impact of different indicators on the comprehensive behavior index to ensure a more accurate abnormal behavior assessment. The comprehensive evaluation unit extracts the comprehensive behavior index of a large number of normal vehicles, calculates the average comprehensive behavior index and standard deviation of normal vehicles, and sets the audit threshold J1 based on this. The threshold is calculated by the following algorithm formula, and can be dynamically adjusted according to the behavior pattern of normal vehicles to ensure that it is consistent with the actual traffic situation. By comparing the comprehensive behavior index Sabn with the audit threshold J1, the system can perform a secondary evaluation. If the comprehensive behavior index is greater than the inspection threshold, the system will prompt the staff to conduct further inspection of the vehicle; if the comprehensive behavior index is lower than the threshold, the vehicle is considered to be passing normally and no human intervention is required.

[0035] Example 7 See also Figure 1 and Figure 2 , a highway ETC toll evasion inspection method based on image recognition, comprising the following steps: S1. By setting up visual sensors at ETC inspection points, vehicle image sets are collected in real time, and then the collected vehicle image sets and traffic flow data are transmitted to the inspection server by building an inspection server; S2, receiving the vehicle image set and traffic flow data in real time in the inspection server, and preprocessing the vehicle image set to obtain a standard image set, and then extracting features from the standard image set through image recognition technology to obtain an inspection feature vector set; S3. Based on the audit feature vector set, a difference calculation is performed to obtain a travel time difference value △T, and a travel time threshold T1 is set at the same time. The travel time threshold T1 is preliminarily compared and evaluated with the travel time difference value △, and based on the preliminary comparison and evaluation results, vehicles with abnormal travel time are marked; S4. Based on the audit feature vector set, the trajectory integrity index Itraj and the image blur abnormality index △Bscore are calculated and output respectively to analyze the trajectory integrity and image blur of the vehicle; S5. By constructing a comprehensive analysis algorithm formula, the obtained travel time difference value △T, trajectory integrity index Itraj and image blur anomaly index △Bscore are input into the comprehensive analysis algorithm formula, and the comprehensive behavior index Sabn is calculated and output, and the audit threshold J1 is set. Then, the audit threshold J1 and the comprehensive behavior index Sabn are subjected to a secondary comparative evaluation, and based on the secondary comparative evaluation results, manual review prompt information is generated.

[0036] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A highway ETC toll evasion inspection system based on image recognition, characterized by: It includes vehicle information collection module, feature extraction module, time consistency analysis module, abnormality monitoring module and comprehensive analysis module; The vehicle information collection module collects vehicle image sets in real time by setting up visual sensors at ETC inspection points, and transmits the collected vehicle image sets and vehicle flow data to the inspection server by building an inspection server; The feature extraction module receives the vehicle image set and the traffic flow data in real time from the inspection server, pre-processes the vehicle image set, obtains the standard image set, and then extracts features from the standard image set through image recognition technology to obtain the inspection feature vector set; The inspection feature vector set includes the license plate recognition interval time Tp, image matching degree Mg and image fuzziness Bg; The time consistency analysis module obtains the travel time difference value △T by performing difference calculation based on the audit feature vector set, and sets the travel time threshold T1, and performs preliminary comparison and evaluation between the travel time threshold T1 and the travel time difference value △T, and marks the vehicles with abnormal travel time based on the preliminary comparison and evaluation results; The time consistency analysis module includes a time consistency detection unit and a consistency evaluation unit; The time consistency detection unit calculates and outputs the travel time difference value △T based on the license plate recognition gap time Tp in the inspection feature vector set of all vehicles, and analyzes the time deviation of the vehicle's time communication; The travel time difference value ΔT is calculated and outputted by the following algorithm formula: ； In the formula, N represents the total number of vehicles passing through the inspection point, N-1 represents the number of adjacent intervals, and Tp i represents the license plate recognition interval time between the i-th inspection points, D represents the road section distance traveled by the vehicle, and Vavg represents the average speed of the road section; The abnormality monitoring module calculates and outputs the trajectory integrity index Itraj and the image blur anomaly index △Bscore based on the audit feature vector set, and analyzes the trajectory integrity and image blur of the vehicle; The comprehensive analysis module constructs a comprehensive analysis algorithm formula, inputs the acquired travel time difference value △T, trajectory integrity index Itraj and image blur anomaly index △Bscore into the comprehensive analysis algorithm formula, calculates and outputs the comprehensive behavior index Sabn, and sets the audit threshold J1, then performs a secondary comparative evaluation on the audit threshold J1 and the comprehensive behavior index Sabn, and generates manual review prompt information based on the secondary comparative evaluation results.

