Traffic accident emergency processing method and device based on autonomous operation of unmanned aerial vehicle

Abstract

The invention provides a traffic accident emergency processing method and device based on autonomous operation of an unmanned aerial vehicle, and the method comprises the steps: carrying out the real-time collection of an image and / or video of an accident scene, carrying out the recognition of a collection result according to a recognition model, and determining the accident type of the accident scene, the information of the accident scene, and an involved vehicle; according to the accident type, the accident scene information and the involved vehicles, the accident severity, the influence range, the surrounding traffic density and the predicted delay time are determined, and the accident scene is scored through a scoring model; determining a risk level corresponding to the current branch according to the scoring result, and generating a processing strategy according to the risk level and the accident scene information; and sending the processing strategy to the vehicle in the accident scene. According to the invention, a full-process framework of autonomous traffic accident processing of the unmanned aerial vehicle is constructed, and intelligent and efficient accident site vehicle management is realized through a closed-loop design of accident identification, graded evaluation and differentiated processing.

Description

Technical Field

[0001] This disclosure relates to the fields of intelligent transportation systems and autonomous control technology for unmanned aerial vehicles (UAVs), and in particular to a method, device, storage medium, and UAV for emergency handling of traffic accidents based on autonomous operation of UAVs. Background Technology

[0002] With the rapid development of Intelligent Transportation Systems (ITS) and autonomous driving technologies, more and more ground vehicles are equipped with the ability to automatically perceive, plan routes, and control themselves. However, in complex urban traffic environments, emergencies such as traffic accidents, traffic congestion, temporary traffic control, and natural disasters still occur frequently. These situations often exceed the local perception and decision-making capabilities of ground vehicles, resulting in weak response capabilities of autonomous driving systems in abnormal situations, or even paralysis or misjudgment, affecting traffic safety and traffic efficiency.

[0003] In recent years, some research and industrial practices have explored the use of unmanned aerial vehicles (UAVs) to assist in intelligent traffic management. For example, UAVs are used for functions such as high-altitude road monitoring, traffic flow statistics, and accident image transmission, enhancing traffic control centers' real-time understanding of road conditions. In some projects, UAVs participate in tasks such as map updates and traffic guidance by sharing information with ground vehicles. Other research is exploring the use of UAVs for remote monitoring and route planning of ground vehicle fleets within closed industrial parks.

[0004] However, most existing solutions rely on manual dispatch centers or preset control strategies, placing low demands on the autonomy of drones. They are typically used as auxiliary sensing devices and lack the ability to proactively intervene and dynamically take over ground vehicles during traffic emergencies. Therefore, to address the problem of insufficient response from ground-based autonomous driving systems in traffic emergencies, there is an urgent need for a drone platform with high autonomy and a closed-loop perception-decision-control capability. This platform should be able to quickly enter abnormal areas and independently assess the on-site environment, identify target vehicles, and take over control without human intervention, thereby achieving a more efficient and safer intelligent traffic emergency response. Summary of the Invention

[0005] The purpose of this disclosure is to provide a method, device, storage medium, and drone for handling traffic accidents based on the autonomous operation of drones, so as to solve the problems existing in the prior art.

[0006] The embodiments of this disclosure adopt the following technical solution: a traffic accident emergency handling method based on autonomous operation of unmanned aerial vehicles, comprising: real-time acquisition of images and / or videos of the accident scene, and identification of the acquisition results according to a recognition model to determine the accident type, accident scene information, and vehicles involved; determining the accident severity, impact range, surrounding traffic density, and estimated delay time according to the accident type, the accident scene information, and the vehicles involved, and scoring the accident scene using a scoring model; determining the risk level corresponding to the current branch according to the scoring results, and generating a processing strategy according to the risk level and the accident scene information; and sending the processing strategy to the vehicles at the accident scene.

