Method and system for monitoring illegal driving of non-motor tricycle

Through the two-stage target detection method and vehicle violation determination logic, the problem of low recognition rate of non-motorized tricycles in the existing technology is solved, and high-accuracy and efficient monitoring of tricycles' violations is achieved, and accurate traffic management data is provided.

CN118968453APending Publication Date: 2024-11-15ISA TECH CO LTD +1
View PDF 12 Cites 0 Cited by

Patent Information

Application Number
CN202411041784.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing image recognition system has a low recognition rate when identifying non-motorized tricycles, which is easy to mistakenly identify small four-wheelers as tricycles. It lacks a special tricycle recognition algorithm, resulting in low recognition accuracy.

Method used

The two-stage target detection method is adopted, first obtaining the road area boundary through the segmentation model, and then two identifications are performed through the target detection model to ensure the accuracy of the bounding box seating standard of the tricycle, and the vehicle violation judgment logic is implemented in combination with the motor vehicle lane boundary.

Benefits of technology

It improves the accuracy and efficiency of tricycles identification, reduces misidentification, and can intelligently monitor and record illegal behaviors of tricycles, providing accurate data support for traffic management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968453A_ABST
    Figure CN118968453A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of target recognition, and provides a non-motor tricycle illegal driving monitoring method and system, and the method comprises the following steps: obtaining to-be-recognized road image data; inputting the image data into a trained segmentation model to identify a road region in the image, and obtaining a lane boundary; inputting the image data into a target detection model, carrying out two-stage tricycle target detection, and carrying out second-stage tricycle target detection on an image segment output by a target detection result of the first stage to obtain bounding box coordinates of a tricycle body; and according to the boundary frame coordinates of the tricycle body and the motor vehicle lane boundary, executing vehicle law violation judgment logic, and outputting a judgment result. Through the two-stage recognition process, the recognition accuracy is improved, and the recognition efficiency is optimized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field related to target recognition, and in particular, to a monitoring method and system for illegal road occupation by non-motorized tricycles. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Non-motorized tricycles have become an important means of transportation for short-distance travel due to their flexibility, convenience and environmental protection. However, due to multiple factors such as unreasonable road design and drivers' weak awareness of traffic safety, there are various illegal phenomena such as non-motorized tricycles driving on motor vehicle lanes, running red lights, changing lanes and turning around at will, and parking on sidewalks, which not only reduces road traffic efficiency and causes traffic jams, but also greatly increases the risk of traffic accidents. Using artificial intelligence technology to monitor the illegal behavior of non-motorized tricycles can reduce the cost of traffic law enforcement, improve law enforcement efficiency, and help improve the level of intelligent traffic management and ensure road traffic safety.

[0004] The inventors found in their research that the existing image recognition systems mainly identify ordinary cars and motorcycles, and lack recognition algorithms specifically for tricycles. Commonly used image recognition models such as YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector) and Faster R-CNN usually use ordinary cars, motorcycles and pedestrians as the main objects in the training data set, and have a low recognition rate for tricycles in image recognition. Existing image recognition technology easily identifies small four-wheeled vehicles as tricycles, resulting in low recognition accuracy. Since tricycles and some small four-wheeled vehicles are similar in appearance, existing algorithms often make mistakes in distinguishing the two. For example, in the YOLOv3 and YOLOv4 models, due to the lack of refinement of input features and target labels, small four-wheeled vehicles are often mistakenly identified as tricycles. Summary of the invention

[0005] In order to solve the above problems, the present disclosure proposes a monitoring method and system for illegal driving of non-motorized tricycles. Through a two-stage recognition process, not only the recognition accuracy is improved, but also the recognition efficiency is optimized.

[0006] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0007] One or more embodiments provide a method for monitoring illegal driving of a non-motorized tricycle, comprising the following steps:

[0008] Acquire road image data to be identified;

[0009] Input the image data into the trained segmentation model to identify the road area in the image and obtain the lane boundary;

[0010] The image data is input into the target detection model to perform two-stage tricycle target detection, and the image segment output by the target detection result of the first stage is used for the second stage tricycle target detection to obtain the bounding box coordinates of the tricycle body;

[0011] According to the boundary box coordinates of the tricycle body and the motor vehicle lane boundary, the vehicle violation determination logic is executed and the determination result is output.

