Multi-camera video joint analysis method and system for vehicle tracking

Abstract

The application relates to the technical field of vehicle tracking, and provides a multi-lens video joint analysis method and system for vehicle tracking. The method comprises the following steps: acquiring position information of a target camera and acquiring corresponding video data sets in real time through a traffic management system; performing vehicle detection to acquire a vehicle detection result; transmitting the vehicle detection result to a cloud processor, acquiring a preset synchronous detection standard, synchronously adjusting the vehicle detection result, and acquiring a standard vehicle detection result; performing vehicle correlation analysis based on the position information to acquire a vehicle correlation result; acquiring a positioning result of a target vehicle relative to the target camera, the positioning result corresponding to a camera; and generating a continuous tracking instruction to continuously track the target vehicle. The application solves the technical problems of slow tracking efficiency and high delay of vehicles in the prior art, and achieves the technical effects of improving information transmission efficiency and more efficiently realizing cross-lens tracking.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle tracking, in particular to a multi-lens video joint analysis method and system for vehicle tracking. BACKGROUND

[0002] Real-time monitoring and tracking technology of vehicles is one of the important technologies of intelligent transportation systems. The computer realizes vehicle detection and tracking by analyzing video sequences recorded by a camera without human intervention or with only a little human intervention. Multi-lens video joint can make the monitoring lens whole, greatly improve the information transmission efficiency, provide data for cross-video intelligent analysis, and more efficiently realize cross-lens tracking, global real-time monitoring of large scenes, and rapid backtracking of historical events.

[0003] In summary, the prior art has the problem of slow tracking efficiency and high delay for vehicles. SUMMARY

[0004] Therefore, it is necessary to provide a multi-lens video joint analysis method and system for vehicle tracking, which can improve the information transmission efficiency and more efficiently realize cross-lens tracking.

[0005] In a first aspect, the present application provides a multi-lens video joint analysis method for vehicle tracking, comprising: acquiring position information of Q target cameras of a target road through a traffic management system, and acquiring corresponding Q video data sets in real time, 2≤Q; inputting the Q video data sets into Q edge nodes of the Q target cameras respectively, performing vehicle detection, and acquiring Q vehicle detection results; transmitting the Q vehicle detection results to a cloud processor, acquiring a preset synchronous detection standard, synchronously adjusting the Q vehicle detection results based on the preset synchronous detection standard, and acquiring P standard vehicle detection results, 2≤P≤Q; performing vehicle correlation analysis on the P standard vehicle detection results based on the position information, and acquiring a vehicle correlation result; acquiring a target vehicle to be tracked, matching the target vehicle with the target correlation result, acquiring a positioning result of the target vehicle for the Q target cameras, the positioning result corresponding to an mth camera; generating a continuous tracking instruction, and sending the continuous tracking instruction to Q-m cameras after the mth camera to continuously track the target vehicle.

[0006] In a second aspect, the present application provides a multi-lens video joint analysis system for vehicle tracking, comprising: a video data set obtaining module, configured to obtain position information of Q target cameras of a target road and real-time obtain corresponding Q video data sets through a traffic management system, 2Q; a vehicle detection result obtaining module, configured to input the Q video data sets into Q edge nodes of the Q target cameras respectively, perform vehicle detection, and obtain Q vehicle detection results; a standard vehicle detection result obtaining module, configured to transmit the Q vehicle detection results to a cloud processor, obtain a preset synchronous detection standard, perform synchronous adjustment on the Q vehicle detection results based on the preset synchronous detection standard, obtain P standard vehicle detection results, 2PQ; a vehicle association result obtaining module, configured to perform vehicle association analysis on the P standard vehicle detection results based on the position information, and obtain a vehicle association result; a positioning result corresponding module, configured to obtain a target vehicle to be tracked, match the target vehicle with a target association result, obtain a positioning result of the target vehicle for the Q target cameras, and the positioning result corresponds to an mth camera; and a target vehicle continuous tracking module, configured to generate a continuous tracking instruction, send the continuous tracking instruction to Q-m cameras after the mth camera, and continuously track the target vehicle.

