Image multi-target recognition system based on composite feature dynamic processing

Abstract

The application discloses an image multi-target recognition system based on composite feature dynamic processing, relates to the technical field of image multi-target recognition, and specifically comprises a multi-target recognition platform. The multi-target recognition platform serves as a core control unit to perform signal interaction. A dynamic recognition early warning unit performs differential analysis on a track. A static recognition early warning unit focuses on a vehicle static stage to realize safety evaluation and risk early warning in the static stage. A microscopic behavior recognition detection unit realizes risk grading and real-time management and control of microscopic behaviors. The application realizes full-chain analysis from dynamic to static and from macroscopic track to microscopic behavior through linkage of three recognition early warning units and the multi-target recognition platform, significantly improves traffic area image analysis efficiency and management and control accuracy, and provides strong technical support for traffic relief and safety management and control.

Description

Technical Field

[0001] This invention relates to the field of image multi-target recognition technology, specifically to an image multi-target recognition system based on dynamic processing of composite features. Background Technology

[0002] In the current field of traffic management, image multi-target recognition technology is the core support for realizing intersection electronic police, speed measurement at checkpoints, violation capture, road traffic flow monitoring, license plate and vehicle attribute recognition, and behavior analysis. The main data sources rely on image acquisition hardware equipment such as high-definition network cameras (IPC), dome cameras (PTZ), and bullet cameras. However, with the acceleration of urbanization, road traffic flow has surged, and the scenario of cars and electric vehicles sharing the road is becoming increasingly common. Traffic management faces multiple challenges, including complex dynamic trajectories, hidden static risks, and diverse micro-behaviors.

[0003] Currently, most image multi-target recognition systems adopt a single feature recognition mode, which specifically includes the following:

[0004] 1. Dynamic recognition only focuses on vehicle speed or simple trajectory judgment, lacks differentiated analysis of the forward and reverse trajectories of the same and different types of vehicles, and cannot accurately distinguish the causes of congestion (such as the impact of reverse traffic and excessive merging volume), resulting in insufficient targeted early warning and low efficiency of control and guidance.

[0005] 2. Static identification often overlooks safety hazards when the vehicle is stationary or moving at low speeds. In particular, it lacks effective detection methods for safety issues related to spacing caused by changes in the outline of electric vehicles, such as the addition of rain canopies, which can easily lead to the omission of static risk points.

[0006] 3. Micro-behavior recognition does not specifically detect the high-risk period of the yellow light phase at intersections. It lacks judgment on the matching of distance and time when vehicles start and accelerate, making it difficult to avoid congestion and accident risks caused by improper micro-behavior in advance.

[0007] To address the aforementioned technical shortcomings, an image multi-target recognition system based on dynamic processing of composite features is proposed. This system aims to overcome the technical deficiencies in traditional image multi-target recognition for traffic control, and to achieve comprehensive and accurate recognition and control of dynamic trajectories, static states, and micro-behaviors in traffic areas. Summary of the Invention

[0008] The purpose of this invention is to solve the problems mentioned above by proposing an image multi-target recognition system based on dynamic processing of composite features.

[0009] The objective of this invention can be achieved through the following technical solution: an image multi-target recognition system based on dynamic processing of composite features, comprising a multi-target recognition platform, wherein the multi-target recognition platform is connected to the following communication links:

[0010] The dynamic identification and early warning unit performs dynamic processing and identification on traffic area images and provides dynamic early warning based on the identification results; the static identification and early warning unit performs static identification and early warning on traffic area images, and has already determined safe road sections.

[0011] After determining the safe road section, the micro-behavior recognition and detection unit performs multi-micro-behavior recognition and detection on the traffic area image.

[0012] Furthermore, the process of dynamically identifying early warning units is as follows:

[0013] RGB and depth images are captured by cameras, and vehicle speed and distance data are collected by radar. The image and radar data are fused to locate the road segment within the current image. Vehicle trajectories are extracted, categorized into car trajectories and electric vehicle trajectories. Road travel direction is determined by arrows indicating the road segment's location, and vehicle trajectories are further classified as either forward or reverse-flowing. Dynamic trajectory analysis is performed based on the traffic area image.

[0014] Obtain the floating span of the boundary distance between adjacent forward trajectories of vehicles of the same type, and at the same time obtain the minimum value of the boundary distance between adjacent forward trajectories of different types.

