Road code and bad driving behavior matching degree detection method

Abstract

According to the road code and bad driving behavior matching degree detection method provided by the invention, a corresponding relation library is constructed, a corresponding relation table of bad driving behavior codes, road types and vehicle types is established, then image recognition is performed on evidence pictures submitted by all monitoring equipment, and the road types and the vehicle types in the evidence pictures are obtained; and identifying and reading the bad driving behavior code in the watermark information, and automatically judging whether the bad driving behavior code in the watermark is accurately matched with the road type and the vehicle type in the picture or not according to the corresponding relation table.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic control, and in particular to a method for detecting the degree of matching between a road code and bad driving behavior. Background Art

[0002] With the acceleration of urbanization and the continuous increase in traffic flow, bad driving behaviors, such as speeding, driving in the wrong direction, running red lights, illegal parking, illegal lane changes, etc., have become important factors affecting road safety, smoothness and order. In order to effectively curb bad traffic driving behaviors and improve the level of traffic management, intelligent transportation systems (ITS) have been widely used. Among them, the intelligent monitoring system, as an important part of the intelligent transportation system, uses high-definition cameras to capture bad traffic driving behaviors on the road in real time and generate corresponding evidence pictures as a basis for subsequent processing. Figure 1 As shown, the intelligent monitoring system will add evidence information in the form of watermarks on the evidence images submitted by the user. The evidence information includes: collection time, vehicle license plate number, vehicle model, and bad driving behavior code. Bad driving behavior codes are defined by the management department according to relevant regulations, such as: 1742: Driving a medium-sized or larger truck on an urban expressway at a speed of 50% or less than 80%. 4712: Driving a school bus on a highway at a speed of 50%.

[0003] However, in practical applications, the judgment process of images captured by monitoring equipment often faces many challenges.

[0004] First, the diversity of road types makes the determination of bad driving behavior more complicated. Different road types (such as highways, urban roads, rural roads, etc.) have different traffic rules and speed limit requirements, so it is necessary to accurately identify the road type in the image to determine whether bad driving behavior applies.

[0005] Secondly, the accuracy of vehicle model recognition is also an important factor affecting the judgment of bad driving behavior. Different vehicle models have differences in speed, load, driving requirements, etc., and incorrect vehicle model recognition may lead to the misuse of bad behavior codes.

[0006] Finally, the accuracy and applicability of bad behavior codes are also key issues in the judgment process. Due to the continuous updating and improvement of traffic laws and regulations, the types and scope of application of bad behavior codes are also constantly changing. Therefore, it is necessary to ensure that the bad behavior codes are consistent with current traffic laws and regulations and accurately match specific bad driving behaviors and road types.

[0007] Because monitoring equipment is not installed uniformly across the country, but is installed in batches in each city at different times according to the needs of urban construction. Every time any judgment condition changes, the system maintenance personnel are required to update the data of the monitoring equipment they are responsible for. Once some equipment is missed, the evidence information in the evidence picture will be wrong. If the evidence information in the evidence picture submitted by the monitoring equipment is wrong, it will directly lead to the invalidity of the evidence picture, which will affect the subsequent processing of the vehicle. Most of the existing technologies are based on manual sampling of the information in the evidence picture, which cannot effectively and comprehensively cover all monitoring equipment. Summary of the invention

[0008] In order to solve the existing problem of mismatch between road codes and bad driving behaviors in evidence pictures submitted by monitoring equipment, which requires manual review to discover and leads to high maintenance costs, the present invention provides a road code and bad driving behavior matching detection method, which automatically discovers that the road code in the picture submitted by the monitoring equipment does not match the bad driving behavior, can more flexibly adapt to changes in various bad driving behavior judgment conditions, and efficiently reduces manual workload.