[0007] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0008] Firstly, through the traffic management system, the position information of Q target cameras of a target road is acquired, and Q corresponding video data sets are acquired in real time, 2≤Q; secondly, the Q video data sets are respectively input into Q edge nodes of the Q target cameras, vehicle detection is performed, and Q vehicle detection results are acquired; next, the Q vehicle detection results are transmitted to a cloud processor, a preset synchronous detection standard is acquired, the Q vehicle detection results are synchronously adjusted based on the preset synchronous detection standard, P standard vehicle detection results are acquired, 2≤P≤Q; then, based on the position information, vehicle correlation analysis is performed on the P standard vehicle detection results, and a vehicle correlation result is acquired; then, a target vehicle to be tracked is acquired, the target vehicle is matched with the target correlation result, a positioning result of the target vehicle for the Q target cameras is acquired, the positioning result corresponds to an mth camera; finally, a continuous tracking instruction is generated, the continuous tracking instruction is sent to Q-m cameras after the mth camera, and the target vehicle is continuously tracked. The application solves the technical problems of slow tracking efficiency and high delay of vehicles in the prior art, and achieves the technical effects of improving information transmission efficiency and more efficiently realizing cross-lens tracking.

[0009] The above description is only a summary of the technical scheme of the application. In order to more clearly understand the technical means of the application, the application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A flowchart of a multi-lens video joint analysis method for vehicle tracking in an embodiment;

[0011] Figure 2 An adjustment target camera flowchart of a multi-lens video joint analysis method for vehicle tracking in an embodiment;

[0012] Figure 3 A structural block diagram of a multi-lens video joint analysis system for vehicle tracking in an embodiment.

[0013] Explanation of reference numerals: video data set acquisition module 11, vehicle detection result acquisition module 12, standard vehicle detection result acquisition module 13, vehicle correlation result acquisition module 14, positioning result corresponding module 15, and target vehicle continuous tracking module 16. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and not to limit the present application.

[0015] As shown in Figure 1 The present application provides a multi-camera video joint analysis method for vehicle tracking, characterized in that it comprises:

[0016] Through the traffic management system, the position information of Q target cameras of the target road is obtained, and the corresponding Q video data sets are obtained in real time, 2≤Q.

[0017] Vehicle tracking refers to the behavior of tracking and monitoring the driving route of the target vehicle; multi-camera video joint refers to the work of multiple cameras forming a camera, which needs to consider that the most time-consuming is the fusion and matching of frame images in the splicing process, and the time of subsequent work should be appropriately reduced, which is to improve the work efficiency and play the role of multi-camera splicing. Because the line is relatively fixed when installing the camera, generally speaking, the position of the camera is not changed, so when extracting multiple camera images, the transformation matrix of the first frame image should be recorded, and then the deformed frame image is fused and cropped by using the first frame image; through the analysis of the multi-camera video joint analysis method, the multi-camera video joint analysis method for vehicle tracking is obtained, which achieves multi-angle observation of the vehicle and improves the recognition degree of the vehicle, and maintains the technical effect of traffic safety.

[0018] The traffic management system is a comprehensive traffic management system that effectively integrates advanced information technology, data communication transmission technology, electronic sensing technology, control technology and computer technology in the whole ground traffic management system, and plays a role in a wide range, all directions, real-time, accurate and efficient. Through the query of the traffic management system, the position information of Q target cameras of the target road is obtained, wherein the target road refers to the traffic lane selected by the staff for studying the multi-camera video joint, and multiple target cameras are arranged in the target road, denoted as Q, and the target road has Q cameras at different positions. At the same time, the multiple target cameras are turned on to shoot the target road, and the corresponding Q video data sets can be obtained in real time, wherein 2≤Q. By accessing the traffic management system, the Q target cameras in the target road are obtained, and the corresponding Q video data sets are obtained by shooting at the same time, which provides data support for obtaining vehicle detection results subsequently.

[0019] Input the Q video data sets into the Q edge nodes of the Q target cameras respectively, perform vehicle detection, and obtain Q vehicle detection results;

[0020] The edge node refers to a service platform constructed at the network edge close to the user, which provides storage, computing, network, and other resources, and sinks part of the key business applications to the access network edge to reduce the width and delay loss caused by network transmission and multi-level forwarding. In this application, the edge node refers to the photographed part intersected by each camera in the Q target cameras; vehicle detection refers to selecting and detecting vehicles in the Q video data sets to obtain Q vehicle detection results, which lays a foundation for subsequent acquisition of standard vehicle detection results by obtaining vehicle detection results.