[0007] This disclosure also provides a traffic accident emergency response device based on autonomous operation of unmanned aerial vehicles (UAVs), comprising: a data acquisition and identification module for real-time acquisition of images and / or videos of the accident scene, and identification of the acquisition results according to an identification model to determine the accident type, accident scene information, and involved vehicles; an on-site scoring module for determining the severity of the accident, the scope of impact, the surrounding traffic density, and the estimated delay time based on the accident type, the accident scene information, and the involved vehicles, and scoring the accident scene using a scoring model; a strategy generation module for determining the risk level corresponding to the current branch based on the scoring results, and generating a processing strategy based on the risk level and the accident scene information; and a communication control module for sending the processing strategy to the vehicles at the accident scene.

[0008] This disclosure also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described traffic accident emergency response method based on autonomous operation of unmanned aerial vehicles.

[0009] This disclosure also provides a drone, which includes at least an image acquisition device, a memory, and a processor. The memory stores a computer program, and the processor executes the computer program in the memory to implement the steps of the above-described traffic accident emergency response method based on the autonomous operation of the drone.

[0010] The beneficial effects of this disclosed embodiment are as follows: It constructs a full-process framework for autonomous handling of traffic accidents by unmanned aerial vehicles (UAVs). Through a closed-loop design of accident identification, hierarchical assessment, and differentiated handling, it achieves intelligent and efficient vehicle management at accident scenes. This avoids the problems of long response chains, poor timeliness, and low data utilization that occur in existing technologies where UAVs typically only undertake image acquisition and transmission tasks, relying on the command center for subsequent judgment and instruction issuance. This embodiment, from accurate accident information collection to the execution of a scoring-based hierarchical strategy, ensures lightweight processing in low-risk scenarios while enabling deep intervention in high-risk scenarios, balancing processing efficiency and rational resource allocation. Furthermore, by addressing the differentiated instruction systems for intelligent and ordinary vehicles, it solves the problem of collaborative scheduling in heterogeneous traffic environments, effectively shortening the impact time of accidents, reducing traffic congestion and the risk of secondary accidents, and providing a systematic solution for dynamic traffic emergency management. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of the traffic accident emergency response method based on the autonomous operation of an unmanned aerial vehicle (UAV) in the first embodiment of this disclosure; Figure 2 This is a schematic diagram of the structure of the traffic accident emergency response device based on the autonomous operation of an unmanned aerial vehicle (UAV) in the second embodiment of this disclosure. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0014] With the rapid development of intelligent transportation systems and autonomous driving technology, more and more ground vehicles are equipped with the ability to automatically perceive, plan routes, and control autonomously. However, in complex urban traffic environments, emergencies such as traffic accidents, traffic congestion, temporary traffic control, and natural disasters still occur frequently. These situations often exceed the local perception and decision-making range of ground vehicles, resulting in weak response capabilities of autonomous driving systems in abnormal situations, or even paralysis or misjudgment, affecting traffic safety and traffic efficiency.

[0015] In recent years, some research and industrial practices have explored the use of unmanned aerial vehicles (UAVs) to assist in intelligent traffic management. For example, UAVs are used for functions such as high-altitude road monitoring, traffic flow statistics, and accident image transmission, enhancing traffic control centers' real-time understanding of road conditions. In some projects, UAVs participate in tasks such as map updates and traffic guidance by sharing information with ground vehicles. Other research is exploring the use of UAVs for remote monitoring and route planning of ground vehicle fleets within closed industrial parks.

[0016] However, most existing solutions rely on manual dispatch centers or preset control strategies, placing low demands on the autonomy of drones. They are typically used as auxiliary sensing devices and lack the ability to proactively intervene and dynamically take over ground vehicles during traffic emergencies. Therefore, to address the problem of insufficient response from ground-based autonomous driving systems in traffic emergencies, there is an urgent need for a drone platform with high autonomy and a closed-loop perception-decision-control capability. This platform should be able to quickly enter abnormal areas and independently assess the on-site environment, identify target vehicles, and take over control without human intervention, thereby achieving a more efficient and safer intelligent traffic emergency response.