[0015] Furthermore, if the reciprocating span of the boundary distance between adjacent vehicles of the same type exceeds the reciprocating span threshold, or if the minimum value of the boundary distance between adjacent vehicles of different types does not exceed the minimum distance threshold, then it is inferred that the dynamic analysis of moving vehicles in the traffic area image is abnormal, and a normal congestion signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0016] If the boundary distance between adjacent vehicles of the same type traveling in the same direction does not exceed the reciprocating floating span threshold, and the minimum distance between adjacent vehicles of different types traveling in the same direction exceeds the minimum distance threshold, then it is inferred that the dynamic analysis of moving vehicles in the traffic area image is normal, and a normal traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0017] Furthermore, the speed reduction span when adjacent reverse-traffic trajectories of dissimilar vehicles meet and overlap is obtained, along with the lane area occupancy percentage of adjacent reverse-traffic trajectories of similar vehicles.

[0018] If the speed reduction span of adjacent reverse trajectories of non-same type vehicles exceeds the reduction span threshold when they meet and overlap, or if the lane area occupied by adjacent reverse trajectories of same type vehicles exceeds the area proportion threshold, it is inferred that the moving reverse trajectory in the traffic area image has a high impact, and a high impact traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0019] If the speed reduction span of adjacent reverse trajectories of non-same type vehicles does not exceed the reduction span threshold when they meet and overlap, and the lane area occupied by adjacent reverse trajectories of same type vehicles does not exceed the area proportion threshold, then it is inferred that the moving reverse trajectory analysis in the traffic area image has low impact, and a low-impact traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0020] Furthermore, after receiving the corresponding type of signal, the multi-target recognition platform conducts traffic control warnings and marks the corresponding location as a control location, and identifies vehicle information; at the same time, it performs joint statistics on various locations in the traffic area image. If a high-impact traffic location appears on the same road and the number of normal congestion locations continues to increase, it indicates that the corresponding inverse trajectory vehicle is a congestion impact point, and targeted control and assisted diversion are carried out; if only the number of normal congestion locations on the same road continues to increase, it indicates the traffic volume of the corresponding road merging point, and the merging point is regarded as a congestion impact point, and targeted diversion and assisted diversion are carried out.

[0021] Furthermore, the process of static identification and early warning unit is as follows:

[0022] Static phase identification is performed on the vehicle trajectory and electric vehicle trajectory. If the corresponding subjects within the vehicle trajectory and electric vehicle trajectory are at the same horizontal position and the speed deviation is within the set speed deviation range, then the current time period is marked as a static phase. When the distance between adjacent subjects shortens within the static phase, the movable safety distance of the adjacent subjects in the opposite direction is obtained. The opposite direction is represented by the relative direction of any adjacent subject to any trajectory in the adjacent trajectory. For example, if there is an electric vehicle trajectory to the right of the vehicle trajectory, then the opposite direction is to the left of the vehicle trajectory. This data is used to detect the movable safety distance of the lane. When the distance between adjacent subjects does not shorten within the static phase, the distance deviation value between the actual outline boundary of the electric vehicle and the rated type outline boundary of the electric vehicle is obtained.

[0023] Furthermore, if the movable safety distance between adjacent subjects in opposite directions exceeds the safety distance threshold, and the distance deviation between the actual outline boundary of the electric vehicle and the outline boundary of the rated type of the electric vehicle in the adjacent subjects does not exceed the distance deviation threshold, then it is inferred that the static stage analysis within the traffic area image is safe, a static safety signal is generated and sent to the multi-target recognition platform; after receiving the static safety signal, the multi-target recognition platform marks the corresponding road segment as a safe road segment.

[0024] If the movable safety distance between adjacent objects in opposite directions does not exceed the safety distance threshold, or if the distance deviation between the actual outline boundary of the electric vehicle and the rated type outline boundary of the electric vehicle exceeds the distance deviation threshold, then a static stage analysis hazard is inferred within the traffic area image, a static hazard signal is generated, and it is sent to the multi-target recognition platform.

