Unmanned aerial vehicle cooperative traffic accident handling system and control method

Abstract

The invention relates to the technical field of traffic assistance, in particular to an unmanned aerial vehicle cooperative traffic accident handling system and a control method, and the method comprises the following steps: starting the system and entering peak daily patrol; the VSDM continuously collects and uploads the traffic flow speed and density of each road section, and the system generates a traffic flow density thermodynamic diagram according to the speed and plans a gridding fixed patrol route for the unmanned aerial vehicle; when the VSDM detects that the average speed of the road section is reduced to exceed a preset threshold value in a preset time window or the unmanned aerial vehicle returns an image to detect that the vehicle is static for a long time, triggering suspected accident early warning and recording an early warning position and associated traffic flow data; if not, returning to S2 to continue patrol; and sending a quick response instruction to the unmanned aerial vehicle closest to the early warning position. According to the invention, the problems of lagging accident discovery and confirmation, low disposal efficiency, insufficient module collaboration and low decision subjectivity are solved.

Description

Technical Field

[0001] This invention relates to a traffic accident handling system and control method using unmanned aerial vehicles (UAVs), belonging to the field of traffic assistance technology. Background Technology

[0002] Currently, emergency response to traffic accidents mainly relies on two technical solutions: one is the traditional model of "manual discovery + traffic police handling," which involves discovering accidents through methods such as drivers of passing vehicles reporting accidents, manual patrols of roadside monitoring, or reports from traffic assistants. The command center then dispatches nearby traffic police to the scene, who guide the accident vehicles to a safe area and handle the situation. The other is a semi-automated solution of "intelligent monitoring + manual dispatch." This solution uses roadside monitoring cameras combined with vehicle recognition algorithms to detect accidents (such as vehicles remaining stationary for extended periods or vehicle collision pattern recognition), or uses induction coils buried in the road surface to detect abnormal traffic flow speeds. When the system determines a suspected accident, it sends an early warning to the traffic command center, where dispatchers manually verify the information and assign traffic police to handle the situation.

[0003] Some advanced solutions introduce drone assistance, but only as a tool for collecting images at the accident scene. That is, before the traffic police arrive, the command center remotely controls the drone to fly to the accident site to take pictures to assess the severity of the accident. The drone does not participate in core processes such as accident confirmation and guidance. The selection of temporary stopping points still relies on the traffic police's on-site experience and judgment, and no intelligent decision-making mechanism has been formed. The following technical problems exist: 1. Delayed accident detection: Traditional manual detection relies on individual reporting, which is highly random and the average detection time exceeds 10 minutes; although semi-automated solutions can monitor and warn through technology, the monitoring cameras are easily affected by weather and obstruction, and the induction coils can only detect speed abnormalities but cannot confirm the accident, requiring secondary manual verification. The overall confirmation cycle is still as long as 5-8 minutes, which may cause the golden time for congestion relief to be missed. 2. Low efficiency in handling traffic accidents: Traffic police are severely affected by traffic congestion when they arrive at the scene. During peak hours on main roads in the city, the average arrival time exceeds 20 minutes. During this time, accident vehicles occupy lanes, causing continuous congestion and even secondary accidents. The selection of temporary parking spots relies on the experience of traffic police, and their unfamiliarity with the parking conditions of surrounding roads may lead to unreasonable guidance routes, further prolonging the congestion time. 3. Insufficient resource collaboration: Each device and module exists as an "information silo," and data cannot be shared in real time between monitoring equipment, drones, and traffic police dispatch systems. Drones are only used as a single image acquisition tool and have not formed a collaborative closed loop with speed detection and decision guidance, resulting in low overall system utilization.