[0012] One or more embodiments provide a monitoring system for illegal driving of a non-motorized tricycle, comprising:

[0013] Image acquisition module: configured to acquire road image data to be identified;

[0014] Lane segmentation module: configured to input image data into the trained segmentation model to identify the road area in the image and obtain the lane boundary;

[0015] The target detection module is configured to input the image data into the target detection model, perform two-stage tricycle target detection, perform the second-stage tricycle target detection on the image segment output by the target detection result of the first stage, and obtain the bounding box coordinates of the tricycle body;

[0016] Judgment module: configured to execute vehicle violation judgment logic according to the boundary box coordinates of the tricycle body and the motor vehicle lane boundary, and output the judgment result.

[0017] An electronic device includes a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the above-mentioned monitoring method for illegal driving of non-motorized tricycles are completed.

[0018] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned method for monitoring illegal driving of non-motorized tricycles are completed.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The target detection process of the present invention adopts a two-stage target detection method, and the preliminary detection results of the target detection model in the first stage are used for secondary recognition of the target in the second stage, ensuring that each target identified as a tricycle has been strictly verified, reducing the occurrence of misidentification and improving the reliability of the recognition result; and, according to the body boundary frame coordinates of the tricycle and the motor vehicle lane boundary, the vehicle violation judgment logic is executed, and the illegal behavior of the tricycle can be intelligently identified and monitored. Through accurate target recognition and illegal behavior judgment logic, the dynamics of tricycles on the road can be monitored in real time, and their illegal behaviors can be identified and recorded in time, providing accurate monitoring data and decision support for traffic management departments.

[0021] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0023] Figure 1 is a flow chart of the monitoring method of Embodiment 1 of the present disclosure;

[0024] Figure 2 It is a flow chart of the two-stage three-wheeled vehicle target detection method of Example 1 of the present disclosure. DETAILED DESCRIPTION

[0025] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0027] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0028] Example 1

[0029] In the technical solutions disclosed in one or more embodiments, Figure 1 to Figure 2 As shown, a method for monitoring illegal driving of a non-motorized tricycle comprises the following steps:

[0030] Step 1: Obtain road image data to be identified;

[0031] Step 2: Input the image data into the trained segmentation model to identify the road area in the image and obtain the lane boundary;

[0032] Step 3: input the image data into the target detection model, perform two-stage tricycle target detection, perform the second-stage tricycle target detection on the image segments output by the target detection results of the first stage, and obtain the bounding box coordinates of the tricycle body;

[0033] Step 4: Execute the vehicle violation determination logic based on the bounding box coordinates of the tricycle body and the motor vehicle lane boundary, and output the determination result.

[0034] In the above scheme, the accuracy of tricycle targets is improved, and the misidentification of small four-wheeled vehicles and other vehicles is significantly reduced through efficient two-stage target detection. The preliminary detection results of the first-stage target detection model are used for secondary recognition of the targets in the second stage to ensure that each target identified as a tricycle has been strictly verified, reducing the occurrence of misidentification and improving the reliability of the recognition results; and, according to the body boundary box coordinates of the tricycle and the motor vehicle lane boundary, the vehicle violation judgment logic is executed, which can intelligently identify and monitor the illegal behavior of the tricycle. Through precise target recognition and illegal behavior judgment logic, the dynamics of tricycles on the road can be monitored in real time, and their illegal behaviors can be identified and recorded in time, providing accurate monitoring data and decision support for traffic management departments.

[0035] In step 1, the image data obtained may be road image data to be identified obtained from a road monitoring system, and the image data may be a single frame image or continuous frame images in a video stream.

[0036] In step 2, a lightweight FastSAM neural network can be used for lane segmentation, which includes the following process:

[0037] Step 21: Input the input image data into the trained FastSAM neural network model to obtain a segmentation mask with the same size as the input image, in which the value of each pixel represents the category to which the pixel belongs, including motor vehicle lane, non-motor vehicle lane, sidewalk, etc.

[0038] Step 22: performing noise processing, result smoothing processing, connected region analysis, and polygon simulation on the segmentation mask to obtain the first lane segmentation result;

[0039] Step 221, smoothing: using image processing techniques such as convolution filtering to smooth the mask to reduce noise and small area misclassification;

[0040] Step 222, connected region analysis: performing connected region analysis on the smoothed mask to identify the connected lane regions and remove isolated regions smaller than a set size;

[0041] Connected regions refer to pixel regions that belong to the same category and are interconnected in the segmentation mask. The connected region analysis algorithm can use 4-connected and 8-connected methods.