[0025] Furthermore, the process of the multi-microscopic behavior recognition and detection unit is as follows:

[0026] Based on the safe road segments determined by the multi-target recognition platform, the safe road segments are located using traffic area images. Simultaneously, the intersections of the safe road segments are identified, and the intersection areas are marked as behavior recognition areas. Within the recognition area, the system extracts the moving subjects starting from the intersection during the yellow light phase and the accelerating moving subjects in the road where passage is about to end, obtaining the shortest distance between the moving subjects during their passage phase. It should be noted that the intersection where passage is about to end represents the red light turning yellow and about to turn green; the road where passage is about to end represents the corresponding green light turning yellow and about to turn red, or a zebra crossing. The system also obtains the deviation between the actual remaining passage time for the road where passage is about to end during the yellow light phase and the yellow light ending time within the recognition area.

[0027] Furthermore, if the shortest distance during the main traffic phase exceeds the shortest distance threshold, and the deviation between the actual remaining time required for traffic on the road to be closed and the yellow light end time exceeds the time deviation threshold, then it is inferred that the micro-behavior recognition detection of the corresponding behavior recognition area in the traffic area image is qualified, a low-risk behavior signal is generated and sent to the multi-target recognition platform, which then continues to monitor the behavior in the behavior recognition area. If the shortest distance during the main traffic phase does not exceed the shortest distance threshold, or the deviation between the actual remaining time required for traffic on the road to be closed and the yellow light end time does not exceed the time deviation threshold, then it is inferred that the micro-behavior recognition detection of the corresponding behavior recognition area in the traffic area image is unqualified, a high-risk behavior signal is generated and sent to the multi-target recognition platform.

[0028] Compared with the prior art, the beneficial effects of the present invention are:

[0029] 1. By fusing image and radar data, the system accurately locates road positions and extracts independent trajectories of cars and electric vehicles, clearly distinguishing between forward and reverse trajectories and significantly improving the accuracy of trajectory recognition. For forward and reverse scenarios involving similar and dissimilar vehicles, multi-dimensional judgment indicators are set, including boundary distance fluctuation range, minimum distance, speed reduction range, and lane occupancy area ratio. Combined with threshold comparisons, this achieves accurate classification of traffic anomalies, avoiding the limitations of single-indicator judgments and improving the accuracy of anomaly identification. By linking various signals with vehicle location information, the system assists the multi-target recognition platform in accurately locating congestion impact points, providing data support for targeted management and guidance, significantly improving the efficiency of traffic congestion relief, and reducing the blindness of traffic control.

[0030] 2. Clearly define the judgment criteria for the static phase to achieve accurate identification of the static phase, avoid misjudgments of low-speed driving and the static phase, and ensure the pertinence of static risk assessment; for the two scenarios of shortened and unshortened distance between adjacent objects in the static phase, introduce two core indicators: movable safety distance and electric vehicle outline deviation, to achieve multi-dimensional assessment of static safety risks, accurately judge lane safety redundancy and changes in the width occupied by electric vehicles, and avoid static risks such as scratches in advance; generate static safety and danger signals by comparing safety thresholds, and the multi-target recognition platform uses this to control dangerous road sections in a timely manner and to remove electric vehicles with illegally installed components, reducing the traffic risk in the static phase from the source. At the same time, mark safe road sections as micro-behavior recognition areas to achieve linkage between static recognition and micro-recognition and improve the consistency of the overall system management.

[0031] 3. Focusing on the high-risk period of yellow lights at intersections, this technology accurately extracts the behavioral characteristics of those starting to cross and those accelerating to finish, filling the gap in existing technology for analyzing micro-behavioral behavior during this critical period and achieving targeted and accurate micro-behavioral identification. By detecting two core indicators—the shortest distance and time deviation of the subject's passage—and combining them with threshold comparisons, it classifies the risks of micro-behavioral behavior, accurately assesses the safety of the subject's passage, and proactively avoids traffic chaos caused by micro-behavioral behaviors such as rushing and rear-end collisions. It accurately captures micro-behavioral risks, enabling real-time control of micro-traffic behavior and providing strong support for traffic management during critical periods at intersections.

[0032] In summary, this invention achieves dynamic processing of composite features for multi-target recognition in traffic area images through the collaborative work of three major identification and early warning units and a multi-target recognition platform. Compared with existing technologies, it has significant technical advantages and practical value. Attached Figure Description

[0033] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0034] Figure 1 This is a system principle block diagram of the present invention;

[0035] Figure 2 This is a flowchart of the method for dynamically identifying early warning units in this invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0038] Please see Figure 1 As shown, the image multi-target recognition system based on composite feature dynamic processing includes a multi-target recognition platform, which is communicatively connected to a dynamic recognition early warning unit, a static recognition early warning unit, and a microscopic behavior recognition and detection unit. In addition, the multi-target recognition platform is communicatively connected to an image acquisition hardware device.