[0004] 4. High degree of subjectivity in decision-making: There is no standardized basis for the selection of temporary parking points, which relies entirely on the personal experience of traffic police. In complex road conditions (such as construction sections and temporary control areas), decision-making errors are prone to occur, resulting in accident vehicles being unable to stop quickly and affecting the efficiency of congestion relief. Therefore, there is an urgent need to improve a traffic accident handling system and control method that is coordinated by drones to solve the above-mentioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a traffic accident handling system and control method in collaboration with unmanned aerial vehicles (UAVs) to solve the problems of delayed accident detection, low handling efficiency, insufficient resource coordination, and strong subjectivity in decision-making.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A traffic accident handling system and control method in collaboration with unmanned aerial vehicles (UAVs) includes the following steps: S1: The system starts up and enters peak-hour routine patrol; S2:VSDM continuously collects and uploads traffic speed and density data for each road segment. The system generates a traffic density heat map based on the speed and plans gridded fixed patrol routes for drones. S3: Abnormal Warning Judgment: When VSDM detects that the average speed of the road segment decreases by more than a preset threshold (e.g., 50%) within a preset time window (e.g., 1 minute) or the drone-transmitted image detects that a vehicle has been stationary for a long time (e.g., more than 30 seconds), a suspected accident warning is triggered and the warning location and associated traffic flow data are recorded. If not, return to S2 to continue patrolling; S4: Dispatch the nearest drone for rapid response: Send a rapid response command to the drone closest to the warning location. The drone adopts a "straight line priority + obstacle avoidance optimization" route to quickly reach the warning location and descends to the confirmation altitude after arrival to take on-site photos and conduct voice inquiries. The collected images and audio are transmitted back to IPDM through RTSP or other real-time channels. S5:IPDM, based on knowledge graphs and large models, generates at least several candidate temporary stops in three steps: candidate point screening, road condition assessment, and feasibility reasoning, and outputs the best temporary stop with the highest score and recommended guidance route. S6: Drone performs ground guidance: After receiving IPDM instructions, it projects guidance arrows above the accident vehicle and broadcasts voice instructions to guide the vehicle to the optimal temporary stopping point, while tracking and adjusting the guidance in real time; S7:ATRM receives accident information and automatically matches the nearest available traffic police officer to dispatch a task. The traffic police officer uses a mobile terminal to navigate to the temporary stop point, completes the accident handling on-site, and uploads the handling results to the system. S8: After the incident is resolved, the system records the results. The drone resumes its daily patrol route, and the system returns to S2 to continue monitoring and patrolling.

[0007] Preferably, the specific thresholds for early warning determination in S2 are: the average speed of the road segment decreases by ≥50% or the single vehicle remains stationary for ≥30 seconds within 1 minute; the latitude and longitude positioning error of the warning location is ≤5 meters; and the system completes the early warning registration within 60 seconds and confirms the arrival of the drone within 5 minutes after triggering the warning.

[0008] Preferably, the priority criteria for selecting candidate temporary parking points in S4 using the knowledge graph include: shortest distance from the warning location, space capacity (sufficient vehicle parking area), minimal impact on main traffic flow, and compliance with road control and construction restrictions. The large model considers the candidate point's travel time, vehicle type (small car / large car), degree of damage, and real-time congestion prediction during evaluation, and outputs a comprehensive feasibility score.

[0009] Preferably, the dispatch rules of ATRM in S6 include: prioritizing the matching of the nearest and available traffic police based on the traffic police's current GPS location, duty status, and historical response time; dispatching also includes preliminary accident information, optimal temporary stop coordinates, and estimated arrival time; traffic police provide feedback on status (received / on the way / arrived / handled) via mobile terminal, and the system updates the handling progress accordingly and forces the upload of a handling report after the handling is completed to end the emergency process.

[0010] Preferably, it includes a vehicle speed detection module (VSDM) for real-time collection of vehicle speed and vehicle distribution density along the road and sending the collected data to the system data bus; The drone patrol module includes at least a high-definition camera, a directional microphone, a ground projection device, a lidar obstacle avoidance sensor, and an RTSP data transmission module. It is used to perform daily patrols, rapid response and on-site guidance, and transmit multimodal data back to the system. The Intelligent Parking Decision Module (IPDM) includes a knowledge graph database and a large model inference unit. The knowledge graph database stores static / dynamic road information such as available parking spots, road grades, traffic rules, and construction / control information. The large model inference unit is used to integrate VSDM real-time data, UAV-transmitted images, and the knowledge graph to output the optimal parking spot and guidance route. The Automatic Traffic Response Module (ATRM) is used to receive accident information, temporary stopping points, and guidance progress output by the IPDM, and automatically match and dispatch processing instructions based on the real-time location and working conditions of traffic police. The traffic command center monitoring platform and interactive bus are used to realize standardized data access, real-time synchronization, visualization and manual intervention between the modules.

