Traffic violation behavior detection system based on unmanned aerial vehicle

Abstract

The invention relates to a traffic violation behavior detection system based on unmanned aerial vehicles. The traffic violation behavior detection system comprises a flight control platform and a plurality of unmanned aerial vehicles in communication connection with the flight control platform, the flight control platform is used for issuing flight instructions to the plurality of unmanned aerial vehicles and receiving image data returned by each unmanned aerial vehicle; performing data processing and trajectory tracking on the image data, extracting key features in a geographic coordinate trajectory of each vehicle, and comparing the key features with violation judgment rules corresponding to the current waypoint to obtain violation results; if the violation result is that violation is confirmed, a surrounding evidence obtaining instruction is sent out; the plurality of unmanned aerial vehicles are used for receiving and executing the flight instruction issued by the flight control platform, autonomously flying according to a preset cruise route, acquiring image data of a waypoint in charge in real time and returning the image data to the flight control platform; and after the surrounding evidence obtaining instruction is received, the surrounding evidence obtaining module is also used for starting the surrounding evidence obtaining mode to fly to the illegal vehicle and obtaining the information of the illegal vehicle. According to the invention, efficient and instant response and disposal of illegal behaviors are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traffic technology, and in particular to a traffic violation behavior detection system based on a UAV. BACKGROUND

[0002] With the development of intelligent traffic management systems, UAVs are gradually applied to the field of traffic patrol and supervision due to their advantages of being flexible and having a wide field of view. At present, traffic monitoring based on UAVs mainly relies on two implementation schemes.

[0003] The first scheme adopts a collaborative operation mode of "remote manual operation + on-site pilot assistance". In this mode, the operator needs to remotely control the UAV to fly and take pictures, and manually monitor and judge the video pictures transmitted back. Since the entire patrol and identification process of this scheme highly depends on manual operation, it has the problems of low patrol efficiency and high labor cost, and it is difficult to realize large-scale and long-term continuous automatic supervision.

[0004] The second scheme uses a UAV to collect historical image data of the cruising area, and then transmits the image to a server for offline analysis and identification. Although this scheme reduces the dependence on real-time manual operation, it cannot discover and intervene in traffic violations in a timely manner because it processes non-real-time historical data, and has obvious deficiencies in the timeliness of event response.

[0005] In summary, the existing technology has limitations in automation and efficiency, or lacks real-time event detection, which restricts the further application of UAVs in automatic identification and efficient disposal of traffic violations.

[0006] Therefore, there is an urgent need for a traffic violation behavior detection system based on a UAV. SUMMARY

[0007] (I) Technical problems to be solved

[0008] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present application provides a traffic violation behavior detection system based on a UAV, which solves the technical problems of low efficiency and high cost caused by dependence on manual operation, and the inability to achieve immediate response and disposal due to non-real-time processing.

[0009] (II) Technical solutions

[0010] In order to achieve the above-mentioned purposes, the main technical solutions adopted by the present application include:

[0011] The present application provides a traffic violation behavior detection system based on a UAV, comprising:

[0012] A flight control platform and multiple drones that are communicatively connected to the flight control platform;

[0013] The flight control platform is used to issue flight commands to the multiple UAVs respectively, and receive image data of each waypoint transmitted back by each UAV; it is also used to process the image data and track the trajectory, generate the geographic coordinate trajectory of each vehicle appearing in the image data; and extract key features from the geographic coordinate trajectory of each vehicle, compare the key features with the violation judgment rules corresponding to the current waypoint, and obtain the violation result based on the logical threshold model; if the violation result is a confirmed violation, then based on the preset scheduling strategy, a surround evidence collection command is issued to at least one target UAV among the multiple UAVs.

[0014] The multiple drones are used to receive and execute flight commands issued by the flight control platform, fly autonomously along a pre-set cruise route, collect image data of the waypoints they are responsible for in real time and transmit it back to the flight control platform; they are also used to activate the surround evidence collection mode after receiving the surround evidence collection command, fly towards the violating vehicle and obtain information about the violating vehicle.

[0015] Optionally, in some embodiments of this application, the flight control platform is further configured to perform route planning before issuing flight commands, specifically including:

[0016] Obtain geographic information and historical traffic violation data for the target area;

[0017] Based on the geographic information and historical traffic violation data, multiple waypoints and the traffic monitoring scenario type corresponding to each waypoint are determined on the geographic map of the target area, and associated violation judgment rules and trigger parameters are configured for each waypoint; based on the status and location of the multiple drones and the location and priority of the multiple waypoints, a cluster task allocation algorithm is adopted to generate a non-conflicting cruise route for each drone that covers all or part of the specified monitoring waypoints.

[0018] Optionally, in some embodiments of this application, when the flight control platform determines multiple monitoring waypoints and the traffic monitoring scenario type corresponding to each monitoring waypoint on the geographic map of the target area, it is specifically used for:

[0019] The flight control platform analyzes the historical traffic violation data to identify areas with high violation rates, road sections with frequent solid line lane changes, or historical locations of traffic accidents, and determines the identified locations as monitoring waypoints; and obtains the traffic supervision scenario type for each monitoring waypoint based on the geographical information.

[0020] The traffic monitoring scenarios include traffic hub intersections, urban commercial areas, areas with frequent illegal parking, or highway emergency lanes.

[0021] Optionally, in some embodiments of this application, the flight control platform configures associated violation judgment rules and triggering parameters for each monitored waypoint, specifically for:

[0022] When the traffic monitoring scenario of the monitored route point is an urban commercial area or a hotspot for illegal parking, a rule for judging illegal parking is configured, and the triggering parameter is that the duration of the vehicle being stationary exceeds a preset threshold.

[0023] When the traffic monitoring scenario of the monitored waypoint is a traffic hub intersection, a solid line lane change judgment rule is configured, and the trigger parameter is the vehicle trajectory crossing the electronic geofence associated with the solid line lane.

[0024] When the traffic monitoring scenario of the monitored waypoint is the emergency lane of a highway, an illegal occupation of the emergency lane judgment rule is configured, and the trigger parameter is that the vehicle trajectory enters the electronic geofence area associated with the rightmost lane of the road and the width is not greater than the preset width.