[0039] Surveillance cameras: High-definition network cameras (IPC), dome cameras (PTZ), bullet cameras

[0040] Application scenarios: Traffic cameras at intersections, speed checks at checkpoints, traffic violation capture, road traffic flow monitoring; a primary data source for license plate recognition, vehicle attribute recognition, and behavior analysis.

[0041] The multi-target recognition platform collects images of traffic areas and performs dynamic feature processing and recognition based on the collected images to improve the efficiency of traffic area image analysis and facilitate traffic area management.

[0042] The multi-target recognition platform generates dynamic recognition early warning signals and sends them to the dynamic recognition early warning unit;

[0043] Please see Figure 2 As shown, after receiving the dynamic identification warning signal, the dynamic identification warning unit performs dynamic processing and identification on the traffic area image, and performs dynamic warning processing based on the identification results.

[0044] The system collects RGB and depth images using cameras and vehicle speed and distance data using radar. It then fuses the image and radar data to locate the road segment in the current image and extracts the vehicle's trajectory, which is divided into car trajectory and electric vehicle trajectory.

[0045] The direction of travel on the road is determined by arrows indicating the location of road segments, and vehicle trajectories are divided into forward and reverse trajectories; dynamic trajectory analysis is performed based on traffic area images.

[0046] Obtain the reciprocating floating span of the boundary distance between adjacent vehicles of the same type traveling in the same direction, and simultaneously obtain the minimum value of the boundary distance between adjacent vehicles of different types traveling in the same direction. Compare the reciprocating floating span of the boundary distance between adjacent vehicles of the same type traveling in the same direction and the minimum value of the boundary distance between adjacent vehicles of different types traveling in the same direction with the reciprocating floating span threshold and the minimum distance threshold, respectively.

[0047] If the boundary distance between adjacent vehicles of the same type in the same direction exceeds the threshold for the reciprocating floating span, or if the minimum value of the boundary distance between adjacent vehicles of different types in the same direction does not exceed the minimum distance threshold, then it is inferred that the dynamic analysis of moving vehicles in the traffic area image is abnormal, and a normal congestion signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0048] If the reciprocating floating span of the boundary distance between adjacent vehicles of the same type does not exceed the reciprocating floating span threshold, and the minimum value of the boundary distance between adjacent vehicles of different types exceeds the minimum distance threshold, then it is inferred that the dynamic analysis of moving vehicles in the traffic area image is normal, and a normal traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0049] Obtain the speed reduction span when adjacent reverse-traffic trajectories of dissimilar vehicles meet and overlap, and simultaneously obtain the lane area percentage corresponding to adjacent reverse-traffic trajectories of similar vehicles. Compare the speed reduction span when adjacent reverse-traffic trajectories of dissimilar vehicles meet and overlap, and the lane area percentage corresponding to adjacent reverse-traffic trajectories of similar vehicles, with the speed reduction span threshold and the area percentage threshold, respectively.

[0050] If the speed reduction span of adjacent reverse trajectories of non-same type vehicles exceeds the reduction span threshold when they meet and overlap, or if the lane area occupied by adjacent reverse trajectories of same type vehicles exceeds the area proportion threshold, it is inferred that the moving reverse trajectory in the traffic area image has a high impact, and a high impact traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0051] If the speed reduction span of adjacent reverse trajectories of non-same type vehicles does not exceed the reduction span threshold when they meet and overlap, and the lane area occupied by adjacent reverse trajectories of same type vehicles does not exceed the area proportion threshold, then it is inferred that the mobile reverse trajectory analysis in the traffic area image has low impact, and a low impact traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

[0052] It should be explained that adjacent forward trajectories mean that both are forward trajectories when compared, while adjacent reverse trajectories mean that one of the trajectories is a reverse trajectories when compared.

[0053] After receiving the corresponding type of signal, the multi-target recognition platform issues traffic control warnings and marks the corresponding location as a control location, and identifies vehicle information. At the same time, it performs joint statistics on various locations in the traffic area image. If a high-impact traffic location appears on the same road and the number of normal congestion locations continues to increase, it indicates that the corresponding inverse trajectory vehicle is a congestion point, and targeted control and assisted traffic management are implemented. If only the number of normal congestion locations on the same road continues to increase, it indicates that the merging point on the corresponding road is a congestion point, and targeted diversion and assisted traffic management are implemented.