[0011] Preferably, the VSDM consists of a microwave radar and video speed measurement equipment set every 500 meters along the road, with a sampling frequency of 1 time / second.

[0012] Preferably, the UAV patrol module has a daily patrol altitude of 50 meters, a rapid response confirmation altitude of 20 meters, and a hovering altitude of 15 meters during guidance; the UAV's rapid response flight speed is preferably 120 km / h and it has the ability to adjust its flight path with "straight line priority + dynamic obstacle avoidance".

[0013] Preferably, the large model inference unit in the IPDM adopts a lightweight LLM to achieve edge / near-edge inference, and the decision-making process includes: ① calling up the available parking points within a preset radius (preferably 3 kilometers) around the warning point based on the knowledge graph and filtering the candidate set; ② evaluating the congestion and travel time of the reachable path of each candidate point based on the real-time traffic conditions of VSDM; ③ evaluating the parking feasibility of the candidate points by combining the vehicle type and damage degree returned by the UAV and outputting the optimal temporary parking point and guidance route with the highest comprehensive score.

[0014] Preferably, the drone projection device is used to project bright indicator arrows on the ground to guide accident vehicles away from the main lane. The projected arrows are 0.5 meters wide and accompanied by voice broadcast instructions. Furthermore, the drone maintains closed-loop communication with the IPDM during the guidance process to adjust the projection direction and broadcast content in real time.

[0015] Preferably, the system implements standardized access to multimodal data at the data interface layer, including but not limited to VSDM velocity stream data, UAV RTSP image stream, knowledge graph structured data and ATRM feedback information, and realizes millisecond-level / second-level data synchronization and command issuance based on the data bus.

[0016] The present invention has at least the following beneficial effects: In this invention, real-time monitoring of vehicle flow speed by VSDM combined with dynamic patrol by drones enables early warning of suspected accidents within 1 minute and confirmation by drones within 5 minutes, shortening the accident confirmation cycle to less than 6 minutes. This buys time for congestion management. Simultaneously, leveraging the aerial mobility of drones, preliminary confirmation and guidance work can be completed instead of manual labor. Accident vehicles can leave the main lane 10-15 minutes before traffic police arrive, significantly reducing the duration of congestion on main roads. Intelligent decision-making for temporary stopping points is achieved through large-scale models and knowledge graphs, avoiding guidance errors caused by empiricism. Furthermore, a unified data interaction platform is constructed to achieve real-time data synchronization and efficient command transmission between VSDM, drones, IPDM, and ATRM, forming a complete collaborative closed loop of "monitoring-confirmation-decision-guidance-handling," improving the overall system efficiency. A standardized temporary stopping point decision model is further established, integrating static road knowledge and real-time traffic data to output objective and optimal temporary stopping points and guidance routes, ensuring the scientific accuracy of the decisions. It effectively solves the problems of delayed accident discovery and confirmation, low handling efficiency, insufficient module coordination, and weak decision-making subjectivity. Attached Figure Description

[0017] Figure 1 This is a flowchart of a control method for a traffic accident handling system using unmanned aerial vehicles (UAVs) according to the present invention. Detailed Implementation

[0018] The following will describe in detail the implementation of this application with reference to the accompanying drawings and embodiments, so that the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0019] Definitions of abbreviations and key terms: UAV: Unmanned Aerial Vehicle, also known as drone, is the core execution device used in this invention for road patrol, accident confirmation, and guidance.

[0020] KG: Knowledge Graph, a semantic network that stores related information such as road parking spots and traffic rules in a structured form.

[0021] LLM: Large Language Model, which has powerful reasoning capabilities, is used in this invention to fuse multi-source data to determine the optimal stopping point.