[0042] Step 223, polygon simulation: polygon simulation is performed on the connected area to further optimize the boundary of the lane area to ensure the accuracy and consistency of the segmentation result; specifically, an edge detection method is used to extract external boundary points for each connected area, and a polygon fitting algorithm is used to fit the boundary points into polygons as the border of the target area.

[0043] Optionally, the polygon fitting algorithm may employ the Douglas-Peucker algorithm, also known as the Ramer-Douglas-Peucker algorithm, which simplifies the boundary by recursively removing unnecessary points.

[0044] Step 23: registering and correcting the first lane segmentation result in combination with the map information so that the shape and position of the identified lane area are consistent with the actual road conditions;

[0045] Specifically, the registration and correction methods include:

[0046] Step 231: align the first segmentation result, i.e., the lane area in the segmentation mask, with the actual road map, and compare the positions and shapes of the two.

[0047] Step 232: the correction method is: by comparing the information in the map with the position and shape of the first segmentation result, the segmentation result is corrected according to the map information so that the shape and position of the lane area are consistent with the actual road conditions;

[0048] The correction process of the segmentation result by using the map information includes adjusting the boundary line, correcting the error, etc.

[0049] Through the above steps, the FastSAM neural network model can efficiently segment the road area and obtain accurate motor vehicle lane boundaries. This process ensures the accuracy and reliability of subsequent target detection and violation judgment.

[0050] In step 3, the first stage target detection model uses YOLO V10, and the target detection process includes the following steps:

[0051] Step 31: for the image data obtained in step 1, feature extraction is performed at different scales using the Feature Pyramid Network (FPN) of YOLO V10, and feature extraction is performed using convolutional layers and residual blocks at each scale;

[0052] The input image is passed through a series of convolutional layers and residual blocks for feature extraction;

[0053] Specifically, the convolution layer uses multiple convolution kernels (Filters) to perform convolution operations on the input image to extract features at different levels. The residual block is used to maintain the network depth while avoiding the gradient vanishing problem.

[0054] YOLO V10 uses the Feature Pyramid Network (FPN) to extract features at different scales to ensure the detection of objects of various sizes. The network generates multiple feature maps, each of which captures image information at different scales.

[0055] Step 32: Divide each feature map into grid units, and predict multiple bounding boxes in each grid unit;

[0056] Specifically, the bounding box includes the center coordinates (x, y), width (w), height (h), and confidence score (confidence score), where the confidence score indicates the probability that the bounding box contains a certain object;

[0057] Step 33, predicting the probability distribution of multiple categories on each bounding box;

[0058] Among them, the probability of each category indicates the possibility that the object in the bounding box belongs to this category.

[0059] Step 34: Perform a non-maximum suppression (NMS) operation to select high-confidence and non-overlapping bounding boxes as the final detection results by setting the confidence threshold and the IOU (Intersection Over Union) threshold. The detection results include the bounding box coordinates, category label, and confidence score of each target.

[0060] Among them, the category label is the vehicle type label, such as non-motorized three-wheeled vehicles, small four-wheeled vehicles, etc.;

[0061] In this embodiment, repeated detection can be reduced through non-maximum suppression (NMS) operation. YOLO V10 adopts the non-maximum suppression algorithm to output preliminary detection results.

[0062] A further technical solution is to extract the non-motorized tricycle image segment according to the target detection result of the first stage, comprising the following steps:

[0063] Step 3.1, filter out the targets with corresponding category labels, that is, the targets of "non-motorized tricycles", such as target 1 and target 3;

[0064] Step 3.2: According to the bounding box of the selected target, the corresponding image segment is cropped as the input image for the second stage target detection;

[0065] Specifically, according to the bounding box (50, 30, 100, 80) of target 1, a corresponding image segment is cropped from the original image; according to the bounding box (300, 220, 350, 270) of target 3, a corresponding image segment is cropped from the original image.

[0066] In step 3, the target detection model of the second stage adopts the PP-LCNet network model. The PP-LCNet network model includes multiple basic blocks, each of which includes a point convolution layer, a depth convolution layer, and an activation layer connected in sequence. Each basic block is stacked as a layer, and SE Block (Squeeze-and-Excitation Block) modules are set between the layers to enhance the feature representation between channels. The SE Block module performs global average pooling, full connection operation, ReLU activation, and Sigmoid activation operations in sequence.

[0067] In this embodiment, a depthwise separable convolution is used to reduce the computational complexity and the number of parameters.

[0068] Specifically, each basic block contains a 1×1 point convolution and a 3×3 depth convolution, as well as a ReLU activation function.