[0017] To address the aforementioned technical problems, the first embodiment of this disclosure provides a traffic accident emergency response method based on the autonomous operation of unmanned aerial vehicles (UAVs). This method is mainly applied to UAVs under the jurisdiction of traffic management departments to assist in emergency response at traffic accident scenes in accordance with traffic management needs. Figure 1 A flowchart of the traffic accident emergency handling method of this embodiment is shown, specifically including steps S10 to S40: S10, collects images and / or videos of the accident scene in real time, and identifies the collected results according to the recognition model to determine the accident type, accident scene information and vehicles involved.

[0018] During its patrol, upon receiving signals of emergencies detected by the city's traffic sensing system or by the drone itself, the drone can autonomously plan the optimal flight path to the accident or congestion area. During flight, the drone uses onboard vision modules, infrared sensors, radar, and other multimodal sensors to perceive ground traffic conditions and performs positioning based on high-precision maps.

[0019] Upon arrival at the accident scene, the drone can use its onboard camera to collect images and / or videos of the scene in real time, obtaining specific information about the accident and analyzing the accident type, scene details, and vehicles involved. Specifically, the drone can run the YOLO V8 traffic incident recognition model, which has been trained on a large amount of data for various abnormal traffic scenarios (such as collisions, wrong-way driving, congestion, and abnormal parking). This model can classify and detect risk events in the scene at a speed of seconds and identify accident scene information and the specific vehicles involved.

[0020] In practical implementation, accident scene information includes at least: the total number of vehicles at the accident scene, the orientation of each vehicle, the extent of damage, its current location, vehicle type, number of passengers, lane occupancy, vehicle speed, and road topology information. The road topology information includes at least: road type, number of lanes, lane width, lane curvature, and median type. Vehicle types include intelligent vehicles and non-intelligent vehicles. In some embodiments, the road topology information may further include: intersection morphology (crossroads, T-junctions, roundabouts, etc.), restricted areas, emergency lanes, non-motorized vehicle lane separation, road signs, markings, guardrails, etc. The accident scene information may also include: the relative positions and density of pedestrians, bicycles, etc.

[0021] S20 determines the severity of the accident, its scope of impact, surrounding traffic density, and estimated delay time based on the accident type, accident scene information, and the vehicles involved, and scores the accident scene using a scoring model.

[0022] By collecting on-site information and analyzing the output of the identification model, the drone can determine the basic situation of the accident scene. Subsequently, the drone will score the degree of danger of the current accident scene based on the identified basic information. In this embodiment, the drone runs a scoring model for evaluating the degree of danger of the accident scene. This scoring model is constructed based on multi-dimensional indicators such as the severity of the traffic incident, the scope of impact, the number of vehicles involved, the surrounding traffic density, and the estimated delay time, assigning a quantified risk value to the current accident scene to characterize its degree of danger.

[0023] In practical implementation, the first step is to determine the five key indicators closely related to traffic risk required for the scoring model input, based on the accident type, accident scene information, and the vehicles involved. Specifically, the accident severity mainly measures the destructiveness and urgency of the accident itself; the impact range mainly reflects the spatial area where the accident interferes with the surrounding traffic flow, usually centered on the accident point and defined in conjunction with lane occupancy and vehicle parking status; the number of vehicles involved mainly refers to the total number of vehicles at the accident scene, including vehicles directly involved in the accident (involved vehicles) and vehicles trapped within the impact range (congested vehicles); the surrounding traffic density represents the number of vehicles per unit road area (including the impact range and surrounding areas), reflecting the saturation level of traffic flow; and the estimated delay time is the total duration for which vehicles cannot pass normally due to the accident, calculated comprehensively based on traffic density, handling difficulty, and other factors.

[0024] In this embodiment, the severity of an accident is primarily determined based on the accident type. Different numerical values ​​are assigned to different accident types to indicate the severity of that type of accident. For example, when the accident type is a collision, the severity score is 1 point; when the accident is a wrong-way driving incident, the severity score is 0.7 points; when the accident is a traffic jam incident, the severity score is 0.5 points; and when the accident is an illegal parking incident, the severity score is 0.3 points.