[0025] Optionally, in some embodiments of this application, the flight control platform performs data processing and trajectory tracking on the image data, specifically for:

[0026] Perform perspective transformation correction on the received image data from the perspective of the UAV to obtain the corrected image data;

[0027] Based on the corrected image data, vehicles are located in each frame using a deep learning object detection model, and the vehicle's bounding box and category are output. Based on the vehicle's bounding box and category in the current frame and the previous frame, a unique trajectory identifier is assigned to and maintained for the same vehicle using a data association algorithm, generating continuous pixel-level trajectories for each vehicle.

[0028] The pixel coordinates in the pixel-level motion trajectory are combined with the real-time position and attitude data of the UAV that acquired the image data and mapped to coordinates in a unified geographic coordinate system to generate the geographic coordinate trajectory of the vehicle.

[0029] Optionally, in some embodiments of this application, the flight control platform assigns and maintains a unique trajectory identifier for the same vehicle through a data association algorithm, generating continuous pixel-level trajectories for each vehicle, specifically used for:

[0030] Based on the state prediction algorithm, the predicted position of the vehicle's bounding box in the next frame is predicted according to the vehicle's trajectory state in the current frame. Then, using the optimal matching algorithm, the vehicle bounding box actually obtained by the deep learning object detection model in the next frame is optimally matched with the predicted position of the bounding box to obtain the matching result.

[0031] Based on the matching result, the trajectory identifier of the current frame is passed to the vehicle bounding box of the successfully matched vehicle in the next frame, until the vehicle leaves the monitoring area of ​​the current waypoint.

[0032] Optionally, in some embodiments of this application, the flight control platform extracts key features from the geographic coordinate trajectory of each vehicle, compares the key features with the violation judgment rules corresponding to the current waypoint, and obtains the violation result based on a logical threshold model, specifically for:

[0033] Based on the geographic coordinate trajectory of the vehicle, key trajectory features are obtained. These key trajectory features include the duration and location of the trajectory stationary position used for illegal parking judgment, the crossing relationship between the trajectory and the preset solid line electronic fence used for solid line lane change judgment, or the intrusion relationship between the trajectory and the preset lane electronic fence used for illegal lane occupation judgment.

[0034] The key features of the trajectory are compared with the violation judgment rules of the current waypoint to obtain the comparison results, and the comparison results are input into the logic threshold model to output the violation results.

[0035] Optionally, in some embodiments of this application, the flight control platform is further used for:

[0036] After the drone completes the initial evidence collection, the violation is recorded, and after a preset delay time, the status of each drone is reassessed, and one drone is assigned to fly back to the violation location for a second round of evidence collection.

[0037] The results of the secondary evidence collection are used to verify whether the violation is ongoing. If it is ongoing, an alarm is triggered; if it has been eliminated, the violation status is marked as ended and the relevant violation information is archived.

[0038] Optionally, in some embodiments of this application, the UAV receives and executes flight commands issued by the flight control platform, autonomously flies along a pre-set cruise route, and collects real-time image data of the waypoints it is responsible for and transmits it back to the flight control platform, specifically for:

[0039] After arriving at the designated waypoint, it enters hover detection mode;

[0040] In the hover detection mode, the UAV hovers at a fixed altitude and adjusts the pitch and yaw angles of the camera device through the onboard gimbal to continuously capture images of the waypoint area from the optimal perspective.

[0041] After completing image acquisition at the current waypoint, the UAV autonomously flies to the next waypoint on the cruise route.

[0042] Optionally, in some embodiments of this application, the drone initiates a surround evidence-gathering mode to fly towards the offending vehicle and obtain information about the offending vehicle, specifically for:

[0043] Receive the geographical coordinates of the illegal vehicle from the flight control platform; based on the geographical coordinates and the drone's real-time positioning, plan and fly to the initial evidence collection position above the illegal vehicle;

[0044] Centered on the offending vehicle, the system maintains a constant radius on the horizontal plane and performs closed-loop orbiting flight at a constant speed and altitude. During the orbiting flight, the gimbal is synchronously controlled to keep the camera continuously pointed at the vehicle and continuously take pictures or record videos to obtain clear images from multiple angles, including the vehicle's license plate number, model, and appearance features.

[0045] (III) Beneficial Effects

[0046] The beneficial effects of this invention are as follows: The traffic violation detection system based on unmanned aerial vehicles (UAVs) of this invention adopts a system architecture that includes a flight control platform and multiple UAVs. The flight control platform performs unified task scheduling, data fusion, and violation judgment, while multiple UAVs perform distributed patrols and collaborative evidence collection. Compared with the prior art, it can achieve synchronous, flexible, and intelligent monitoring of a large-scale road network, significantly improving the coverage and efficiency of supervision, and achieving the technical effects of automated detection, intelligent judgment, rapid response, and accurate evidence collection of traffic violations. Attached Figure Description

[0047] Figure 1 This is a structural diagram of a drone-based traffic violation detection system according to an embodiment of this application;

[0048] Figure 2 This is a diagram showing the overall data flow of the flight control platform in a drone-based traffic violation detection system according to an embodiment of this application. Detailed Implementation

[0049] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Flight control platform: The flight control platform enables real-time control, autonomous decision-making, and mission management of the UAV. Its core functions include flight attitude stabilization, flight path planning, sensor data processing, equipment scheduling, and data transmission, and it is widely used in aerial photography, inspection, surveying, security, and other fields.

[0051] Deep learning object detection model: It uses deep learning algorithms to locate and identify target objects in images or videos, and outputs the object's category, bounding box, and confidence score.

[0052] Flight routes and waypoints: Flight route and waypoint planning for drones is a core element in achieving autonomous flight missions. By pre-setting paths and action commands, it can efficiently complete the needs of scenarios such as aerial photography, surveying, and inspection.

[0053] Pixel coordinates: A two-dimensional coordinate system used to locate the position of each pixel in a digital image. Its core elements include the origin, coordinate axis direction, coordinate representation method, and physical meaning.