[0009] The technical solution of the present invention is as follows: a method for detecting the matching degree between a road code and bad driving behavior, characterized in that it comprises the following steps: S1: build a corresponding relationship library; The parameter data stored in the corresponding relationship library includes: road type, vehicle type, bad driving behavior and bad driving behavior code; In the corresponding relationship library, a corresponding relationship table between the bad driving behavior code and the road type and vehicle type is constructed. S2: Obtain the surveillance images collected and submitted by the surveillance equipment, recorded as: evidence images to be detected; S3: road type detection; Performing road type detection on the image in the evidence picture to be detected, identifying the road type of the road in the picture, which is recorded as: road type to be determined; The road types include: expressways, urban roads and other types of roads; S4: vehicle type recognition; Performing vehicle recognition on the image in the evidence picture to be detected to obtain the vehicle to be detected; Determine the vehicle type based on the vehicle to be detected, which is recorded as: the vehicle type type to be determined; The vehicle types include: medium-sized and larger passenger-carrying motor vehicles and passenger cars; S5: Identify bad driving behavior codes; Based on the information data in the evidence picture to be detected, the bad driving behavior code in the evidence picture submitted by the monitoring device is identified, which is recorded as: behavior code to be determined; S6: taking the road type to be determined and the vehicle type to be determined to be satisfied at the same time as a joint determination condition, and using the correspondence table to confirm whether there is a correspondence between the behavior code to be determined and the joint determination condition; If there is a corresponding relationship, it means that the behavior code to be judged is correct, and the matching degree detection result is output; this detection ends; Otherwise, it indicates that there is a possibility of error in the behavior code to be judged; the automatic detection ends, the matching detection result is output, and the manual review process is triggered.

[0010] It is further characterized by: In step S6, before triggering the manual review process, the following re-verification steps are performed: a1: judging the bad driving behavior of the vehicle to be detected in the picture based on the evidence picture to be detected, and recording the obtained result as: bad driving behavior for verification; a2: querying the bad driving behavior code corresponding to the bad driving behavior for verification based on the corresponding relationship database, recorded as: verification behavior code; a3: Determine whether the verification behavior code is consistent with the behavior code to be determined; If they are consistent, it means that the behavior code to be judged is correct, and the matching test result is output; this test ends Otherwise, the matching test result is output and the manual review process is triggered; It also includes the following steps: S7: monitor the changes of the relevant regulations on road types, vehicle types and bad driving behaviors in real time, and update the relevant contents in the corresponding relationship library in real time once any changes affecting the three parameter data occur; In step S3, a road type recognition model is constructed based on the CNN model to recognize the road type in the evidence image to be detected; In step S4, a vehicle detection model is constructed based on YOLO to identify the vehicle in the evidence image to be detected; a vehicle type classification model is constructed based on CNN to classify and identify the type of the vehicle to be detected; The matching test result includes: road type, vehicle type, bad behavior code and whether bad driving behavior is determined.

[0011] The present application provides a method for detecting the matching degree of road codes and bad driving behaviors. The method constructs a corresponding relationship library, which is a corresponding relationship table of bad driving behavior codes, road types and vehicle models, and then performs image recognition on all evidence pictures submitted by monitoring devices to obtain the road type and vehicle model in the evidence pictures, and reads the bad driving behavior code in the watermark information. According to the corresponding relationship table, it is automatically determined whether the bad behavior code in the watermark accurately matches the road type and vehicle model in the picture; because the watermark information of most evidence pictures is accurate, all correct evidence pictures can be quickly screened through the corresponding relationship table, and evidence pictures that may have mismatches between road codes and bad driving behaviors can be efficiently locked, effectively reducing the manual workload. Moreover, once any parameter in the road type, vehicle model, or bad driving behavior changes, flexible adaptation to the change can be achieved by maintaining the corresponding relationship library and the corresponding relationship table, without modifying too much logic code. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic diagram of the structure of the evidence diagram submitted for monitoring equipment in the prior art; Figure 2 This is a structural diagram of the road code and bad driving behavior matching detection system. DETAILED DESCRIPTION

[0013] like Figure 2 As shown, the road code and bad driving behavior matching detection method in this application can be used as a subsystem of the traffic management platform and set in the server group of the traffic management platform. After the intelligent monitoring system sends the evidence images collected by the monitoring equipment within its range to the traffic management platform, it is first sent to the subsystem running the road code and bad driving behavior matching detection method of this application for detection, and then subsequent processing is performed. Specifically, this application includes a road code and bad driving behavior matching detection method, which includes the following steps.

[0014] S1: build a corresponding relationship library; The parameter data stored in the corresponding relationship library include: road type, vehicle type, bad driving behavior and bad driving behavior code; In the correspondence library, a correspondence table between bad driving behavior codes and road types and vehicle models is constructed.

[0015] The corresponding relationship database used in this method can be built based on an existing database or can be independently built based on existing database technology. In the database and the corresponding relationship table, when specifically stored, the road type, bad driving behavior, and vehicle type are stored in the form of road code, bad driving behavior code, and vehicle type code, respectively. The specific code can be a pure digital code or a combination of letters and numbers. When comparing road types, bad driving behaviors, and vehicle types in calculations, the comparison is based on the code, which greatly simplifies the calculation process.