[0025] The extent of the impact is primarily determined based on the lane occupancy, orientation, and road topology information of the vehicles involved. For example, all vehicles at the accident scene can be considered as a rectangle (length = ...). Width = Center point location The road centerline is used as the x-axis (with the positive direction along the road), and the perpendicular direction to the road is the y-axis. For each vehicle, its axis-aligned rectangle is projected onto the road axis (x) and the lateral (y) directions to obtain the total longitudinal span and the total lateral span; the half-axes a and b are equal to half the span, and a safety buffer is added. Finally, appropriate adjustments are made according to the road complexity or the median amplification factor.

[0026] For each vehicle involved Road projection length to the x-axis (halfway):

[0027] To the y-axis (half):

[0028] in, The vehicle length is in meters (m). The width of the vehicle is in meters (m). The vehicle's heading angle (the angle between the vehicle and the road centerline, in rad).

[0029] Longitudinal (along the road) minimum / maximum boundaries:

[0030]

[0031] Horizontal minimum / maximum boundaries:

[0032]

[0033] in, .

[0034] Total longitudinal span and total transverse span:

[0035]

[0036] Half shaft:

[0037]

[0038] in, and For safety buffer (m), the amplified semi-axis is used to cover sensing error / movement diffusion, and a and b are the semi-major axis (m) and semi-minor axis (m) of the ellipse, respectively.

[0039] Adjustments should be made based on road complexity and median strip:

[0040]

[0041] in , C is a small amplification factor, C is the road complexity factor (dimensionless), and I is the median strip factor (e.g., no median = 1.0, with median = 1.2).

[0042] Area affected:

[0043] in, Area affected by the accident .

[0044] Use the Sigmoid function to Mapped to normalized score :

[0045] Where the threshold and slope Fitting from historical labeled data, (e.g.) Shi De ).

[0046] The surrounding traffic density is primarily determined based on the total number of vehicles, vehicle speed, road type, and the area of ​​influence. In some embodiments, the surrounding traffic density can be calculated using the following formula:

[0047] in, Indicates the surrounding traffic density , This represents the total number of vehicles in the current area. This represents the number of vehicles that merge into the affected area within the sampling interval. To determine the average speed of vehicles within the affected area, This is the default free-flow speed for the current road type. Finally, the surrounding traffic density is evaluated using the pre-defined scoring rules of the piecewise linear function. The score is determined by the value of the score. To facilitate threshold setting, each score is used as the threshold value. Number of vehicles As a metric, the normalized density is defined. , The reference density threshold can be set according to the road type; for example, for urban secondary roads, it can be set to [value missing]. Urban main / secondary roads can be taken Highways are acceptable Based on the normalized density, the surrounding traffic density score... .

[0048] Subsequently, based on the lane occupancy of the vehicles involved, the number of lanes, the severity of the accident, the scope of impact, and the surrounding traffic density, the estimated delay time and its score are determined. Specifically, this embodiment will determine the estimated delay time... Classified as having incurred delays and potential delays ,Right now .

[0049] Define remaining capacity:

[0050] in, The total number of lanes The number of lanes already occupied is 's', and 's' represents the single-lane capacity (vehicles / s). This is the efficiency reduction factor (accident disturbance). Prevent division by zero.

[0051] Delays have occurred :

[0052] in, This is a dimensionless calibration coefficient used to correct biases introduced during modeling.

[0053] Potential delays :

[0054] in, The arrival time of the rescue team (s). The distance to the nearest traffic police / rescue point (in meters) can be obtained via GIS. For rescue speed (m / s), The dimensionless calibration coefficient is used for congestion correction.

[0055] The delay score is obtained using the Sigmoid continuous mapping. :

[0056]

[0057] in, The baseline duration for normalizing delay time is set based on different scenarios and experiences (e.g., urban main roads). Setting it to 1800s means that 30 minutes corresponds to approximately 0.5 minutes. Control the steepness of the curve (e.g., 900s).

[0058] After determining the scores for the five key indicators mentioned above, they are input into the scoring model, which can be implemented using a linear weighted model, for example:

[0059] in, The current accident scene is rated. The scores indicate the severity. The score indicates the scope of influence. For the normalized score involving the number of vehicles N (e.g. , (The reference number of vehicles for normalization is set according to historical distribution). The score represents the surrounding traffic density. The score is the estimated delay time. to The weights of the corresponding indicators, And satisfy .