[0054] Drone traffic patrol is an intelligent traffic management method that uses drones equipped with high-definition cameras, infrared devices, and loudspeaker modules to conduct aerial patrols, violation monitoring, and accident handling in scenarios such as roads, highways, and waterways. Its core value lies in breaking through the limitations of traditional ground patrols and improving processing efficiency and emergency response speed through an "air-ground linkage" model, making it an important component of the digital transportation system.

[0055] However, most drone patrol scenarios currently rely on remote human control, requiring operators to possess combined skills in flight control, equipment maintenance, and data interpretation. Currently, there is a shortage of certified drone pilots in the industry, particularly a lack of cross-disciplinary talent familiar with traffic regulations and AI recognition technology. This talent shortage leads to low drone patrol efficiency in some areas, and the cost of traffic patrols cannot be reduced.

[0056] To address the above issues, the industry currently widely adopts the following two technologies:

[0057] First, the construction of related supporting facilities, such as drone airports and flight service platform command centers, is accelerating, forming a collaborative operation mode covering "remote control + on-site pilots"; second, traffic image information within the patrol range is collected by drones and output to the server for image processing, and traffic abnormality events are identified through a traffic intelligent recognition model.

[0058] However, the current collaborative operation mode of "remote control + on-site pilot" relies on manual remote control, which is inefficient and costly. While the mode of using historical drone cruise images + view recognition can reduce the dependence on manual control to some extent, it generally suffers from problems such as the inability to recognize data in real time and low efficiency.

[0059] To address these issues, this invention proposes a traffic violation detection system based on unmanned aerial vehicles (UAVs), aiming to solve the core problems of existing traffic monitoring technologies, such as fixed monitoring range, numerous blind spots, low evidence collection efficiency, and difficulty in achieving real-time, intelligent supervision of large-scale road networks. This invention employs a system architecture where a flight control platform and multiple UAVs work collaboratively. The flight control platform is responsible for intelligent task scheduling and flight path planning for multiple UAVs, and performs real-time processing of their transmitted video streams, vehicle trajectory tracking, and automatic violation judgment based on electronic fences and logical rules. Once a violation is confirmed, the flight control platform dynamically schedules the optimal UAV to fly to the scene for automatic surround evidence collection. This technical solution achieves full-process automation from wide-area patrol monitoring and intelligent analysis to precise evidence collection, thereby significantly expanding the effective monitoring range, improving the efficiency of violation detection and handling, and enhancing the accuracy and timeliness of evidence collection.

[0060] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0061] Figure 1 This is a structural diagram of a drone-based traffic violation detection system according to an embodiment of this application. Figure 1 As shown, the UAV-based traffic violation detection system of this application includes:

[0062] Flight control platform and multiple drones that communicate with the flight control platform.

[0063] The flight control platform is essentially a high-performance computing and command center. Its core hardware includes: a communication server equipped with high-power radio frequency equipment and multiple network interfaces to establish stable, low-latency control and data links with multiple drones, enabling real-time command issuance and concurrent reception of multiple high-definition video streams; a data processing server cluster composed of servers equipped with high-performance GPUs for vehicle detection and identification, real-time image correction, and vehicle trajectory tracking calculations; a data storage and management server for storing high-precision maps, electronic geofence data, preset violation judgment rules, and archiving all original videos, vehicle trajectories, and evidence of violations; and a task scheduling and monitoring workstation providing operators with a graphical interface for monitoring drone status, reviewing violations, and intervening in tasks.

[0064] As the aerial mobile sensing unit of the system, the UAV mainly includes: flight and navigation hardware, the core of which is a flight controller integrating a high-precision GPS or Beidou module and an inertial measurement unit, ensuring that the UAV can fly and hover accurately and autonomously along a preset route; a mission payload module, equipped with a high-precision three-axis mechanical stabilization gimbal and high-performance optoelectronic camera equipment. The gimbal can effectively eliminate flight jitter, and the camera equipment (such as a 4K camera) usually has optical zoom capability, which can capture wide-angle images during cruise and zoom to capture details such as vehicle license plates during evidence collection. It is also equipped with a thermal imager and millimeter-wave radar: the 4K high-definition camera collects real-time video streams, the thermal imager assists in target identification in night / low visibility scenes, and the millimeter-wave radar supplements distance and speed data; an airborne communication module, including a dedicated data transmission radio for receiving key flight commands and a 4G / 5G or broadband image transmission module for transmitting high-definition video streams; and a power and energy module, which uses high-energy-density batteries and high-efficiency motors to meet the endurance requirements of long-term cruise and sudden mobile evidence collection missions.

[0065] In addition, the video streams captured by the drone are encrypted and transmitted to the flight control platform via a 5G / 4G image transmission link, keeping the latency within 200ms and ensuring the real-time performance of the algorithm.

[0066] In practice, the flight control platform centrally processes information, enabling more intelligent global decisions and in-depth analysis; while the drone, acting as a mobile "sentinel," can achieve rapid arrival and multi-angle evidence collection. The collaboration between the two ultimately achieves a highly efficient closed loop of "full-area monitoring, intelligent detection, and precise strikes," realizing automated and intelligent supervision of traffic violations over a vast area with limited hardware resources.

[0067] In this embodiment, see Figure 2 The flight control platform is used to issue flight commands to multiple drones and receive image data of each waypoint transmitted back by each drone. It is also used to process the image data and track the trajectory, generating the geographic coordinate trajectory of each vehicle appearing in the image data; extracting key features from the geographic coordinate trajectory of each vehicle, comparing the key features with the violation judgment rules corresponding to the current waypoint, and obtaining the violation result based on the logical threshold model; if the violation result is confirmed, it issues a surround evidence collection command to at least one target drone among the multiple drones based on the preset scheduling strategy.

[0068] In practical implementation, the flight control platform is also used to perform route planning before issuing flight commands, specifically including:

[0069] Obtain geographic information and historical traffic violation data for the target area;

[0070] Based on geographic information and historical traffic violation data, multiple waypoints and the traffic monitoring scenario type corresponding to each waypoint are determined on the geographic map of the target area, and associated violation judgment rules and trigger parameters are configured for each waypoint. Based on the status and location of multiple drones and the location and priority of multiple waypoints, a cluster task allocation algorithm is adopted to generate a cruise route for each drone that is non-conflicting and covers all or part of the specified monitoring waypoints.