[0022] VSDM: Vehicle Speed ​​Detection Module, a collection of hardware units used to collect real-time vehicle speed data for various road segments.

[0023] IPDM: Intelligent Temporary Parking Decision Module, a core module that integrates large models and knowledge graphs to achieve accurate decision-making for temporary parking locations.

[0024] ATRM: Artificial Traffic Police Response Module, a human-computer interaction module that receives system instructions and completes on-site handling.

[0025] RTSP: Real-Time Streaming Protocol, a network protocol used for real-time transmission of footage captured by drones. Example

[0026] like Figure 1 As shown, this embodiment provides a traffic accident handling system and control method in collaboration with unmanned aerial vehicles (UAVs), which is described in the context of peak hours on urban main roads. The example scenario is a main road about 6km long, with a set of vehicle speed detection modules (VSDM, a combination of microwave radar and video speed measurement) deployed every 500m along the road, and several of them configured.

[0027] System Deployment and Main Components: 1.1 Vehicle Speed ​​Detection Module (VSDM) 1.1.1 Deployment method: A set of sensors is deployed every 500m along the road, and each set includes a microwave radar and a video speed measurement device.

[0028] 1.1.2 Sampling frequency: 1Hz (1 time / second).

[0029] 1.1.3 The output data includes event or sampling time, average vehicle speed of road segment, vehicle density, etc.

[0030] 1.1.4 Note: The project implementation can map the fields of the above output data to structured formats such as JSON and Protobuf.

[0031] 1.2 Unmanned Aerial Vehicle (UAV) Patrol Module 1.2.1 Hardware configuration: Several industrial-grade multi-rotor UAVs, each equipped with a high-definition camera, directional microphone, ground directional projection equipment, lidar obstacle avoidance unit, RTSP backhaul module and GPS / IMU.

[0032] 1.2.2 Flight parameters: patrol altitude 50m, confirmation altitude 20m, guidance hovering altitude 15m; the maximum safe flight speed for rapid response is preferably 120 km / h, and straight-line priority is adopted with supplementary flight.

[0033] 1.3 Intelligent Temporary Stop Decision Module (IPDM) 1.3.1 Composition: Knowledge Graph Database (KG) and Lightweight Large Language Model (LLM) inference unit.

[0034] 1.3.2 Knowledge Graph Content: Static and semi-static metadata such as road parking locations (latitude and longitude, area, entrance and exit attributes, etc.), road grade, traffic rules, construction and temporary control information, and supports real-time updates.

[0035] 1.4 ATRM (Automatic Traffic Police Response) Module 1.4.1 Function: Receives accident information, optimal temporary stopping point and guidance progress output by IPDM, automatically matches and dispatches handling instructions based on the real-time location and duty status of traffic police, and supports mobile terminal feedback.

[0036] 1.5 Traffic Control Center and Data Bus 1.5.1 Functionality: Enables multimodal data access, message queues, visual monitoring, manual intervention access, and audit log functionality.

[0037] 1.5.2 Performance target examples: Critical path message latency target <200ms; System availability target ≥99.5%.

[0038] 2. Routine patrol phase (peak hours) 2.1 The VSDM continuously collects and reports the average speed and traffic density of each road segment to the command center; the platform generates a traffic density heat map based on the speed value, with red indicating dense traffic (speed < 20km / h), yellow indicating moderate traffic (20km / h ≤ speed ≤ 60km / h), and green indicating smooth traffic (speed > 60km / h). 3. Emergency Response Phase (Suspected → Confirmed → Handling) 3.1 Abnormal Early Warning Triggering Conditions 3.1.1 Example of conditions: When the VSDM detects a decrease in the average speed of the road segment of ≥50% within a preset time window (example 1 min), or when the UAV transmits images and detects a single vehicle stationary for ≥30 s, the system triggers a suspected accident warning; the system records the warning location and associates it with the surrounding real-time traffic flow data and heat map.

[0039] 3.1.2 Location accuracy: The latitude and longitude positioning error of the warning point is preferably ≤5 m.