[0054] The multi-target recognition platform generates a static recognition warning signal and sends it to the static recognition warning unit; after receiving the static recognition warning signal, the static recognition warning unit performs static recognition warning on the traffic area image;

[0055] Static phase identification is performed on car trajectories and electric vehicle trajectories. If the corresponding subjects within the car trajectory and electric vehicle trajectory are at the same horizontal position and the speed deviation is within the set speed deviation range, then the current time period is marked as a static phase. It should be explained that time periods with corresponding subjects at the same horizontal position in different trajectories or different road segments are all merged into the static phase. The subject is represented by the car or electric vehicle of the corresponding type of trajectory.

[0056] When the distance between adjacent entities shortens during the static phase, the movable safety distance of the adjacent entities in the opposite direction is obtained. The opposite direction is the relative direction of any trajectory in the adjacent trajectories to the adjacent entities. For example, if there is a battery vehicle trajectory to the right of the car trajectory, then the opposite direction is the left side of the car trajectory. This data is used to detect the movable safety distance of the lane.

[0057] When the distance between adjacent main bodies is not shortened during the static stage, the distance deviation between the actual outline boundary of the electric vehicle and the outline boundary of the rated type of the electric vehicle in the adjacent main bodies is obtained. It should be explained that the electric vehicle may have components such as rain shelters installed, which will cause changes in width.

[0058] The movable safety distance between adjacent entities in opposite directions and the distance deviation between the actual outline boundary of the electric vehicle and the rated type outline boundary of the electric vehicle in adjacent entities are compared with the safety distance threshold and the distance deviation threshold, respectively:

[0059] If the movable safety distance between adjacent objects in opposite directions exceeds the safety distance threshold, and the distance deviation between the actual outline boundary of the electric vehicle and the outline boundary of the rated type of the electric vehicle in the adjacent objects does not exceed the distance deviation threshold, then it is inferred that the static stage analysis within the traffic area image is safe, a static safety signal is generated and sent to the multi-target recognition platform; after receiving the static safety signal, the multi-target recognition platform marks the corresponding road segment as a safe road segment.

[0060] If the movable safety distance between adjacent entities in opposite directions does not exceed the safety distance threshold, or if the distance deviation between the actual outline boundary of the electric vehicle and the outline boundary of the rated type of the electric vehicle in the adjacent entities exceeds the distance deviation threshold, then a static phase analysis hazard is inferred in the traffic area image, a static hazard signal is generated and sent to the multi-target recognition platform. After receiving the static hazard signal, the multi-target recognition platform controls this road section and adds or removes electric vehicles of the same type.

[0061] After a safe road section is determined, a multi-micro behavior recognition and detection signal is generated and sent to the multi-micro behavior recognition and detection unit;

[0062] After receiving the multi-micro behavior recognition and detection signal, the multi-micro behavior recognition and detection unit performs multi-micro behavior recognition and detection on the traffic area image to infer whether there is congestion in the traffic area image.

[0063] Based on the safe road segments determined by the multi-target recognition platform, the safe road segments are located using traffic area images. At the same time, the intersections of the safe road segments are determined, and the intersection areas are marked as behavior recognition areas.

[0064] Extract the moving subject starting from the intersection during the yellow light phase of the indicator lights within the recognition area and the accelerating moving subject in the road where passage is about to end, and obtain the shortest distance of the subject during the passage phase; it should be explained that the intersection where passage is about to end represents the red light turning into the yellow light phase and about to turn green; the road where passage is about to end represents the road or zebra crossing that is about to turn into the yellow light phase and about to turn red.

[0065] Obtain the deviation between the actual remaining time required for traffic to end on the road within the recognition area during the yellow light phase and the yellow light end time;

[0066] The deviations between the shortest distance during the main traffic phase and the actual remaining time required for traffic to end on the road, and the yellow light end time, are compared with the shortest distance threshold and the time deviation threshold, respectively:

[0067] If the shortest distance during the main traffic phase exceeds the shortest distance threshold, and the deviation between the actual remaining time required for traffic on the road to be closed and the yellow light end time exceeds the time deviation threshold, then it is inferred that the micro-behavior recognition detection of the corresponding behavior recognition area in the traffic area image is qualified, a low-risk behavior signal is generated and sent to the multi-target recognition platform, and the multi-target recognition platform continues to monitor the behavior in the behavior recognition area after receiving it.