[0060] The weight values ​​in this embodiment can be determined using empirical methods, specifically by combining the opinions of traffic management experts to provide heuristic weights, such as major accidents being more important than congestion; or by using data-driven methods, using historical labeled data, and learning the optimal weights through regression models, information gain, or AHP (analytic hierarchy process).

[0061] S30: Based on the scoring results, determine the risk level corresponding to the current branch, and generate a handling strategy based on the risk level and accident scene information.

[0062] After determining the score of the current accident scene, the risk level of the current accident can be evaluated according to the pre-set risk level evaluation criteria. Based on the risk level results, corresponding handling strategies for the current accident scene can be generated. Specifically, the risk level evaluation criteria can be used to assess the risk level through thresholds. If the score is less than the first threshold, the risk level is determined to be Level 1, and the corresponding handling strategy is to generate regional broadcast information to prompt all vehicles at the accident scene to leave the current area as soon as possible. If the score is greater than or equal to the first threshold and less than the second threshold, the risk level is determined to be Level 2. All vehicles at the accident scene are classified, and alarm prompts are generated for vehicles of different levels according to the classification results to handle the accident scene. If the score is greater than or equal to the second threshold, the risk level is determined to be Level 3. At this time, control commands are generated for each vehicle to achieve refined vehicle management and quickly clear the accident scene.

[0063] It should be noted that the specific values ​​of the first threshold and the second threshold can be determined according to actual needs. For example, the first threshold is 0.3 and the second threshold is 0.6. This embodiment does not impose specific numerical restrictions. If the scores are calculated according to a percentage system, the values ​​of the first threshold and the second threshold can be converted accordingly.

[0064] Corresponding to accident scenes with different risk levels, this embodiment generates different processing strategies, ensuring lightweight processing for low-risk scenarios while enabling deep intervention in high-risk scenarios, thus balancing processing efficiency and rational resource allocation. In some embodiments, for accident scenes with Level 1 risk, due to their low risk level and minimal impact on traffic, drones can generate regional broadcast information and use a voice broadcast system to inform vehicles at the accident scene to leave as soon as possible, without requiring special control measures.

[0065] For accidents of level two risk, which have a certain impact on traffic conditions, the management objective of the drone should be to quickly remove vehicles unrelated to the accident and instruct vehicles involved that may affect the traffic of other vehicles to move to a safe area. In some embodiments, vehicles at the accident scene can be categorized into movable vehicles, immovable vehicles, and other vehicles based on their current location, degree of damage, and lane occupancy. Subsequently, alarm prompts are generated for each type of vehicle. The main purpose of the alarm prompts is to provide suggestive guidance to the corresponding vehicles, using simple, one-step instructions to enable the vehicles to complete the action. For example, for movable vehicles involved in the accident, specific location movement prompts and personnel evacuation prompts are generated, guiding them to move to areas that do not affect main traffic, such as emergency lanes or shoulders, and providing assistance in prompting them to engage the handbrake, turn off the engine, and evacuate to the outside of the guardrail after the vehicle is moved to the specific area. For immovable vehicles involved in the accident, the main focus is on generating personnel evacuation prompts to guide the occupants of the vehicle to transfer to prevent secondary injuries. For other vehicles unrelated to the accident but only affected by it, guidance prompts such as specific location movement prompts, speed limit prompts, and sequence prompts are provided, for example, instructing them to leave the accident scene through a specific lane, limiting their movement speed within the accident scene, and specifying the order of passage between different vehicles to avoid secondary congestion.

[0066] In some special scenarios, the alarm information generated by drones can also include environmental prompts, such as special prompts for rainy or snowy weather, or nighttime conditions.