[0040] 3.2 Recent Rapid Response of Unmanned Aerial Vehicles 3.2.1 Scheduling logic: The system selects the UAV closest to the warning location and whose current resources are available (battery power and status allow) from the UAV cluster, and issues a rapid response command to it.

[0041] 3.2.2 Flight Strategy: The selected UAV will travel to the target in a "straight-line priority + dynamic obstacle avoidance" mode, with a maximum flight speed of 120 km / h. If a no-fly zone or extreme weather conditions render the target unreachable, a retreat strategy will be triggered.

[0042] 3.3 Arrival and On-site Confirmation of the Drone 3.3.1 Actions after arrival: The drone descends to the confirmed altitude (approximately 20 m), circles around the accident site to capture keyframes, and conducts voice inquiries and records voice responses via a directional microphone.

[0043] 3.3.2 Data feedback: Keyframes and audio are fed back to IPDM in real time via RTSP or event channel. The image analysis module performs target detection and damage assessment on the keyframes.

[0044] 3.4 IPDM Intelligent Temporary Stop Decision 3.4.1 Step 1: Knowledge Graph Retrieval - Retrieve the set of available stops within a 3 km radius of the warning point from the KG and obtain the static attributes of each candidate point (latitude and longitude, available area, entrances and exits, right-of-way and temporary control information, etc.).

[0045] 3.4.2 Step 2: Traffic condition assessment – ​​Calculate the estimated travel time and congestion level of each candidate point by combining real-time traffic data from VSDM and short-term congestion prediction.

[0046] 3.4.3 Third step: Feasibility reasoning - LLM The image analysis results (vehicle type, damage level, etc.) are fused with KG and road condition data from multiple sources. The candidate points are normalized and scored according to preset weights, and the optimal temporary stop, alternative points, recommended guidance routes and confidence scores with the highest comprehensive scores are output.

[0047] 3.5 Drone Projection Guidance 3.5.1 Projection and Broadcast: After receiving the IPDM command, the UAV projects a bright guide arrow (0.5m wide) about 15m above the accident vehicle and broadcasts short sentences to prompt the driver to follow the ground guidance to leave the main lane.

[0048] 3.5.2 Real-time tracking: During the guidance process, the drone stays about 50m in front of the vehicle, continuously tracking the vehicle based on vision and positioning, and adjusting the projection direction and voice content in real time to ensure the guidance is dynamic and effective.

[0049] 3.6 ATRM Dispatch and Response Closed Loop 3.6.1 Dispatch Rules: ATRM prioritizes matching the nearest and available traffic police officer based on the traffic police officer's current GPS location, duty status, and historical response performance; the dispatch content includes basic accident information, optimal temporary stop coordinates, navigation route, and estimated arrival time.

[0050] 3.6.2 Traffic Police Feedback: Traffic police officers report the status (received, en route, arrived, handled) via mobile APP, and upload a handling report after handling is completed; the system updates the handling progress and records the event closed-loop log accordingly.

[0051] 3.6.3 End Process: After the disposal is completed, the system records the disposal results and key data, the drone resumes its daily patrol route, and the system returns to the daily patrol status.

[0052] 4. Criteria, Time Series, and Numerical Values 4.1 Example of warning judgment threshold: VSDM if the average speed of the road segment decreases by ≥50% or a single vehicle remains stationary for ≥30 seconds within a 1-minute window.

[0053] 4.2 Example of time-limited objectives: System issues early warning ≤ 60 s; UAV arrives and completes keyframe transmission ≤ 5 min; 5. Redundancy and rollback strategies 5.1 Unreachable or delayed arrival of drones: If the estimated arrival time of the drone after the instruction is issued is >5 minutes or restricted by no-fly zones / extreme weather, the system will automatically: recommend the optimal temporary parking point to the nearest road traffic assistant or dispatch traffic police; pop up a manual intervention prompt in the command center; and record the event as a special process sample for post-event analysis.