[0016] S2: Obtain the monitoring pictures collected and submitted by the monitoring equipment, which are recorded as: evidence pictures to be detected.

[0017] S3: road type detection; Perform road type detection on the image in the evidence picture to be detected, and identify the road type of the road in the picture, which is recorded as: road type to be determined; Road types include: expressways, urban roads and other types of roads.

[0018] Road type detection can be implemented directly using the trained models in other subsystems of the existing traffic management platform, or it can be implemented by analyzing the scene features in the picture based on the existing image processing technology and machine learning algorithms. The specific steps include the following steps.

[0019] Preprocessing: Use OpenCV to convert the image into grayscale, apply Gaussian blur to remove noise, and finally use Canny edge detection to extract image edges.

[0020] Feature extraction and classification: Design a CNN model, such as VGG, ResNet, etc., with preprocessed images as input. Use ImageNet or other datasets containing road type classification for pre-training, and then fine-tune on a specific road type dataset. Finally, the classification layer outputs the road type.

[0021] S4: vehicle type recognition; Perform vehicle recognition on the image in the evidence picture to be detected to obtain the vehicle to be detected; Determine the vehicle type based on the vehicle to be detected, which is recorded as: the vehicle type type to be determined; Vehicle types include: medium-sized and larger passenger-carrying motor vehicles and small passenger cars.

[0022] The vehicle detection model and vehicle type classification model can also be directly implemented using the trained models in other subsystems of the traffic control platform. This can reduce the cost of system construction. It can also be based on existing image processing technology and pre-trained vehicle type recognition models. Specifically include the following.

[0023] The vehicle detection method is to use target detection algorithms such as YOLO and SSD to directly detect the vehicle in the original image and obtain its bounding box. The vehicle image is cropped as the input for the next step.

[0024] The vehicle model classification method is: train a CNN-based vehicle model classification model, the input is the cropped vehicle image, and the model output is the vehicle model category.

[0025] S5: Identify bad driving behavior codes; Based on the information data in the evidence picture to be detected, the bad driving behavior code in the evidence picture submitted by the monitoring device is identified and recorded as: behavior code to be judged.

[0026] S6: The road type to be determined and the vehicle type to be determined are both satisfied as a joint determination condition, and a correspondence table is used to confirm whether there is a correspondence between the behavior code to be determined and the joint determination condition; If there is a corresponding relationship, it means that the behavior code to be judged is correct, and the matching test result is output; this test ends; Otherwise, it means that there is a possibility that the behavior code to be judged is wrong; the automatic detection ends, the matching test result is output, and the manual review process is triggered. The specific matching test results include: road type, vehicle type, bad behavior code, and whether bad driving behavior is confirmed.

[0027] In order to ensure the accuracy of the output test results and reduce the workload of manual verification, before triggering the manual review process, this method first implements the following re-verification steps to re-verify the accuracy of the test results.

[0028] a1: Based on the evidence image to be detected, the bad driving behavior of the vehicle to be detected in the image is judged, and the result is recorded as: bad driving behavior for verification; the specific bad driving behavior recognition can be achieved based on the existing trained bad driving behavior detection model or any known bad driving behavior detection method; a2: query the bad driving behavior code corresponding to the bad driving behavior for verification based on the corresponding relationship database, recorded as: verification behavior code; a3: Determine whether the behavior code for verification is consistent with the behavior code to be determined; If they are consistent, it means that the behavior code to be judged is correct, and the matching test result is output; this test ends Otherwise, the matching test result is output and the manual review process is triggered; For example, if a medium-sized or larger passenger-carrying motor vehicle exceeds the speed limit by more than 20% on a highway but the speeding code is mistakenly used as a speeding code for a passenger car, the manual review process will be triggered.

[0029] For images with inconsistent or questionable detection, the system will automatically submit them to manual review. Manual reviewers will make final judgments on the images based on their professional knowledge and experience to ensure the accuracy and fairness of bad driving behavior. Based on the test results and possible manual review results, the system outputs the final bad driving behavior judgment results, including road type, vehicle model, bad behavior code, and whether bad driving behavior is confirmed.

[0030] In order to ensure that the method can accurately test the evidence image, the method also includes the following steps.

[0031] S7: Real-time monitoring of changes in regulations on road types, vehicle types, and bad driving behaviors. Once any changes that affect the three parameter data occur, the relevant content in the corresponding relationship library is updated in real time to ensure that the corresponding relationship table in this method is strictly matched with the latest regulations. In addition, in this method, only the corresponding relationship library and the corresponding relationship table need to be maintained to easily and flexibly adapt to various changes, reducing the manual workload of system maintenance.