[0067] For accident scenes corresponding to Level 3 risk, drones are used for on-site takeover to handle emergencies. In this case, the drones generate control commands for each vehicle at the accident scene, achieving unified vehicle control from a global perspective, enabling real-time aerial decision-making and on-site control, significantly improving emergency response efficiency. Specifically, the drones first prioritize vehicles at the accident scene based on their current location, damage level, lane occupancy, vehicle type, and number of passengers. Then, according to the priority order, combined with path planning algorithms, control commands are generated for each vehicle to guide it to leave the accident scene sequentially.

[0068] In some embodiments, control commands may be represented in the following format:

[0069] in, Indicates acceleration. Indicates the direction of travel. Indicates driving speed. Indicates the target point. This indicates the vehicle's driving priority. In this embodiment, the control commands can be output based on a command generation model. This model is trained using a neural network. Specifically, simulation platforms such as CARLA and LGSVL can be used to generate large-scale traffic accident scenarios. Path planning algorithms such as MPC and A* are used to generate control commands as supervisory labels for training. The final model takes vehicle priority, current location, damage level, lane occupancy, vehicle type, and passenger capacity as inputs and control commands as outputs.

[0070] S40 sends handling strategies to vehicles at the accident scene.

[0071] Once the drone generates a processing strategy, it can send the strategy to vehicles at the accident scene to achieve emergency management. Depending on the different levels of the accident and the corresponding strategies, the drone will use different methods to send the processing strategy. Specifically, for level-one risk accidents, the drone can distribute the processing strategy through regional broadcasts, using voice instructions and laser guidance. Vehicles within the scene can then autonomously leave the accident area based on the voice or laser guidance.

[0072] For accident scenarios with level 2 risk, the drone generates different types of handling strategies for vehicles of different levels and instructs the corresponding vehicles to execute the handling strategies through voice instructions and laser guidance. It should be noted that when the drone is actually classifying, it can record the license plate numbers of vehicles of different levels as the grouping results. When issuing the handling strategy, it can move to the vicinity of the corresponding vehicle in the low-altitude area in combination with the license plate number and vehicle location to instruct the vehicle to execute the handling strategy.

[0073] If the accident scenario is Level 3, independent control commands need to be generated and issued for each vehicle. In this case, the vehicle type can be considered, and the handling can be differentiated between intelligent and non-intelligent vehicles. In some embodiments, for non-intelligent vehicles, the handling strategy for Level 2 risk scenarios can be referenced. Control commands can be displayed to the driver of the non-intelligent vehicle using voice prompts and / or laser projection to guide the driver to move to the corresponding location and restrict its movement path and speed, thereby achieving refined and accurate vehicle control. For intelligent vehicles, drones can directly take over the control authority of the intelligent vehicle and issue control commands directly to the vehicle's control system, so that the intelligent vehicle can directly execute the control commands.

[0074] It's important to note that drone takeover requires the vehicle owner's consent. Takeover can only proceed after the owner's confirmation. If the owner refuses, the drone can treat the vehicle as a "dynamic obstacle" and optimize its handling strategies for other vehicles it might affect, thus avoiding the obstacle. Furthermore, before taking over the intelligent vehicle, the drone needs to obtain the vehicle's current status via V2X (Vehicle-to-Everything) communication, including but not limited to remaining battery power, intelligent driving system protocols, and intelligent driving system self-check results. This determines whether the vehicle can perform actions such as starting, stopping, changing lanes, and obstacle avoidance under drone control. If the vehicle's current battery power or system self-check results indicate insufficient power to complete control commands, the drone will also mark it as a "dynamic obstacle." If the vehicle's intelligent driving system protocol does not support remote drone control, voice or laser guidance can be used to direct the owner away from the accident scene. It should be understood that the above-mentioned method of issuing processing strategies by drones is only a possible preferred example in actual implementation. Depending on the accessories carried by the drone, the accident scene environment, vehicle conditions, etc., the drone may also use other methods such as vehicle broadcasting and screen display to issue processing strategies. Any method that can complete on-site vehicle control can be used as a specific implementation method of this embodiment, and this embodiment does not impose any specific restrictions.