[0021] In the first edge node, key frames are extracted from the first video data set to obtain a first detection image;

[0022] Features are extracted from the first detection image through a feature extraction channel to obtain detection features;

[0023] Based on the detection features, the first detection image is divided into a bounding box through a target positioning channel;

[0024] Through a vehicle recognition channel, the target in the bounding box is classified, and a first vehicle detection result of the first edge node is obtained according to the classification result.

[0025] The first edge node refers to any one of the edge nodes, denoted as the first edge node. The first video data set, the first detection image, and the first vehicle detection result are all data corresponding to the first edge node. Key frame extraction refers to obtaining key frames in a shot by using shot segmentation on a video sequence, which in this application refers to obtaining key frames in the first video data set to obtain the first detection image. The feature extraction channel refers to a system that uses a computer to extract information characteristic of the first detection image. The detection features include but are not limited to the length, width, and other characteristics of the vehicle. The target positioning channel refers to a positioning module of the target vehicle in the first detection image, which is used to determine the bounding box of the first detection image. In target detection, a bounding box is usually used to describe the target position. The vehicle recognition channel refers to determining the detection and recognition of the vehicle in the bounding box. The target in the bounding box is classified through the vehicle recognition channel, and the first vehicle detection result of the first edge node is obtained according to the classification result. By obtaining the first vehicle detection result, the Q vehicle detection results in the Q edge nodes are obtained according to the above technology.

[0026] In the Q edge nodes, historical image data in the history time is called to obtain Q historical image data sets;

[0027] The first historical image data set in the Q historical image data set is used to construct the first vehicle identification model in the first edge node, and the first vehicle identification model is synchronized to the vehicle identification channel;

[0028] Continue to construct vehicle identification models in other edge nodes respectively, and synchronize to the corresponding vehicle identification channel.

[0029] The historical image data refers to the image data photographed in the Q edge nodes in the past time period. The historical image data is analyzed and integrated to obtain the Q historical image data set. One of the Q historical image data sets is selected as the first historical image data set to construct the first vehicle identification model in the first edge node. The first vehicle identification model is used to identify the image of the vehicle in the first historical image data set. The first vehicle identification model is synchronized to the vehicle identification channel. According to the above technical means, vehicle identification models are constructed in other edge nodes and synchronized to the vehicle identification channel corresponding to the edge node. By constructing the vehicle identification model and synchronizing it to the vehicle identification channel, the vehicle detection result corresponding to the edge node is obtained.

[0030] The first historical image data set is subjected to feature extraction and boundary box division, and the existence of a vehicle in the boundary box is labeled according to the division result to obtain a first vehicle identification result set;

[0031] Based on the BP neural network, the network structure of the first vehicle identification model is constructed;

[0032] The first historical image data set and the first vehicle identification result set are used to train the first vehicle identification model. The model parameters are adjusted and updated during the training process to obtain the first vehicle identification model that meets the preset requirements.

[0033] The first historical image data set is subjected to feature extraction, bounding box division, and annotation of whether a vehicle exists in the bounding box according to the division result, to obtain a first vehicle identification result set; a network structure of the first vehicle identification model is constructed based on a BP neural network in machine learning, the first vehicle identification model includes a plurality of simple units simulating neurons in the human brain, the first vehicle identification model can form parameters such as connection weights and threshold values between the simple units in a supervised training process, the first vehicle identification model after training can perform complex nonlinear logical operations according to input data, and output the operated data, the input data is the first historical image data set, and the output data is a vehicle identification result. The first vehicle identification model is trained and verified by the first historical image data set and the first vehicle identification result set, to obtain the first vehicle identification model, the first vehicle identification model is supervised trained by the first historical image data set, when the model output result tends to be in a convergence state, the output result accuracy of the first vehicle identification model is verified by the first vehicle identification result set, to obtain a preset requirement, which can be defined by a person skilled in the art based on actual conditions, for example, when the output result accuracy is greater than or equal to 95%, the first vehicle identification model meets the requirement. When the output result accuracy of the first vehicle identification model meets the preset requirement, the first vehicle identification model is obtained.