[0071] Furthermore, when the flight control platform determines multiple monitoring waypoints and the corresponding traffic monitoring scenario type for each monitoring waypoint on the geographic map of the target area, it is specifically used for:

[0072] The flight control platform analyzes historical traffic violation data to identify areas with high violation rates, road sections with frequent solid line lane changes, or historical locations of traffic accidents, and determines the identified locations as monitoring waypoints; and obtains the traffic supervision scenario type for each monitoring waypoint based on geographic information;

[0073] Traffic monitoring scenarios include traffic hub intersections, urban commercial areas, hotspots for illegal parking, or highway emergency lanes.

[0074] In the embodiments of this application, the flight control platform configures associated violation judgment rules and trigger parameters for each monitored waypoint, specifically for:

[0075] When the traffic monitoring scenario of the monitored route point is an urban commercial area or a hotspot for illegal parking, configure the vehicle illegal parking judgment rule, and the trigger parameter is that the duration of the vehicle being stationary exceeds a preset threshold.

[0076] When the traffic monitoring scenario of the monitored route point is a traffic hub intersection, configure a solid line lane change judgment rule, and the trigger parameter is the vehicle trajectory crossing the electronic geofence associated with the solid line lane.

[0077] When the traffic monitoring scenario of the monitored waypoint is the emergency lane of the highway, the rule for judging illegal occupation of the emergency lane is configured, and the trigger parameter is that the vehicle trajectory enters the electronic geofence area associated with the rightmost lane of the road and the width is not greater than the preset width.

[0078] In the specific implementation process, the flight control platform's route planning function first relies on historical traffic violation data and high-precision geographic information of the target area. For example, in the actual deployment in a new district of a certain city, the platform analyzed historical data and identified a large number of illegal parking records on the roads surrounding the Southeast Trade City during weekday lunchtimes, while the intersection of the two ramps of the ring expressway was a high-incidence point for solid line lane changes. Based on this, the system automatically set the section of road above the east gate of the Trade City and the ring expressway ramp on the geographic information system map as two core monitoring flight points, and marked them as "illegal parking hotspot area" and "traffic hub intersection" respectively, according to their geographical location and violation characteristics. Subsequently, the flight control platform dynamically configures differentiated violation judgment rules and parameters for each flight point according to the scenario type: for the Trade City flight point, the system loads the "vehicle illegal parking judgment rule" and sets the trigger parameter to "the vehicle is stationary within the electronic no-parking fence for more than 120 seconds"; for the ring expressway ramp flight point, it loads the "solid line lane change judgment rule" and sets the trigger parameter to "the vehicle trajectory projection intrudes into the solid line electronic fence". Finally, the platform, considering the current locations of the three available drones (at the base and the western patrol point, respectively), their battery levels (85%, 70%, and 95%), and the task priorities of each waypoint (midday commercial area illegal parking monitoring has the highest priority), uses a cluster task allocation algorithm to optimize calculations and generate three conflict-free patrol routes: a drone with sufficient battery power is assigned to fly directly from the base to the commercial city waypoint for continuous monitoring; a drone from the western area is dispatched to the nearest waypoint at the ring road ramp; and the third drone serves as a mobile force, going to provide support after completing the current low-priority route. Through this process, the system achieves a complete intelligent route deployment, from data-driven intelligent waypoint deployment and automatic matching of scenario-based rules to dynamic planning of multi-drone tasks.

[0079] The cluster task allocation algorithm is configured to execute a real-time scheduling method based on multi-dimensional constraints. The core of this algorithm lies in the flight control platform defining each waypoint to be monitored as a monitoring task with a geographical location, monitoring scenario type, and priority, and continuously maintaining a status list containing the real-time location, remaining flight time, and current task load of each UAV. During initial task allocation or dynamic adjustment, the algorithm calculates based on a comprehensive cost function. This function primarily considers the estimated flight time of the UAV to the task point, the matching degree between the energy consumption required to execute the task and the UAV's remaining battery power, and the length of the UAV's existing task queue, in order to seek the allocation scheme with the shortest total response time and the most balanced task load globally.

[0080] Therefore, by comprehensively considering multi-dimensional constraints such as UAV status, mission location, and priority, and focusing on optimizing the comprehensive cost function, this algorithm can intelligently and dynamically generate efficient and conflict-free cruise routes for multiple UAVs. This achieves optimal matching between UAV waypoints, maximizes the synchronous monitoring coverage of a large-scale road network, and significantly improves the intelligence level of UAV swarm scheduling and overall service efficiency.

[0081] Furthermore, the flight control platform processes the image data and performs trajectory tracking, specifically for:

[0082] Perform perspective transformation correction on the received image data from the perspective of the UAV to obtain the corrected image data;

[0083] Based on the corrected image data, vehicles are located in each frame using a deep learning object detection model, and the vehicle's bounding box and category are output. Based on the vehicle's bounding box and category in the current frame and the previous frame, a unique trajectory identifier is assigned to and maintained for the same vehicle using a data association algorithm, generating continuous pixel-level trajectories for each vehicle.

[0084] The pixel coordinates in the pixel-level motion trajectory are combined with the real-time position and attitude data of the drone that acquired the image data and mapped to coordinates in a unified geographic coordinate system to generate the vehicle's geographic coordinate trajectory.

[0085] In practical implementation, for example, a drone hovers approximately 80 meters above an intersection, capturing traffic flow at a certain angle. First, the flight control platform performs perspective transformation correction on the received video frames. In the original footage, due to the drone's tilted perspective, the lane lines at the intersection appear to converge with a "nearer is larger, farther is smaller" shape; a car actually traveling in a straight line appears to have a curved trajectory in the image. After perspective transformation correction, the image is converted into a standard orthographic view, i.e., a vertical overhead view. The lane lines in the image become parallel and equal-width straight lines, allowing the position and movement of vehicles to be accurately represented in a unified two-dimensional coordinate system.