[0054] 5.2 Communication Interruption Situation: Critical event information will be prioritized for redundancy using dual-channel backhaul (4G / 5G and satellite link); if both fail, the system will notify ATRM with brief textual elements (latitude and longitude, event time, brief description) and trigger the manual handling process.

[0055] 5.3 False Alarm Handling and Model Iteration: All false alarm samples and their correction results are stored in the database for subsequent offline and online training, threshold adjustment and model upgrade, forming a continuous learning loop.

[0056] 6. Privacy, Security and Compliance 6.1 Data Protection Measures: Image and voice data shall be encrypted and stored and access controlled in accordance with applicable laws and regulations, and minimum retention periods and configurable de-identification strategies shall be implemented; Use only within the scope authorized by law and necessary for incident handling; unauthorized disclosure is prohibited.

[0057] 6.2 Flight Compliance: Drone flight plans must be coordinated with urban management and civil aviation authorities. The system should include no-fly zones, dynamic approval, and minimum safe altitude constraints to ensure airspace compliance and flight safety.

[0058] 6.3 Identity and Audit: The command center and traffic police mobile terminal adopt a certificate-based identity verification mechanism; the system should record all dispatch, feedback and handling operations for post-event auditing.

[0059] This embodiment effectively shortens the time for accident confirmation and initial traffic management (from the traditional >10 min to ≤6 min) through a collaborative closed loop of VSDM early warning, rapid drone confirmation, KG+LLM multi-source fusion decision-making, drone projection guidance, and ATRM dispatch, thereby improving road restoration efficiency and reducing the risk of secondary accidents. Based on the structured available parking point information provided by the knowledge graph and the reasoning ability of the large model, the selection of temporary parking points is made more scientific and standardized, reducing guidance errors caused by traffic police subjective judgment. Drone projection and voice guidance can provide real-time aerial guidance for accident vehicles before the arrival of traffic police, significantly improving initial traffic management efficiency and alleviating congestion on the main lane.

[0060] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A control method for a traffic accident handling system coordinated by unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: The system starts up and enters its peak-hour routine patrol. VSDM continuously collects and uploads traffic speed and density data for each road segment. The system generates a traffic density heat map based on the speed and plans gridded fixed patrol routes for drones. When the VSDM detects that the average speed of the road segment has decreased by more than a preset threshold within a preset time window, or when the drone transmits images and detects that a vehicle has been stationary for a long time, it triggers a suspected accident warning and records the warning location and associated traffic flow data; otherwise, it returns to S2 to continue patrolling. A rapid response command is sent to the drone closest to the warning location. The drone quickly arrives at the warning location using a "straight line priority + obstacle avoidance optimization" route and descends to the confirmation altitude after arrival to take on-site photos and conduct voice inquiries. The collected images and audio are transmitted back to IPDM via RTSP or other real-time channels. IPDM, based on knowledge graphs and large models, generates at least several candidate stop points in three steps: candidate point screening, road condition assessment, and feasibility reasoning, and outputs the best stop point with the highest score and recommended guidance route. After receiving the IPDM command, the drone projects a guide arrow above the accident vehicle and broadcasts a voice message, instructing the vehicle to drive to the optimal temporary stopping point, while simultaneously tracking and adjusting the guidance in real time. ATRM receives accident information and automatically matches the nearest available traffic police officer to dispatch a task. The traffic police officer uses a mobile device to navigate to the temporary stop point, completes the accident handling on-site, and uploads the handling results to the system. After the incident is resolved, the system records the results, the drone resumes its daily patrol route, and the system returns to S2 to continue monitoring and patrolling.

2. The control method of a traffic accident handling system coordinated by unmanned aerial vehicles (UAVs) according to claim 1, characterized in that: The specific thresholds for early warning determination in S2 are: an average speed decrease of ≥50% on the road segment within 1 minute or a single vehicle stationary time of ≥30 seconds; and a latitude and longitude positioning error of the early warning location of ≤5 meters. After triggering the warning, the system completes the warning registration within 60 seconds and confirms the arrival of the drone within 5 minutes.