[0034] The Q vehicle detection results are transmitted to a cloud processor, a preset synchronous detection standard is obtained, the Q vehicle detection results are synchronously adjusted based on the preset synchronous detection standard, P standard vehicle detection results are obtained, 2≤P≤Q;

[0035] The cloud processor is a virtual computer, which can be connected to the computer through the Internet for use and management, the Q vehicle detection results are transmitted to the cloud processor, and a preset synchronous detection standard is obtained, wherein the preset synchronous detection standard is set by a worker, which is not limited here, synchronous adjustment means normalization of different data, that is, the Q vehicle detection results come from different cameras, and there may be various problems such as unclearness caused by light, angle inconsistency, shielding and the like, each result is synchronously adjusted from this angle, if normalization cannot be performed, the data is deleted, P standard vehicle detection results are obtained, 2≤P≤Q, P is less than Q because if normalization cannot be performed, the data will be deleted, and the remaining data P will be less than Q. The Q vehicle detection results are synchronously adjusted based on the preset synchronous detection standard, to obtain P standard vehicle detection results, which contributes to subsequent vehicle correlation analysis.

[0036] A preset time threshold is obtained according to the preset synchronous detection standard;

[0037] extracting collection time from the Q vehicle detection results, calculating a time difference between collection data and detection time, when the time difference is less than a preset time threshold, retaining the Q vehicle detection results;

[0038] According to the preset synchronous detection standard, a preset image clarity threshold is obtained;

[0039] The Q vehicle detection results are subjected to clarity detection, when the clarity detection result is greater than the preset image clarity threshold, the Q vehicle detection results are retained, or image enhancement processing is performed;

[0040] When the time difference is greater than or equal to the preset time threshold, and / or when the clarity detection result is less than or equal to the preset image clarity threshold, the corresponding vehicle detection result is eliminated.

[0041] The preset time threshold is a time threshold set by the staff, the collection time of the Q vehicle detection results is extracted, the time difference between the collection data and the detection time is calculated, when the time difference is less than the preset time threshold, the Q vehicle detection results are retained; then according to the image clarity threshold set by the staff, if the clarity detection result is greater than the preset image clarity threshold, the Q vehicle detection results are retained, if there are unclear or occlusion problems in the vehicle detection results due to light, but not to the extent that it cannot be repaired, image enhancement or restoration technology is used to solve these problems to improve the quality of the vehicle detection results as much as possible; when the time difference is greater than or equal to the preset time threshold, and / or when the clarity detection result is less than or equal to the preset image clarity threshold, the corresponding vehicle detection result is eliminated.

[0042] Based on the position information, vehicle correlation analysis is performed on the P standard vehicle detection results to obtain vehicle correlation results;

[0043] Based on the position information, vehicle correlation analysis is performed on the P standard vehicle detection results, for example, vehicle detection results in the first edge node and vehicle monitoring results in other edge nodes are subjected to correlation analysis, that is, from large-scale data, the process of discovering the implicit relationship and rule between objects, in this application, it refers to obtaining images of the same vehicle from the edge node to obtain vehicle correlation results, which lays a foundation for subsequent tracking of vehicles by obtaining vehicle correlation results.

[0044] Obtaining a target vehicle to be tracked, matching the target vehicle with the target correlation result, obtaining a positioning result of the target vehicle for the Q target cameras, the positioning result corresponding to the mth camera;

[0045] The target vehicle to be tracked is selected by the staff, and the target vehicle is compared and matched with the target association result. The target vehicle is in the camera shooting area of the mth camera in the Q target cameras, and the corresponding mth camera is determined according to the camera shooting area. The vehicle position of the target vehicle is located, which supports the subsequent continuous tracking.

[0046] A test vehicle is arranged on the target road, and the test vehicle includes a positioning device.

[0047] The test vehicle is started, and the test vehicle is positioned in real time through the positioning device to obtain a real-time positioning result.

[0048] Data of the test vehicle is collected through the Q target cameras to obtain a data collection result.

[0049] The data collection result is sent to the cloud processor for vehicle positioning, and the delay coefficient is obtained in combination with the real-time positioning result.