[0086] Furthermore, the deep learning object detection model is a specially optimized, high-efficiency convolutional neural network, whose core architecture is typically based on an improved single-stage detector. The model's input is a normalized video frame corrected for perspective transformation, i.e., corrected image data. It first extracts multi-level image features through a lightweight backbone network. To accurately locate various targets, from large vehicles in the foreground to small motorcycles in the background, the model employs a multi-scale prediction mechanism. This involves setting detection heads on deep and shallow feature maps, outputting bounding boxes and class probabilities for targets of different sizes in parallel. Addressing the unique challenges of small targets, occlusion, and multiple viewpoints inherent in UAV traffic monitoring, the model integrates a feature pyramid network to enhance small target recognition and introduces an attention mechanism to improve the detection rate of partially occluded vehicles in dense traffic flow. Through training and fine-tuning on a large amount of real traffic scene data covering different weather conditions, lighting, and angles, the model ultimately achieves high-precision, real-time vehicle detection and classification for each frame, stably outputting structured results containing pixel-level coordinates and vehicle type information, providing a reliable perceptual foundation for subsequent continuous trajectory tracking and behavior analysis.

[0087] In this embodiment, the deep learning object detection model, through its lightweight design and multi-scale detection mechanism, significantly improves the accuracy of identifying small vehicles, dense traffic flow, and partially occluded vehicles from an overhead view taken by a drone, while ensuring high processing speed to meet real-time analysis requirements. Furthermore, the model, trained on multi-scenario data, exhibits good environmental adaptability and can operate stably under different lighting and weather conditions. This provides a reliable and continuous perception foundation for subsequent vehicle trajectory tracking and behavior violation judgment, ensuring the realization of the core functions of the entire system.

[0088] Specifically, the flight control platform assigns and maintains a unique trajectory identifier for the same vehicle through a data association algorithm, generating continuous pixel-level trajectories for each vehicle, specifically for:

[0089] Based on the state prediction algorithm, the bounding box prediction position of the vehicle in the next frame is predicted according to the trajectory state of the vehicle in the current frame; using the optimal matching algorithm, the vehicle bounding box actually obtained by the deep learning object detection model in the next frame is optimally matched with the bounding box prediction position to obtain the matching result.

[0090] Based on the matching results, the trajectory identifier of the current frame is passed to the bounding box of the vehicle that is successfully matched in the next frame, until the vehicle leaves the monitoring area of ​​the current waypoint.

[0091] For example, the state prediction algorithm is the Kalman filter algorithm. The core objective of the Kalman filter algorithm is to maintain the continuity of a vehicle's identity when it is briefly occluded or when fluctuations are detected. This algorithm is based on the fundamental assumption that vehicles maintain approximately uniform motion over extremely short time intervals. For each tracked vehicle trajectory, the algorithm uses its state in the current frame and, through a built-in dynamic model (uniform motion model), calculates the region where it is most likely to appear in the next frame of the video, generating a corresponding "predicted bounding box." This predicted box is not a simple positional translation, but rather includes the algorithm's estimation of the vehicle's motion trend and a quantification of the reliability of the prediction. This prediction result provides crucial prior reference for subsequent data association steps, enabling the system to quickly and accurately identify which detected target belongs to the same vehicle as the historical trajectory among numerous actual detection boxes in the next frame, thus effectively solving the identity matching problem in multi-target intertwined environments.

[0092] Furthermore, after completing the Kalman filter algorithm prediction, two sets of key data are obtained: one set consists of M predicted bounding boxes from historical trajectories, each box associated with a unique trajectory ID and its predicted location; the other set consists of N newly detected actual vehicle bounding boxes in the current frame, output by the deep learning model, with location and confidence scores but no identification yet. These two sets of data constitute a typical bipartite graph matching problem.

[0093] To perform the matching, the algorithm first constructs an M×N cost matrix. Each element in the matrix represents the "cost" or "distance" of matching the i-th predicted box with the j-th detected box. This cost is typically determined by calculating the complement of the intersection-union ratio (IoU) between the predicted and detected boxes. IoU measures the degree of overlap between two rectangular regions; a higher IoU means a greater likelihood that the two boxes represent the same vehicle, and therefore a lower matching cost. If two boxes do not overlap spatially, their IoU is 0, and the cost is 1. Next, the optimal matching algorithm, the Hungarian algorithm, is applied to this cost matrix. The goal of this algorithm is to find an optimal matching combination such that the sum of the total matching costs of all successfully paired predicted-detected box pairs is globally minimized. This is equivalent to finding the most likely corresponding observed target in the current frame for each historical trajectory while adhering to the "one-to-one correspondence" principle. The algorithm efficiently solves for this optimal allocation scheme through a series of row reduction, column reduction, and covering zero elements.

[0094] After matching, the results are divided into three categories:

[0095] Successful match: The predicted bounding box and the detection bounding box are successfully paired. The system uniquely transmits the trajectory ID carried by the predicted bounding box to the corresponding detection bounding box, completing the trajectory continuation, and updates the Kalman filter with the actual observation position of this detection bounding box.

[0096] Unmatched detection boxes: New detection boxes that could not be associated with any predicted boxes. This usually means that a new vehicle has entered the monitoring field of view. The system will initialize new trajectories for these detection boxes and assign them entirely new trajectory IDs.

[0097] Unmatched predicted bounding box: A historical trajectory predicted bounding box that failed to be associated with any detection box. This may indicate that the vehicle has left the monitored area or is completely obscured. The system records the number of failed matches for this predicted bounding box. When consecutive failures exceed a certain threshold, the trajectory is considered terminated, and its ID is released.

[0098] By employing the aforementioned data association mechanism based on the collaborative operation of state prediction and optimal matching algorithms, a precise, continuous, and unique motion trajectory is constructed and maintained for each vehicle. The predictive power of Kalman filtering allows the system to reliably estimate the position of a vehicle even when it is briefly obscured or when frames are missed, thus ensuring trajectory continuity. The optimal global matching of the Hungarian algorithm ensures that, in the case of multiple vehicles intertwined and operating in parallel, the system can correctly associate each historical trajectory with its current real-time observation, achieving high-precision identity transfer and effectively avoiding ID jumps or mistracking. The combination of these two methods enables the system to robustly handle the appearance of new vehicles and the disappearance of old vehicles, achieving stable and real-time tracking of the motion state of all vehicles within the monitored area. This provides a high-quality and reliable data foundation for subsequent violation judgment based on complete trajectories.