[0069] The PP-LCNet network model is used for the second stage of target detection, which includes the following steps:

[0070] Step 3-1: Normalize and resize the image segments extracted in step 3.2 to meet the input requirements of the PP-LCNet network model;

[0071] Specifically, the input image fragments are normalized and the pixel values ​​are scaled from 0-255 to 0-1 to meet the input requirements of the PP-LCNet model;

[0072] Adjust the input image segments to the fixed size required by the PP-LCNet model (e.g. 224x224, or other suitable sizes). This step can make the image segments of the same size, which is convenient for batch processing and speeds up calculations;

[0073] Step 3-2: Input the adjusted image segments into the trained PP-LCNet network model to obtain the category probability distribution corresponding to each image segment.

[0074] In this embodiment, the YOLO V10 model is first used to quickly detect the entire image, reducing the data range for subsequent processing; and the PP-LCNet network performs fine classification in a smaller range, thereby ensuring a high-precision recognition result. In this embodiment, the YOLO V10 model and the PP-LCNet network (or PP-LCNet) are superimposed to recognize the target image twice to improve the accuracy of recognition; it is equivalent to a pre-judgment by the YOLO V10 model, which reduces the processing range of the data, and then the pre-judgment result is judged again by the PP-LCNet network (or PP-LCNet), which can improve the recognition accuracy and the recognition efficiency.

[0075] A further technical solution also includes the steps of training the above segmentation model and the two-stage target detection model, which specifically include the following:

[0076] S1. Collect data and annotate to build training data sets;

[0077] Obtain real-time video data from traffic surveillance cameras, including non-motorized tricycles and motor vehicles, covering a variety of weather conditions (sunny, rainy, foggy, etc.), different time periods (daytime, nighttime), and various traffic flow conditions (peak hours, off-peak hours);

[0078] Obtain image data: extract frame images from the video and directly take static pictures through the camera to ensure data diversity;

[0079] Data labeling: Use professional labeling tools (such as LabelImg, VIA, etc.) to label the image (mark non-motorized tricycles, motor vehicles, and road areas). Each labeling box should accurately indicate the boundaries of the target.

[0080] S2. Model training and evaluation test: The three models mentioned above, FastSAM model, YOLO v10 model, and PP-LCNet model, are trained according to the data of the training set;

[0081] A further technical solution is to continuously learn during the use of the model, continuously update and optimize the model according to the recognition results of the model, and improve the detection accuracy and reliability.

[0082] In step 4, the boundary box of the non-motorized tricycle is mapped to the road area for judgment. The logic for judging vehicle violations is determined according to specific traffic rules. The judgment logic for occupying the road is as follows:

[0083] Completely in the motor vehicle lane: If the entire bounding box of the non-motor vehicle is within the motor vehicle lane area, it is determined to be driving on the motor vehicle lane;

[0084] Completely not in the motor vehicle lane: If the bounding box of the non-motor vehicle is completely outside the motor vehicle lane area, it is determined that it is not driving on the motor vehicle lane;

[0085] Partially in the motor vehicle lane: If the non-motor vehicle's boundary box is partially in the motor vehicle lane area, it can be determined whether it is driving on the motor vehicle lane based on specific circumstances (such as threshold settings). Finally, the determination result is output and an alarm is triggered or a violation record is automatically generated.

[0086] Example 2

[0087] Based on Example 1, this embodiment provides a monitoring system for illegal driving of non-motorized tricycles, including:

[0088] Image acquisition module: configured to acquire road image data to be identified;

[0089] Lane segmentation module: configured to input image data into the trained segmentation model to identify the road area in the image and obtain the lane boundary;

[0090] The target detection module is configured to input the image data into the target detection model, perform two-stage tricycle target detection, perform the second-stage tricycle target detection on the image segment output by the target detection result of the first stage, and obtain the bounding box coordinates of the tricycle body;

[0091] Judgment module: configured to execute vehicle violation judgment logic according to the boundary box coordinates of the tricycle body and the motor vehicle lane boundary, and output the judgment result.

[0092] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0093] Example 3

[0094] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps of the monitoring method for illegal driving of a non-motorized tricycle in embodiment 1 are completed.

[0095] Example 4

[0096] This embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the monitoring method for illegal driving of a non-motorized tricycle in embodiment 1 are completed.