[0050] A test vehicle is arranged on the target road, and the test vehicle can be any vehicle and includes a positioning device. The test vehicle is started, and the test vehicle is positioned in real time through the positioning device to obtain real-time position information of the vehicle, which is recorded as a real-time positioning result. Data of the test vehicle is collected through the Q target cameras to obtain a data collection result, wherein the data collection result includes position information of the test vehicle and has a time identifier. The data collection result is sent to the cloud processor for vehicle positioning, and the delay coefficient is obtained in combination with the real-time positioning result, that is, the delay information of the data collection result is obtained according to the real-time positioning result and the data collection result of the test vehicle at the same time, which is recorded as a delay coefficient. By obtaining the delay coefficient, the continuous tracking of the target vehicle is contributed.

[0051] As shown in Figure 2 The adjustment content for synchronously adjusting the Q vehicle detection results is obtained, and a first adjustment index is obtained.

[0052] The continuity of the vehicle association result is evaluated, and a continuity evaluation result is obtained as a second adjustment index.

[0053] The Q target cameras are adjusted in combination with the first adjustment index and the second adjustment index.

[0054] According to the above-mentioned adjustment content of the synchronization adjustment of the Q vehicle detection results, a first adjustment index is obtained, the continuity of the vehicle association result is evaluated, and a continuity evaluation result is obtained, wherein the continuity evaluation result refers to whether the target vehicle continues to travel, as a second adjustment index; and the Q target cameras are adjusted in combination with the first adjustment index and the second adjustment index. This makes a preparation for subsequent tracking of the target vehicle.

[0055] A continuous tracking instruction is generated, and the continuous tracking instruction is sent to the Q-m cameras after the mth camera to continuously track the target vehicle.

[0056] The continuous tracking instruction refers to a command for tracking the target vehicle issued by the cloud processor; the continuous tracking instruction is sent to the Q-m cameras after the mth camera to continuously track the target vehicle. By synchronously adjusting the Q vehicle detection results, a correct tracking instruction is obtained, and the purpose of accurately tracking the target vehicle is achieved. The present application solves the technical problems of slow tracking efficiency and high delay of vehicles in the prior art, and achieves the technical effects of improving information transmission efficiency and more efficiently realizing cross-lens tracking.

[0057] As shown in Figure 3 The embodiment of the present application also provides a multi-lens video joint analysis system for vehicle tracking, which comprises:

[0058] A video data set obtaining module 11 is configured to obtain position information of Q target cameras of a target road through a traffic management system, and obtain corresponding Q video data sets in real time, 2≤Q;

[0059] A vehicle detection result obtaining module 12 is configured to input the Q video data sets into Q edge nodes of the Q target cameras respectively, perform vehicle detection, and obtain Q vehicle detection results;

[0060] A standard vehicle detection result obtaining module 13 is configured to transmit the Q vehicle detection results to a cloud processor, obtain a preset synchronous detection standard, synchronously adjust the Q vehicle detection results based on the preset synchronous detection standard, and obtain P standard vehicle detection results, 2≤P≤Q;

[0061] A vehicle association result obtaining module 14 is configured to perform vehicle association analysis on the P standard vehicle detection results based on the position information, and obtain a vehicle association result;

[0062] A positioning result corresponding module 15 is configured to acquire a target vehicle to be tracked, match the target vehicle with a target association result, acquire a positioning result of the target vehicle for the mth camera among the Q target cameras;

[0063] A target vehicle continuous tracking module 16 is configured to generate a continuous tracking instruction and send the continuous tracking instruction to the Q-m cameras after the mth camera to continuously track the target vehicle.

[0064] Further, the embodiment of the present application further comprises:

[0065] A detection image acquisition module is configured to perform key frame extraction on a first video data set at a first edge node to acquire a first detection image;

[0066] A detection feature acquisition module is configured to perform feature extraction on the first detection image through a feature extraction channel to acquire detection features;

[0067] A bounding box division module is configured to divide a bounding box of the first detection image through a target positioning channel based on the detection features;

[0068] A vehicle detection result acquisition module is configured to perform category determination on a target in the bounding box through a vehicle recognition channel and acquire a first vehicle detection result of the first edge node according to a determination result.

[0069] Further, the embodiment of the present application further comprises:

[0070] A historical image data set acquisition module is configured to acquire historical image data in a historical time in the Q edge nodes to acquire Q historical image data sets;

[0071] A vehicle recognition model synchronization module is configured to use a first historical image data set in the Q historical image data sets to construct a first vehicle recognition model in the first edge node and synchronize the first vehicle recognition model to the vehicle recognition channel;

[0072] A vehicle recognition model continuous synchronization module is configured to continue to construct vehicle recognition models in other edge nodes respectively and synchronize to corresponding vehicle recognition channels.