[0099] In the embodiments of this application, the flight control platform extracts key features from the geographic coordinate trajectory of each vehicle, compares the key features with the violation judgment rules corresponding to the current waypoint, and obtains the violation result based on the logical threshold model, specifically for:

[0100] Based on the vehicle's geographic coordinate trajectory, key trajectory features are obtained. These key trajectory features include the duration and location of the trajectory stationary state used for illegal parking judgment, the relationship between the trajectory and the preset solid line electronic fence used for solid line lane change judgment, or the relationship between the trajectory and the preset lane electronic fence used for illegal lane occupation judgment.

[0101] The key features of the trajectory are compared with the violation judgment rules of the current waypoint to obtain the comparison results. The comparison results are then input into the logic threshold model to output the violation results.

[0102] The logic threshold model transforms pre-defined violation judgment rules, such as "no parking" and "no lane change," into specific logical judgment statements, which include one or more "conditions" and corresponding "thresholds" that must be met. The model only outputs a "violation" judgment when the input trajectory key feature data simultaneously satisfies AND logic or sequentially satisfies all pre-defined conditions. This design allows the system to accurately define complex violations; for example, requiring vehicles not only to enter the emergency lane but also to be in motion for a certain period, thereby significantly reducing misjudgments caused by vehicles legally using lanes, such as for emergency lane avoidance, and improving the accuracy and fairness of the processing.

[0103] For example, consider monitoring "illegal occupation of the emergency lane on a highway." The system's preset rule at the corresponding waypoint is: non-emergency / rescue vehicles entering the emergency lane and not making a brief stop constitutes a violation. The corresponding logical threshold model is programmed to simultaneously meet the following three conditions: the vehicle's trajectory point continuously falls within the geofence area marked "emergency lane"; the vehicle type is non-emergency / rescue vehicle; and the vehicle's average speed within the emergency lane is greater than a low threshold, recorded as >5 km / h. When these three conditions are met, the logical threshold model immediately outputs "Violation Result: Illegal occupation of the emergency lane confirmed" and triggers the subsequent drone evidence collection process.

[0104] Furthermore, historical data can be used to assist in the judgment of the violation target in the violation results: for example, the license plate recognition algorithm can be used to verify whether the vehicle is on the "long-term illegal parking blacklist", or the data from surrounding surveillance cameras can be combined to confirm whether the lane-changing behavior is continuous.

[0105] Optionally, in some embodiments of this application, the flight control platform is also used for:

[0106] After the drone completes the initial evidence collection, the violation is recorded, and after a preset delay time, the status of each drone is reassessed, and one drone is assigned to fly back to the violation location for a second round of evidence collection.

[0107] The results of the secondary evidence collection are used to verify whether the violation is ongoing. If it is ongoing, an alarm is triggered; if it has been eliminated, the violation status is marked as ended and the relevant violation information is archived.

[0108] In an optional embodiment of this application, in order to construct a complete chain of evidence and achieve closed-loop management of violations, the flight control platform also executes a refined secondary verification and event final state management process after the initial evidence collection. Specifically:

[0109] After the drone completes its initial surround scan of the violating vehicle and transmits high-definition evidence, such as a clear license plate, vehicle model, and close-up of the violation, the flight control platform immediately generates a structured violation event record. This record includes the violation type, time, precise geographical location, vehicle characteristics, and an index of the initial video / image capture. Subsequently, the system initiates a configurable delay timer. This delay is pre-set based on the type of violation, road segment attributes, and processing requirements. For example, it can be set to 5-10 minutes for illegal parking and 2-5 minutes for obstructing the emergency lane, providing a reasonable "correction period."

[0110] When the delay period ends, the system automatically triggers a secondary verification task. The flight control platform first reassesses the real-time status of the entire drone swarm, taking into account factors such as: the distance of each drone to the violation location; selecting drones with a lighter current task load and sufficient remaining battery power to support evidence collection and return; and ensuring that the secondary verification does not significantly interfere with higher-priority real-time monitoring tasks. Based on the above assessment, the platform uses its scheduling algorithm to assign the most suitable drone—which could be the original evidence-collecting drone or another, better one—to leave the current task flow and fly to the violation location to perform a secondary perimeter evidence collection. The core purpose of this evidence collection differs from the first, focusing on verifying the status: whether the violation continues, such as whether illegally parked vehicles are still in place or whether emergency lanes are still blocked.

[0111] If the violation persists: The system determines that the violation is ongoing and escalates the action accordingly. In addition to marking the event in the event log, the system can automatically trigger an alarm. Alarm methods include, but are not limited to: visual and audible alerts on the command center's large screen, automatically generating an incident report containing location and evidence and pushing it to nearby officers or the traffic police's operational system, and even remotely warning and driving away violating vehicles by linking with roadside loudspeakers or information boards.

[0112] If the violation has been rectified: the system determines that the violation status has been corrected and marks the relevant violation event as closed. The system will archive the entire event, including evidence of the initial violation and evidence of secondary verification, forming a complete chain of evidence, and store it in the database for future querying, statistics, or subsequent processing.

[0113] In a traffic violation detection system based on unmanned aerial vehicles (UAVs) according to an embodiment of this application, multiple UAVs are used to receive and execute flight commands issued by the flight control platform, fly autonomously along a pre-set cruise route, collect image data of the waypoints they are responsible for in real time and transmit it back to the flight control platform; they are also used to activate the surround evidence collection mode after receiving the surround evidence collection command, fly towards the violating vehicle and obtain information about the violating vehicle.

[0114] The drone receives and executes flight commands from the flight control platform, autonomously flies along a pre-set cruise route, and collects real-time image data of the waypoints it is responsible for, transmitting it back to the flight control platform. Specifically, it is used for:

[0115] After arriving at the designated waypoint, it enters hover detection mode;

[0116] In hover detection mode, the drone hovers at a fixed altitude and adjusts the pitch and yaw angles of the camera equipment through the onboard gimbal to continuously collect images of the waypoint area from the optimal perspective.