[0097] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0098] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A method for monitoring illegal driving of non-motorized tricycles, characterized in that: The steps include: Acquire road image data to be identified; Input the image data into the trained segmentation model to identify the road area in the image and obtain the lane boundary; The image data is input into the target detection model to perform two-stage tricycle target detection, and the image segment output by the target detection result of the first stage is used for the second stage tricycle target detection to obtain the bounding box coordinates of the tricycle body; According to the boundary box coordinates of the tricycle body and the motor vehicle lane boundary, the vehicle violation determination logic is executed and the determination result is output.

2. A method for monitoring illegal driving of a non-motorized tricycle as claimed in claim 1, characterized in that: The lightweight FastSAM neural network is used for lane segmentation, which includes the following processes: Input the input image data into the trained FastSAM neural network model to obtain a segmentation mask with the same size as the input image; The segmentation mask is subjected to noise processing, smoothing processing, connected region analysis, and polygon simulation to obtain the first lane segmentation result; The first lane segmentation result is registered and corrected in combination with the map information so that the shape and position of the identified lane area are consistent with the actual road conditions.

3. A method for monitoring illegal driving of a non-motorized tricycle as claimed in claim 2, characterized in that: The mask is smoothed using convolution filtering to reduce noise and small area misclassifications; Perform connected region analysis on the smoothed mask to identify the coherent lane areas and remove isolated areas smaller than the set size. The connected areas are simulated with polygons, and the edge detection method is used to extract the external boundary points of each connected area. The boundary points are fitted into polygons using the polygon fitting algorithm as the border of the target area.

4. A method for monitoring illegal driving of a non-motorized tricycle as claimed in claim 2, characterized in that: The first lane segmentation result is registered and corrected. Specifically: the first segmentation result, i.e., the lane area in the segmentation mask, is registered with the actual road map, and the positions and shapes of the two are compared; the segmentation result is corrected according to the map information so that the shape and position of the lane area are consistent with the actual road conditions.

5. A method for monitoring illegal driving of a non-motorized tricycle as claimed in claim 1, characterized in that: The target detection model in the first stage uses YOLO V10. The target detection process includes the following steps: For the acquired image data, the feature pyramid network is used to extract features at different scales, and convolutional layers and residual blocks are used for feature extraction at each scale; Divide each feature map into grid cells and predict multiple bounding boxes in each grid cell; Predict the probability distribution of multiple categories on each bounding box; A non-maximum suppression operation is performed. By setting the confidence threshold and IOU threshold, high-confidence and non-overlapping bounding boxes are screened out as the final detection results. The detection results include the bounding box coordinates, category label, and confidence score of each target.

6. A method for monitoring illegal driving of a non-motorized tricycle as claimed in claim 1, characterized in that: Extracting non-motorized tricycle image segments based on the target detection results of the first stage includes the following steps: Filter out targets with corresponding category labels; According to the bounding box of the filtered target, the corresponding image segment is cropped as the input image for the second stage target detection.

7. A method for monitoring illegal driving of a non-motorized tricycle as claimed in claim 1, characterized in that: The target detection model in the second stage adopts the PP-LCNet network model. The PP-LCNet network model includes multiple basic blocks, each of which includes a point convolution layer, a depth convolution layer and an activation layer connected in sequence; each basic block is stacked as a layer, and SE Block modules are set between layers to enhance the feature representation between channels. The SE Block module performs global average pooling, full connection operations and activation operations in sequence.

8. A monitoring system for illegal driving of non-motorized tricycles, characterized in that: include: Image acquisition module: configured to acquire road image data to be identified; Lane segmentation module: configured to input image data into the trained segmentation model to identify the road area in the image and obtain the lane boundary; The target detection module is configured to input the image data into the target detection model, perform two-stage tricycle target detection, perform the second-stage tricycle target detection on the image segment output by the target detection result of the first stage, and obtain the bounding box coordinates of the tricycle body; Judgment module: configured to execute vehicle violation judgment logic according to the boundary box coordinates of the tricycle body and the motor vehicle lane boundary, and output the judgment result.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method for monitoring illegal driving of a non-motorized tricycle as described in any one of claims 1 to 7 are completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a method for monitoring illegal driving of a non-motorized tricycle as described in any one of claims 1-7.

Citation Information

Patent Citations

  • Non-motorized vehicle lane traffic violation driving identification method based on deep learning

    CN110909598A

  • Vehicle line pressing violation judgment method and device based on deep learning and storage medium

    CN111382704A

  • Illegal behavior recognition management method and system for two / three-wheeled vehicle

    CN111832519A

  • Traffic illegal behavior detection method for non-motor vehicles and drivers and system

    CN113160575A

  • Traffic person and vehicle non-target detection method for small target

    CN114332780A