[0073] Further, the embodiment of the present application further comprises:

[0074] a vehicle recognition result set obtaining module, configured to perform feature extraction and boundary box division on the first historical image data set, and label whether a vehicle exists in the boundary box according to the division result, and obtain a first vehicle recognition result set;

[0075] a network structure constructing module, configured to construct a network structure of the first vehicle recognition model based on a BP neural network;

[0076] a vehicle recognition model module, configured to train the first vehicle recognition model by using the first historical image data set and the first vehicle recognition result set, and adjust and update model parameters during the training process to obtain the first vehicle recognition model meeting preset requirements.

[0077] Further, the embodiment of the application further includes:

[0078] a preset time threshold obtaining module, configured to obtain a preset time threshold according to the preset synchronous detection standard;

[0079] a vehicle detection result retaining module, configured to extract collection time from the Q vehicle detection results, calculate a time difference between the collection data and the detection time, and retain the Q vehicle detection results when the time difference is less than the preset time threshold;

[0080] an image definition threshold obtaining module, configured to obtain a preset image definition threshold according to the preset synchronous detection standard;

[0081] a definition detection module, configured to perform definition detection on the Q vehicle detection results, and retain the Q vehicle detection results or perform image enhancement processing when the definition detection result is greater than the preset image definition threshold;

[0082] a vehicle detection result eliminating module, configured to eliminate the corresponding vehicle detection result when the time difference is greater than or equal to the preset time threshold and / or when the definition detection result is less than or equal to the preset image definition threshold.

[0083] Further, the embodiment of the application further includes:

[0084] a test vehicle setting module, configured to set a test vehicle on the target road, and the test vehicle comprises a positioning device;

[0085] The real-time positioning result acquisition module is configured to start the test vehicle, perform real-time positioning on the test vehicle through the positioning device, and acquire a real-time positioning result.

[0086] The data acquisition result acquisition module is configured to perform data acquisition on the test vehicle through the Q target cameras and acquire a data acquisition result.

[0087] The delay coefficient acquisition module is configured to send the data acquisition result to the cloud processor for vehicle positioning, acquire a delay coefficient in combination with the real-time positioning result.

[0088] Further, the embodiments of the present application further include:

[0089] The adjustment index acquisition module is configured to acquire an adjustment content for synchronously adjusting the Q vehicle detection results and acquire a first adjustment index.

[0090] The continuity evaluation module is configured to perform continuity evaluation on the vehicle association result and acquire a continuity evaluation result as a second adjustment index.

[0091] The camera adjustment module is configured to adjust the Q target cameras in combination with the first adjustment index and the second adjustment index.

[0092] For specific embodiments of the multi-lens video joint analysis system for vehicle tracking, reference can be made to the embodiments of the multi-lens video joint analysis method for vehicle tracking described above, which will not be repeated here. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above modules.