[0068] If the shortest distance during the main traffic phase does not exceed the shortest distance threshold, or if the deviation between the actual remaining time required for traffic to end and the yellow light end time does not exceed the time deviation threshold, it is inferred that the micro-behavior recognition detection in the corresponding behavior recognition area within the traffic area image is unqualified, generating a high-risk behavior signal and sending it to the multi-target recognition platform. It should be explained that high-risk behaviors pose a direct risk to road congestion and indirectly affect road traffic by increasing the risk of accidents. After receiving the high-risk behavior signal, the multi-target recognition platform controls the current behavior recognition area, capturing and warning vehicles that cross the line and providing real-time behavior warnings to pedestrians, i.e., through voice broadcasting.

[0069] This invention aims to solve the technical problems of low accuracy of multi-target image recognition, insufficient control targeting, and lack of full-chain analysis in traditional traffic management, and to achieve comprehensive and accurate recognition and control of dynamic trajectories, static states, and micro-behaviors in traffic areas.

[0070] Thresholds, preset values, preset ranges, etc. are set for result comparison and analysis to determine whether they are good or bad. The value of these thresholds is determined by a combination of large-scale model analysis of sample data and human experience. They can also be adjusted appropriately based on seasonal or common-sense influences.

[0071] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An image multi-target recognition system based on dynamic processing of composite features, characterized in that, This includes a multi-target recognition platform, whose communication connections include: The dynamic identification and early warning unit dynamically processes and identifies traffic area images, and performs dynamic early warning processing based on the identification results; The static identification and early warning unit performs static identification and early warning on traffic area images, and has completed the determination of safe road sections; After determining the safe road section, the micro-behavior recognition and detection unit; Perform multi-micro behavior recognition and detection on traffic area images.

2. The image multi-target recognition system based on composite feature dynamic processing according to claim 1, characterized in that, The process of dynamically identifying early warning units is as follows: RGB and depth images are captured by cameras, and vehicle speed and distance data are collected by radar. The image and radar data are fused to locate the road segment within the current image. Vehicle trajectories are extracted, categorized into car trajectories and electric vehicle trajectories. Road travel direction is determined by arrows indicating the road segment's location, and vehicle trajectories are further classified as either forward or reverse-flowing. Dynamic trajectory analysis is performed based on the traffic area image. Obtain the floating span of the boundary distance between adjacent forward trajectories of vehicles of the same type, and at the same time obtain the minimum value of the boundary distance between adjacent forward trajectories of different types.

3. The image multi-target recognition system based on composite feature dynamic processing according to claim 2, characterized in that, If the boundary distance between adjacent vehicles of the same type in the same direction exceeds the threshold for the reciprocating floating span, or if the minimum value of the boundary distance between adjacent vehicles of different types in the same direction does not exceed the minimum distance threshold, then it is inferred that the dynamic analysis of moving vehicles in the traffic area image is abnormal, and a normal congestion signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle. If the boundary distance between adjacent vehicles of the same type traveling in the same direction does not exceed the reciprocating floating span threshold, and the minimum distance between adjacent vehicles of different types traveling in the same direction exceeds the minimum distance threshold, then it is inferred that the dynamic analysis of moving vehicles in the traffic area image is normal, and a normal traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

4. The image multi-target recognition system based on composite feature dynamic processing according to claim 3, characterized in that, Obtain the speed reduction span when adjacent reverse-traffic trajectories of dissimilar vehicles meet and overlap, and simultaneously obtain the lane area percentage occupied by adjacent reverse-traffic trajectories of similar vehicles: If the speed reduction span of adjacent reverse trajectories of non-same type vehicles exceeds the reduction span threshold when they meet and overlap, or if the lane area occupied by adjacent reverse trajectories of same type vehicles exceeds the area proportion threshold, it is inferred that the moving reverse trajectory in the traffic area image has a high impact, and a high impact traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle. If the speed reduction span of adjacent reverse trajectories of non-same type vehicles does not exceed the reduction span threshold when they meet and overlap, and the lane area occupied by adjacent reverse trajectories of same type vehicles does not exceed the area proportion threshold, then it is inferred that the moving reverse trajectory analysis in the traffic area image has low impact, and a low-impact traffic signal is generated and sent to the multi-target recognition platform along with the location of the corresponding vehicle.