2. According to the image recognition-based highway ETC toll evasion inspection system of claim 1, it is characterized by: The vehicle information acquisition module includes a vehicle image acquisition unit and a data transmission unit; The vehicle image acquisition unit collects vehicle image sets in real time by setting up several inspection points at the high-speed ETC inspection place and installing several visual sensors at each inspection point; The visual sensor includes a camera; The data transmission unit transmits the collected vehicle image set to the inspection server by wirelessly connecting the visual sensor's own communication module to the inspection server through the 5G network.

3. According to the image recognition-based highway ETC toll evasion inspection system of claim 2, it is characterized by: The feature extraction module includes a data processing unit and a feature extraction unit; The data processing unit receives the vehicle image set in real time from the audit server and automatically performs uniform processing on the vehicle image set through an image processing tool, wherein the uniform processing includes resolution uniformization, size uniformization, background interference removal and image feature enhancement processing to obtain a standard image set; The resolution unification and size unification are achieved by using image processing tools to unify the image resolutions of the vehicle image set into images of the same resolution, and at the same time scaling the size of the vehicle image set to the same size; The background interference removal is performed by using an image processing tool to crop the vehicle main area, remove the background part, and retain the vehicle main part; The image feature enhancement process is performed by performing contrast enhancement on the license plate characters in the vehicle image set; The feature extraction unit obtains the inspection feature vector set by extracting features from the standard image set, and then extracts character features from the license plates in the standard image set by using an optical character recognition algorithm (OCR), and labels each group of inspection feature vector sets according to the character features of each vehicle; The license plate recognition interval time Tp is to extract the character features of the license plate in the standard image set based on the optical character recognition algorithm OCR, analyze the clarity according to the edge sharpness, color contrast and resolution of the character features, and generate the license plate recognition confidence Cp, and the range is [0,1], to assist in judging whether the blurred license plate is used to evade tolls; The license plate recognition gap time Tp is recorded by the camera, the vehicle's license plate, timestamp and position, and compared with the continuous records of the standard image set of vehicles with the same license plate number, the recognition time interval of adjacent monitoring points is calculated to extract the license plate recognition gap time Tp. The specific algorithm formula is: , where t current Indicates the timestamp of the current audit point, t previous Indicates the timestamp of the last audit point; The image matching degree Mg is obtained by using a deep learning model to extract the global and local feature vectors of the standard image set, including color distribution, contour shape, and roof loading, and then comparing the global and local feature vectors taken at different inspection points, and using cosine similarity to calculate the extraction. The specific algorithm formula is: ; Wherein, F1 represents the global and local feature vectors of the first audit point, F2 represents the global and local feature vectors of the second audit point, and the image matching degree Mg ranges from [0,1]; The image blur Bg is obtained by performing edge detection on the standard image set to extract the blur degree of the high standard image set, and the range is [0, 1].

4. The highway ETC toll evasion inspection system based on image recognition according to claim 1 is characterized by: The consistency evaluation unit sets the passing time threshold T1 based on the normal passing time of the two inspection points, and then performs a preliminary comparison and evaluation on the passing time threshold T1 and the passing time difference value △T, analyzes the abnormal time of the vehicle passing the inspection point, and marks the abnormal passing vehicle. The specific evaluation content is as follows; When the travel time difference value △T> the travel time threshold T1, it means that the vehicle's travel time is abnormal, and the current vehicle is marked as an abnormal travel vehicle; When the travel time difference value △T≤ the travel time threshold T1, it means that the vehicle's travel time is normal, and the current vehicle is not marked at this time.

5. The highway ETC toll evasion inspection system based on image recognition according to claim 1 is characterized by: The anomaly monitoring module includes a trajectory integrity monitoring unit and an image fuzziness analysis unit; The track integrity monitoring unit extracts the inspection feature vector of the license plate characters of the marked abnormal passing vehicles, and calculates and outputs the track integrity index Itraj of the abnormal passing vehicles according to the image matching degree Mg in the inspection feature vector, so as to analyze the track integrity of the abnormal vehicles; The trajectory integrity index Itraj is calculated and output by the following algorithm formula; ； In the formula, Mg i represents the image matching degree of the i-th inspection point, and N represents the total number of vehicles passing through the inspection point.