[0032] This method first detects the road type (such as highways, urban roads, etc.) and vehicle type (such as medium-sized and larger passenger-carrying motor vehicles, passenger cars, etc.) in the evidence image through image processing technology, and then automatically determines whether the bad behavior code accurately matches the current bad driving behavior based on the correspondence between the predefined bad behavior code and the vehicle type and road type. If the detection result shows a mismatch or there is a question, the manual review process is triggered to ensure the accuracy of the judgment. The detection method of the present invention not only reduces the workload of manual judgment, but also improves the accuracy and fairness of the judgment of bad driving behavior, providing strong technical support for intelligent traffic management and law enforcement. After using the technical solution of the present invention, the automatic recognition and matching of road types, vehicle types and bad behavior codes in the pictures captured by the monitoring equipment is realized by comprehensively using image processing technology, machine learning algorithms and manual review mechanisms, thereby improving the accuracy and efficiency of bad driving behavior judgment. At the same time, the method also provides a flexible update and maintenance mechanism based on the correspondence table to adapt to the continuous changes and improvements of traffic regulations.

Claims

1. A method for detecting the matching degree between road code and bad driving behavior, characterized in that: It includes the following steps: S1: build a corresponding relationship library; The parameter data stored in the corresponding relationship library includes: road type, vehicle type, bad driving behavior and bad driving behavior code; In the corresponding relationship library, a corresponding relationship table between the bad driving behavior code and the road type and vehicle type is constructed. S2: Obtain the surveillance images collected and submitted by the surveillance equipment, recorded as: evidence images to be detected; S3: road type detection; Performing road type detection on the image in the evidence picture to be detected, identifying the road type of the road in the picture, which is recorded as: road type to be determined; The road types include: expressways, urban roads and other types of roads; S4: vehicle type recognition; Performing vehicle recognition on the image in the evidence picture to be detected to obtain the vehicle to be detected; Determine the vehicle type based on the vehicle to be detected, which is recorded as: the vehicle type type to be determined; The vehicle types include: medium-sized and larger passenger-carrying motor vehicles and passenger cars; S5: Identify bad driving behavior codes; Based on the information data in the evidence picture to be detected, the bad driving behavior code in the evidence picture submitted by the monitoring device is identified, which is recorded as: behavior code to be determined; S6: taking the road type to be determined and the vehicle type to be determined to be satisfied at the same time as a joint determination condition, and using the correspondence table to confirm whether there is a correspondence between the behavior code to be determined and the joint determination condition; If there is a corresponding relationship, it means that the behavior code to be judged is correct, and the matching degree detection result is output; this detection ends; Otherwise, it indicates that there is a possibility of error in the behavior code to be judged; the automatic detection ends, the matching detection result is output, and the manual review process is triggered.

2. According to claim 1, a method for detecting the matching degree between a road code and bad driving behavior, characterized in that: In step S6, before triggering the manual review process, the following re-verification steps are performed: a1: judging the bad driving behavior of the vehicle to be detected in the picture based on the evidence picture to be detected, and recording the obtained result as: bad driving behavior for verification; a2: querying the bad driving behavior code corresponding to the bad driving behavior for verification based on the corresponding relationship database, recorded as: verification behavior code; a3: Determine whether the verification behavior code is consistent with the behavior code to be determined; If they are consistent, it means that the behavior code to be judged is correct, and the matching test result is output; this test ends Otherwise, the matching test result is output and the manual review process is triggered.

3. The method for detecting the matching degree between a road code and bad driving behavior according to claim 1, characterized in that: It also includes the following steps: S7: Monitor the changes in the relevant regulations on road types, vehicle types and bad driving behaviors in real time. Once any changes that affect the three parameter data occur, update the relevant content in the corresponding relationship library in real time.

4. The method for detecting the matching degree between a road code and bad driving behavior according to claim 1, characterized in that: In step S3, a road type recognition model is constructed based on the CNN model to identify the road type in the evidence image to be detected.

5. The method for detecting the matching degree between a road code and bad driving behavior according to claim 1, characterized in that: In step S4, a vehicle detection model is constructed based on YOLO to identify the vehicle in the evidence image to be detected; A vehicle type classification model is built based on CNN to classify and identify the type of vehicle to be detected.

6. The method for detecting the matching degree between a road code and bad driving behavior according to claim 1, characterized in that: The matching test result includes: road type, vehicle type, bad behavior code and whether bad driving behavior is determined.

Images