3. The control method for a UAV-assisted traffic accident handling system according to claim 1, characterized in that: In S4, the priority criteria for selecting candidate temporary parking points using the knowledge graph include: shortest distance from the warning location, space capacity, minimal impact on main traffic flow, and compliance with road control and construction restrictions. The large model considers the travel time, vehicle type, damage level, and real-time congestion prediction of candidate points during evaluation and outputs a comprehensive feasibility score.

4. The control method for a UAV-assisted traffic accident handling system according to claim 1, characterized in that: In S6, the dispatch rules of ATRM include: prioritizing the matching of the nearest and available traffic police based on the traffic police's current GPS location, duty status, and historical response time; dispatching also includes preliminary accident information, the coordinates of the optimal temporary stop point, and the estimated arrival time; traffic police provide feedback on the status via mobile terminal, and the system updates the handling progress accordingly and forces the upload of a handling report after the handling is completed to end the emergency process.

5. A traffic accident handling system based on any one of claims 1-4, characterized in that: This includes a vehicle speed detection module (VSDM), which is used to collect real-time data on vehicle speed and vehicle distribution density along the road and send the collected data to the system data bus. The drone patrol module includes at least a high-definition camera, a directional microphone, a ground projection device, a lidar obstacle avoidance sensor, and an RTSP data transmission module. It is used to perform daily patrols, rapid response and on-site guidance, and transmit multimodal data back to the system. The Intelligent Parking Decision Module (IPDM) includes a knowledge graph database and a large model inference unit. The knowledge graph database stores road parking locations, road grades, traffic rules, construction and control information, and static and dynamic road information. The large model inference unit is used to integrate VSDM real-time data, UAV-transmitted images, and the knowledge graph to output the optimal parking location and guidance route. The ATRM (Automatic Traffic Police Response) module is used to receive accident information, temporary stopping points, and guidance progress output by IPDM (Information Technology Management System), and automatically match and dispatch processing instructions based on the real-time location and working conditions of traffic police. The traffic command center monitoring platform and interactive bus are used to realize standardized data access, real-time synchronization, visualization and manual intervention between the modules.

6. A traffic accident handling system in collaboration with unmanned aerial vehicles (UAVs) according to claim 5, characterized in that: The VSDM consists of microwave radar and video speed measurement equipment arranged in groups of 500 meters along the road, with a sampling frequency of 1 time per second.

7. A traffic accident handling system in collaboration with unmanned aerial vehicles (UAVs) according to claim 5, characterized in that: The drone patrol module has a daily patrol altitude of 50 meters, a rapid response confirmation altitude of 20 meters, and a hovering altitude of 15 meters during guidance. The optimal flight speed for the UAV is 120 km / h, and it has the ability to adjust its flight path with "straight line priority + dynamic obstacle avoidance".

8. A traffic accident handling system in collaboration with unmanned aerial vehicles (UAVs) according to claim 5, characterized in that: The large model inference unit in the IPDM adopts a lightweight LLM to achieve edge and near-edge inference, and the decision-making process includes: Based on the knowledge graph, call up available parking points within a preset radius around the warning point and filter the candidate set; Based on VSDM real-time traffic assessment, congestion and travel time of reachable routes for each candidate point are evaluated; The feasibility of parking at candidate points is assessed by combining the vehicle type and damage level data returned by the drone, and the optimal temporary parking point and guidance route with the highest comprehensive score are output.

9. A traffic accident handling system in collaboration with unmanned aerial vehicles (UAVs) according to claim 5, characterized in that: The drone projection device is used to project bright indicator arrows on the ground to guide accident vehicles away from the main lane. The projected arrows are 0.5 meters wide and are accompanied by voice broadcast instructions. The drone maintains closed-loop communication with the IPDM during the guidance process and adjusts the projection direction and broadcast content in real time.

10. A traffic accident handling system in collaboration with unmanned aerial vehicles (UAVs) according to claim 5, characterized in that: The system implements standardized access to multimodal data at the data interface layer, including but not limited to VSDM velocity stream data, UAV RTSP image stream, knowledge graph structured data and ATRM feedback information, and achieves millisecond / second-level data synchronization and command issuance based on this data bus.

Images