5. The image multi-target recognition system based on composite feature dynamic processing according to claim 4, characterized in that, After receiving the corresponding type of signal, the multi-target recognition platform issues traffic control warnings and marks the corresponding location as a control location, and identifies vehicle information. At the same time, it performs joint statistics on various locations in the traffic area image. If a high-impact traffic location appears on the same road and the number of normal congestion locations continues to increase, it indicates that the corresponding inverse trajectory vehicle is a congestion point, and targeted control and assisted diversion are implemented. If only the number of normal congestion locations on the same road continues to increase, it indicates that the merging point on the corresponding road is a congestion point, and targeted diversion and assisted diversion are implemented.

6. The image multi-target recognition system based on composite feature dynamic processing according to claim 1, characterized in that, The process of static identification and early warning unit is as follows: Static phase identification is performed on the vehicle trajectory and electric vehicle trajectory. If the corresponding subjects within the vehicle trajectory and electric vehicle trajectory are at the same horizontal position and the speed deviation is within the set speed deviation range, then the current time period is marked as a static phase. When the distance between adjacent subjects shortens within the static phase, the movable safety distance of the adjacent subjects in the opposite direction is obtained. The opposite direction is represented by the relative direction of any adjacent subject to any trajectory in the adjacent trajectory. For example, if there is an electric vehicle trajectory to the right of the vehicle trajectory, then the opposite direction is to the left of the vehicle trajectory. This data is used to detect the movable safety distance of the lane. When the distance between adjacent subjects does not shorten within the static phase, the distance deviation value between the actual outline boundary of the electric vehicle and the rated type outline boundary of the electric vehicle is obtained.

7. The image multi-target recognition system based on composite feature dynamic processing according to claim 6, characterized in that, If the movable safety distance between adjacent subjects in opposite directions exceeds the safety distance threshold, and the distance deviation between the actual outline boundary of the electric vehicle and the outline boundary of the rated type of the electric vehicle in the adjacent subjects does not exceed the distance deviation threshold, then it is inferred that the static stage analysis of the traffic area image is safe, a static safety signal is generated and sent to the multi-target recognition platform. After receiving a static safety signal, the multi-target recognition platform marks the corresponding road segment as a safe road segment; If the movable safety distance between adjacent objects in opposite directions does not exceed the safety distance threshold, or if the distance deviation between the actual outline boundary of the electric vehicle and the rated type outline boundary of the electric vehicle exceeds the distance deviation threshold, then a static stage analysis hazard is inferred within the traffic area image, a static hazard signal is generated, and it is sent to the multi-target recognition platform.

8. The image multi-target recognition system based on composite feature dynamic processing according to claim 1, characterized in that, The process of the multi-micro behavior recognition and detection unit is as follows: Based on the safe road segments determined by the multi-target recognition platform, the safe road segments are located using traffic area images. Simultaneously, the intersections of the safe road segments are identified, and the intersection areas are marked as behavior recognition areas. Within the recognition area, the system extracts the moving subjects starting from the intersection during the yellow light phase and the accelerating moving subjects in the road where passage is about to end, obtaining the shortest distance between the moving subjects during their passage phase. It should be noted that the intersection where passage is about to end represents the red light turning yellow and about to turn green; the road where passage is about to end represents the corresponding green light turning yellow and about to turn red, or a zebra crossing. The system also obtains the deviation between the actual remaining passage time for the road where passage is about to end during the yellow light phase and the yellow light ending time within the recognition area.

9. The image multi-target recognition system based on composite feature dynamic processing according to claim 8, characterized in that, If the shortest distance during the main traffic phase exceeds the shortest distance threshold, and the deviation between the actual remaining time required for traffic to end on the road and the yellow light end time exceeds the time deviation threshold, then it is inferred that the micro-behavior recognition detection of the corresponding behavior recognition area in the traffic area image is qualified, a low-risk behavior signal is generated and sent to the multi-target recognition platform, which then continues to monitor the behavior in the behavior recognition area. If the shortest distance during the main traffic phase does not exceed the shortest distance threshold, or the deviation between the actual remaining time required for traffic to end on the road and the yellow light end time does not exceed the time deviation threshold, then it is inferred that the micro-behavior recognition detection of the corresponding behavior recognition area in the traffic area image is unqualified, a high-risk behavior signal is generated and sent to the multi-target recognition platform.

Images