[0117] After completing image acquisition at the current waypoint, the drone autonomously flies to the next waypoint on the cruise route.

[0118] For example, the flight control platform sets up a monitoring waypoint 100 meters above the main road near the school gate, with the mission attribute being "illegal parking hotspot area". The drone autonomously flies to the point, enters hover detection mode at an altitude of 60 meters, and maintains a stable position while resisting crosswinds. At this time, the gimbal integrated on the flight control platform activates the preset "road segment monitoring" strategy. First, it adjusts the pitch angle to 40 degrees so that the camera's field of view can simultaneously cover the no-parking yellow line area near the roadside and the two adjacent driving lanes. Then, the gimbal performs a slow, automatic horizontal scan at a speed of 15 degrees per minute within a range of 30 degrees to the left and right of the waypoint. This is equivalent to dynamically monitoring a road segment of about 100 meters using a fixed point. Afterward, the drone begins to continuously shoot high-definition video. During the 3-minute hovering acquisition process, for example, the following scene was fully recorded: a black sedan drives up and stops within the no-parking yellow line; the driver gets out and leaves; during this period, the slow scan of the gimbal ensures that the vehicle is continuously tracked in the frame for a sufficient duration. All video data is transmitted back in real time. After the preset 3-minute acquisition time ends, or upon receiving a command from the flight control platform, the drone exits hover mode, accelerates smoothly, and flies to the next waypoint.

[0119] Through the aforementioned "hovering detection mode," the drone transforms its original mobile observation platform into a temporary, stable, and intelligent "aerial fixed monitoring point," providing high-quality, continuous, and complete image input for backend analysis. This is the primary physical basis for ensuring the accuracy of subsequent vehicle detection, trajectory tracking, and violation determination.

[0120] Furthermore, the drone activates its surround-spot evidence-gathering mode to fly towards the offending vehicle and obtain its information, specifically for:

[0121] Receive the geographical coordinates of the illegal vehicle from the flight control platform; based on the geographical coordinates and the drone's real-time positioning, plan and fly to the initial evidence collection location above the illegal vehicle;

[0122] Centered on the offending vehicle, the system maintains a constant radius on the horizontal plane and performs closed-loop orbiting flight at a constant speed and altitude. During the orbiting flight, the gimbal is synchronously controlled to keep the camera continuously pointed at the vehicle and continuously take pictures or record videos to obtain clear images from multiple angles, including the vehicle's license plate number, model, and appearance features.

[0123] During the orbital flight, the gimbal on the drone and the camera equipment simultaneously initiate intelligent tracking. Based on the relative position of the drone and the vehicle, the gimbal uses a built-in tracking algorithm to adjust the pitch and yaw axes in real time and automatically, ensuring that the optical axis of the camera equipment is always precisely locked onto the vehicle, offsetting the changes in perspective caused by the drone's own movement, and achieving the effect of "aiming while moving".

[0124] During this process, high-definition camera equipment performs multimodal evidence collection: it typically records the entire process continuously at a high frame rate and can automatically trigger high-resolution still photo capture at specific angles to the vehicle, such as facing the front, rear, or side of the vehicle. Through this collaborative operation, the system can efficiently acquire multi-angle, all-round visual evidence containing clear vehicle license plate numbers, complete vehicle model features, vehicle exterior details, and the scene of the violation, and transmit the data back in real time to form a complete electronic evidence chain.

[0125] Furthermore, in the specific implementation process, super-resolution reconstruction technology can be used to improve the clarity of license plates and vehicle details, ensuring the identifiability of evidence, while also ensuring continuous timestamps, matching of geographical coordinates and waypoints, and no video editing. Low-quality images (such as blurry or obscured images) are automatically replaced, and a standardized evidence report is generated (including violation type, time, location, vehicle information, complete trajectory, and high-definition photos / videos).

[0126] In addition, when the image shifts due to airflow during drone flight, the system performs coordinate correction every 5 minutes: by identifying static references such as road markings (lane lines, stop lines) and streetlights in the video, it calculates the pixel offset and corrects the vehicle trajectory coordinates.

[0127] Similarly, the system regularly (monthly) reviews historical violation data: analyzes misjudgment cases, such as non-motorized vehicles being identified as illegally parked, and abnormal trajectories caused by shadows, and updates the deep learning object detection model through transfer learning to improve adaptability to complex scenarios.

[0128] This application discloses a traffic violation detection system based on unmanned aerial vehicles (UAVs). By constructing a detection system that coordinates multiple UAVs with a flight control platform, it achieves full automation of the entire process of traffic violation detection, from "wide-area detection, intelligent judgment, automatic evidence collection to closed-loop management." It innovatively combines the mobility of UAV swarms with a flight control platform, utilizing data-driven flight path planning, precise vehicle trajectory tracking, violation judgment based on multi-condition logic, and dynamic task scheduling. This not only greatly expands the geographical scope of monitoring and improves processing response speed and coverage efficiency, but also significantly improves the accuracy of violation identification through a rigorous secondary verification mechanism and electronic evidence chain generation. Ultimately, it effectively solves the problems of blind spots in traditional fixed monitoring and low efficiency of manual patrols in an intelligent and unmanned manner, providing an efficient and reliable technical means for modern traffic management.

[0129] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0130] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0131] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "over," or "on top" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," or "beneath" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0132] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0133] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A traffic violation detection system based on unmanned aerial vehicles (UAVs), characterized in that, include: A flight control platform and multiple drones that are communicatively connected to the flight control platform; The flight control platform is used to issue flight commands to the multiple UAVs respectively, and receive image data of each waypoint transmitted back by each UAV; it is also used to process the image data and track the trajectory, generate the geographic coordinate trajectory of each vehicle appearing in the image data; and extract key features from the geographic coordinate trajectory of each vehicle, compare the key features with the violation judgment rules corresponding to the current waypoint, and obtain the violation result based on the logical threshold model; if the violation result is a confirmed violation, then based on the preset scheduling strategy, a surround evidence collection command is issued to at least one target UAV among the multiple UAVs. The multiple drones are used to receive and execute flight commands issued by the flight control platform, fly autonomously along a pre-set cruise route, collect image data of the waypoints they are responsible for in real time and transmit it back to the flight control platform; they are also used to activate the surround evidence collection mode after receiving the surround evidence collection command, fly towards the violating vehicle and obtain information about the violating vehicle.

2. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The flight control platform is also used to perform route planning before issuing flight commands, specifically including: Obtain geographic information and historical traffic violation data for the target area; Based on the geographic information and historical traffic violation data, multiple waypoints and the traffic monitoring scenario type corresponding to each waypoint are determined on the geographic map of the target area, and associated violation judgment rules and trigger parameters are configured for each waypoint; based on the status and location of the multiple drones and the location and priority of the multiple waypoints, a cluster task allocation algorithm is adopted to generate a non-conflicting cruise route for each drone that covers all or part of the specified monitoring waypoints.

3. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, When the flight control platform determines multiple monitoring waypoints and the corresponding traffic monitoring scenario type for each monitoring waypoint on the geographic map of the target area, it is specifically used for: The flight control platform analyzes the historical traffic violation data to identify areas with high violation rates, road sections with frequent solid line lane changes, or historical locations of traffic accidents, and determines the identified locations as monitoring waypoints; and obtains the traffic supervision scenario type for each monitoring waypoint based on the geographical information. The traffic monitoring scenarios include traffic hub intersections, urban commercial areas, areas with frequent illegal parking, or highway emergency lanes.

4. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, The flight control platform configures associated violation judgment rules and trigger parameters for each monitored waypoint, specifically for: When the traffic monitoring scenario of the monitored route point is an urban commercial area or a hotspot for illegal parking, a rule for judging illegal parking is configured, and the triggering parameter is that the duration of the vehicle being stationary exceeds a preset threshold. When the traffic monitoring scenario of the monitored waypoint is a traffic hub intersection, a solid line lane change judgment rule is configured, and the trigger parameter is the vehicle trajectory crossing the electronic geofence associated with the solid line lane. When the traffic monitoring scenario of the monitored waypoint is the emergency lane of a highway, an illegal occupation of the emergency lane judgment rule is configured, and the trigger parameter is that the vehicle trajectory enters the electronic geofence area associated with the rightmost lane of the road and the width is not greater than the preset width.

5. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The flight control platform processes and tracks the image data, specifically for: Perform perspective transformation correction on the received image data from the perspective of the UAV to obtain the corrected image data; Based on the corrected image data, vehicles are located in each frame using a deep learning object detection model, and the vehicle's bounding box and category are output. Based on the vehicle's bounding box and category in the current frame and the previous frame, a unique trajectory identifier is assigned to and maintained for the same vehicle using a data association algorithm, generating continuous pixel-level trajectories for each vehicle. The pixel coordinates in the pixel-level motion trajectory are combined with the real-time position and attitude data of the UAV that acquired the image data and mapped to coordinates in a unified geographic coordinate system to generate the geographic coordinate trajectory of the vehicle.

6. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 5, characterized in that, The flight control platform assigns and maintains a unique trajectory identifier for each vehicle through a data association algorithm, generating continuous pixel-level trajectories for each vehicle, specifically for: Based on the state prediction algorithm, the predicted position of the vehicle's bounding box in the next frame is predicted according to the vehicle's trajectory state in the current frame. Then, using the optimal matching algorithm, the vehicle bounding box actually obtained by the deep learning object detection model in the next frame is optimally matched with the predicted position of the bounding box to obtain the matching result. Based on the matching result, the trajectory identifier of the current frame is passed to the vehicle bounding box of the successfully matched vehicle in the next frame, until the vehicle leaves the monitoring area of ​​the current waypoint.

7. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The flight control platform extracts key features from the geographic coordinate trajectory of each vehicle, compares these key features with the violation judgment rules corresponding to the current waypoint, and obtains the violation result based on a logical threshold model. Specifically, this is used for: Based on the geographic coordinate trajectory of the vehicle, key trajectory features are obtained. These key trajectory features include the duration and location of the trajectory stationary position used for illegal parking judgment, the crossing relationship between the trajectory and the preset solid line electronic fence used for solid line lane change judgment, or the intrusion relationship between the trajectory and the preset lane electronic fence used for illegal lane occupation judgment. The key features of the trajectory are compared with the violation judgment rules of the current waypoint to obtain the comparison results, and the comparison results are input into the logic threshold model to output the violation results.

8. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 1 or 2, characterized in that, The flight control platform is also used for: After the drone completes the initial evidence collection, the violation is recorded, and after a preset delay time, the status of each drone is reassessed, and one drone is assigned to fly back to the violation location for a second round of evidence collection. The results of the secondary evidence collection are used to verify whether the violation is ongoing. If it is ongoing, an alarm is triggered; if it has been eliminated, the violation status is marked as ended and the relevant violation information is archived.

9. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The UAV receives and executes flight commands issued by the flight control platform, autonomously flies along a pre-set cruise route, and collects real-time image data of the waypoints it is responsible for and transmits it back to the flight control platform. Specifically, it is used for: After arriving at the designated waypoint, it enters hover detection mode; In the hover detection mode, the UAV hovers at a fixed altitude and adjusts the pitch and yaw angles of the camera device through the onboard gimbal to continuously capture images of the waypoint area from the optimal perspective. After completing image acquisition at the current waypoint, the UAV autonomously flies to the next waypoint on the cruise route.

10. The traffic violation detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The drone activates its surround-and-collection mode, flies towards the offending vehicle, and obtains information about it. Specifically, it is used for: Receive the geographical coordinates of the illegal vehicle from the flight control platform; based on the geographical coordinates and the drone's real-time positioning, plan and fly to the initial evidence collection position above the illegal vehicle; Centered on the offending vehicle, the system maintains a constant radius on the horizontal plane and performs closed-loop orbiting flight at a constant speed and altitude. During the orbiting flight, the gimbal is synchronously controlled to keep the camera continuously pointed at the vehicle and continuously take pictures or record videos to obtain clear images from multiple angles, including the vehicle's license plate number, model, and appearance features.

Images