6. The highway ETC toll evasion inspection system based on image recognition according to claim 5 is characterized by: The image blur analysis unit extracts the inspection feature vector of the license plate characters of the abnormally passing vehicles from the marked abnormally passing vehicles, and calculates and outputs the image blur abnormality index △Bscore based on the image blur Bg in the inspection feature vector to monitor the abnormal blur of the vehicle image; The image blur abnormality index △Bscore is calculated and output by the following algorithm formula; ； In the formula, Bg i Represents the image blur of the i-th inspection point.

7. The highway ETC toll evasion inspection system based on image recognition according to claim 1 is characterized by: The comprehensive analysis module includes a comprehensive analysis unit and a comprehensive evaluation unit; The comprehensive analysis unit constructs a comprehensive algorithm formula, extracts the travel time difference value △T, the trajectory integrity index Itraj and the image blur abnormality index △Bscore of the marked abnormal passing vehicle, inputs them into the comprehensive analysis algorithm formula, calculates and outputs the comprehensive behavior index Sabn, and comprehensively analyzes the abnormal situation of the vehicle; The comprehensive behavior index Sabn is calculated and output by the following algorithm formula; ； Where a1, a2, and a3 represent the adjustment coefficients of the travel time difference value △T, the trajectory integrity index Itraj, and the image blur anomaly index △Bscore, respectively.

8. The highway ETC toll evasion inspection system based on image recognition according to claim 7 is characterized by: The comprehensive evaluation unit extracts the comprehensive behavior index Sabn of a large number of normal vehicles, calculates the average value and standard deviation of normal vehicles, and obtains the inspection threshold J1. The specific algorithm formula is: ,in, Represents the average comprehensive behavior index Sabn value of normal vehicles, represents the standard deviation of the comprehensive behavior index Sabn of normal vehicles, k represents the adjustment coefficient of the standard deviation of the comprehensive behavior index Sabn of normal vehicles; After the audit threshold J1 and the comprehensive behavior index Sabn are compared and evaluated twice, the abnormal situation of the vehicle is comprehensively analyzed, and manual review prompt information is generated based on the evaluation results. The specific evaluation content is as follows; When the comprehensive behavior index Sabn> the inspection threshold J1, it means that the vehicle is abnormal. At this time, a manual review prompt message is generated, and the inspection server prompts the staff to conduct an ETC inspection on the abnormal vehicle at the next inspection point; When the comprehensive behavior index Sabn ≤ the audit threshold J1, it indicates that the vehicle is normal. At this time, there is no need to generate manual review prompt information and monitoring continues.

9. A highway ETC toll evasion inspection method based on image recognition, applied to a highway ETC toll evasion inspection system based on image recognition as claimed in any one of claims 1 to 8, characterized in that: The following steps are involved: S1. By setting up visual sensors at ETC inspection points, vehicle image sets are collected in real time, and then the collected vehicle image sets and traffic flow data are transmitted to the inspection server by building an inspection server; S2, receiving the vehicle image set and the traffic flow data in real time in the inspection server, preprocessing the vehicle image set to obtain a standard image set, and then extracting features from the standard image set by image recognition technology to obtain an inspection feature vector set; S3. Based on the audit feature vector set, a difference calculation is performed to obtain a travel time difference value △T, and a travel time threshold T1 is set at the same time. The travel time threshold T1 is preliminarily compared and evaluated with the travel time difference value △, and based on the preliminary comparison and evaluation results, vehicles with abnormal travel time are marked; S4. Based on the audit feature vector set, the trajectory integrity index Itraj and the image blur abnormality index △Bscore are calculated and output respectively to analyze the trajectory integrity and image blur of the vehicle; S5. By constructing a comprehensive analysis algorithm formula, the obtained travel time difference value △T, trajectory integrity index Itraj and image blur anomaly index △Bscore are input into the comprehensive analysis algorithm formula, and the comprehensive behavior index Sabn is calculated and output, and the audit threshold J1 is set. Then, the audit threshold J1 and the comprehensive behavior index Sabn are subjected to a secondary comparative evaluation, and based on the secondary comparative evaluation result, a manual review prompt information is generated.

Images