[0093] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0094] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A multi-camera video joint analysis method for vehicle tracking, characterized in that, The application relates to a traffic management system and a vehicle tracking method thereof. The application comprises the following steps: acquiring position information of Q target cameras on a target road through the traffic management system, and acquiring corresponding Q video data sets in real time, 2<=Q; inputting the Q video data sets into Q edge nodes of the Q target cameras respectively, performing vehicle detection, and acquiring Q vehicle detection results; transmitting the Q vehicle detection results to a cloud processor, acquiring a preset synchronous detection standard, synchronously adjusting the Q vehicle detection results based on the preset synchronous detection standard, and acquiring P standard vehicle detection results, 2<=P<=Q; performing vehicle correlation analysis on the P standard vehicle detection results based on the position information, and acquiring a vehicle correlation result; acquiring a target vehicle to be tracked, matching the target vehicle with the target correlation result, and acquiring a positioning result of the target vehicle for the Q target cameras, the positioning result corresponding to an mth camera; generating a continuous tracking instruction, sending the continuous tracking instruction to Q-m cameras after the mth camera, and continuously tracking the target vehicle; setting a test vehicle on the target road, the test vehicle comprising a positioning device; starting the test vehicle, performing real-time positioning on the test vehicle through the positioning device, and acquiring a real-time positioning result; performing data acquisition on the test vehicle through the Q target cameras, and acquiring a data acquisition result; 2. The method of claim 1, wherein, sending the data acquisition result to the cloud processor for vehicle positioning, combining the real-time positioning result, and acquiring a delay coefficient. The vehicle detection comprises the following steps: extracting key frames from a first video data set in a first edge node, and acquiring a first detection image; performing feature extraction on the first detection image through a feature extraction channel, and acquiring detection features; dividing a bounding box of the first detection image through a target positioning channel based on the detection features; 3. The method of claim 2, wherein, performing category determination on a target in the bounding box through a vehicle recognition channel, and acquiring a first vehicle detection result of the first edge node according to the determination result. The vehicle recognition channel comprises the following steps: in the Q edge nodes, acquiring Q historical image data sets in a historical time; using a first historical image data set in the Q historical image data sets to construct a first vehicle recognition model in the first edge node, and synchronizing the first vehicle recognition model to the vehicle recognition channel; 4. The method of claim 3, wherein, continuously constructing vehicle recognition models in other edge nodes and synchronizing the vehicle recognition models to corresponding vehicle recognition channels. The first vehicle recognition model in the first edge node is constructed by the following steps: performing feature extraction and bounding box division on the first historical image data set, marking whether a vehicle exists in the bounding box according to a division result, and acquiring a first vehicle recognition result set; constructing a network structure of the first vehicle recognition model based on a BP neural network; training the first vehicle recognition model by using the first historical image data set and the first vehicle recognition result set, adjusting and updating model parameters in the training process, and obtaining the first vehicle recognition model satisfying a preset requirement.

5. The method of claim 1, wherein, The Q vehicle detection results are synchronously adjusted based on the preset synchronous detection standard, including: According to the preset synchronous detection standard, a preset time threshold is obtained; From the Q vehicle detection results, the collection time is extracted, and the time difference between the collection data and the detection time is calculated. When the time difference is less than the preset time threshold, the Q vehicle detection results are retained; According to the preset synchronous detection standard, a preset image clarity threshold is obtained; The Q vehicle detection results are subjected to clarity detection. When the clarity detection result is greater than the preset image clarity threshold, the Q vehicle detection results are retained, or image enhancement processing is performed; When the time difference is greater than or equal to the preset time threshold, and / or when the clarity detection result is less than or equal to the preset image clarity threshold, the corresponding vehicle detection result is removed.

6. The method of claim 1, wherein, Further comprising: Obtain the adjustment content of the Q vehicle detection results, and obtain the first adjustment index; The continuity of the vehicle association result is evaluated, and the continuity evaluation result is obtained as the second adjustment index; The first adjustment index and the second adjustment index are combined to adjust the Q target cameras.

7. A multi-camera video joint analysis system for vehicle tracking, characterized by, Including: A video data set obtaining module is used to obtain the position information of the Q target cameras of the target road through the traffic management system, and real-time Q video data sets are obtained, 2≤Q; A vehicle detection result obtaining module is used to input the Q video data sets into the Q edge nodes of the Q target cameras respectively, perform vehicle detection, and obtain Q vehicle detection results; A standard vehicle detection result obtaining module is used to transmit the Q vehicle detection results to a cloud processor, obtain a preset synchronous detection standard, synchronously adjust the Q vehicle detection results based on the preset synchronous detection standard, and obtain P standard vehicle detection results, 2≤P≤Q; A vehicle association result obtaining module is used to perform vehicle association analysis on the P standard vehicle detection results based on the position information, and obtain a vehicle association result; A positioning result corresponding module is used to obtain a target vehicle to be tracked, match the target vehicle with a target association result, obtain a positioning result of the target vehicle for the Q target cameras, and the positioning result corresponds to the mth camera; A target vehicle continuous tracking module is used to generate a continuous tracking instruction, send the continuous tracking instruction to the Q-m cameras after the mth camera, and continuously track the target vehicle; The positioning result corresponding module is also used to set a test vehicle on the target road, and the test vehicle includes a positioning device; Start the test vehicle, and perform real-time positioning on the test vehicle through the positioning device to obtain a real-time positioning result; Data of the test vehicle are collected through the Q target cameras to obtain a data collection result; The data acquisition result is sent to the cloud processor for vehicle positioning, and a delay coefficient is obtained in combination with the real-time positioning result.

Images