[0075] This embodiment constructs a full-process framework for autonomous handling of traffic accidents by unmanned aerial vehicles (UAVs). Through a closed-loop design of accident identification, hierarchical assessment, and differentiated handling, it achieves intelligent and efficient vehicle management at accident scenes. This avoids the problems of long response chains, poor timeliness, and low data utilization that occur in existing technologies where UAVs typically only undertake image acquisition and transmission tasks, relying on the command center for subsequent judgment and instruction issuance. From accurate accident information collection to the execution of scoring-based hierarchical strategies, this embodiment ensures lightweight processing in low-risk scenarios while enabling deep intervention in high-risk scenarios, balancing processing efficiency and rational resource allocation. Furthermore, by addressing the differentiated instruction systems for intelligent and ordinary vehicles, it solves the collaborative scheduling challenges in heterogeneous traffic environments, effectively shortening accident impact time, reducing traffic congestion and the risk of secondary accidents, and providing a systematic solution for dynamic traffic emergency management.

[0076] In some embodiments, when a drone is performing on-site emergency response, if an abnormal state occurs such as a short-term bottleneck or overload of the onboard computer's computing resources, or if it enters a frequency reduction mode due to excessive temperature, the drone can send the collected image or video data to the city edge server via the wireless image transmission module. The city edge server will then temporarily take over some of the drone's computing tasks and send the processed vehicle control commands, which have a smaller data volume, to the drone, thereby ensuring the continuity and stability of the emergency response task.

[0077] Furthermore, during emergency response, drones can simultaneously upload the collected on-site information and generated processing strategies to the city's traffic perception system in real time. This allows for overall control by the system, and the drones can pause or resume the processing strategy generation or distribution process at any time based on operational instructions from the system. At the same time, drones can communicate with medical and fire departments in real time. Upon discovering injured or trapped individuals at an accident scene, they can promptly issue alarms and upload images, assisting medical and fire departments in reaching the scene as quickly as possible to complete rescue operations.

[0078] Based on the same inventive concept, the second embodiment of this disclosure provides a traffic accident emergency response device based on the autonomous operation of a drone. This device can be applied to drones to assist in emergency response at traffic accident scenes in accordance with traffic management requirements. Figure 2 The diagram illustrates the structure of the traffic accident emergency response device in this embodiment, which mainly includes: a data acquisition and identification module 10, used to acquire images and / or videos of the accident scene in real time, and identify the acquisition results according to an identification model to determine the accident type, accident scene information, and vehicles involved; an on-site scoring module 20, used to determine the severity of the accident, the scope of impact, the surrounding traffic density, and the estimated delay time based on the accident type, accident scene information, and vehicles involved, and to score the accident scene using a scoring model; a strategy generation module 30, used to determine the risk level corresponding to the current branch based on the scoring results, and to generate a processing strategy based on the risk level and accident scene information; and a communication control module 40, used to send the processing strategy to vehicles at the accident scene.

[0079] Specifically, the recognition model is trained based on the YOLO V8 model; the accident scene information includes at least the following: the total number of vehicles at the accident scene, the orientation of each vehicle, the degree of damage, the current location, the vehicle type, the number of passengers, the lane occupancy status, the vehicle speed, and the road topology information; among which, the road topology information includes at least the following: road type, number of lanes, lane width, lane curvature, and median strip type; vehicle types include: intelligent vehicles and non-intelligent vehicles.

[0080] In some embodiments, the on-site scoring module 20 is specifically used to: determine the score of accident severity based on the accident type; determine the scope of impact and its score based on the lane occupancy, orientation, and road topology information of the vehicles involved; determine the surrounding traffic density and its score based on the total number of vehicles, vehicle speed, road type, and scope of impact; determine the estimated delay time and its score based on the lane occupancy, number of lanes, accident severity, scope of impact, and surrounding traffic density of the vehicles involved; and use the score of accident severity, the score of scope of impact, the score of surrounding traffic density, and the score of estimated delay time as input to the scoring model, and output the score of the accident scene.

[0081] In some embodiments, the strategy generation module 30 is specifically used to: determine the risk level as Level 1 when the score is less than a first threshold, and generate regional broadcast information to prompt all vehicles at the accident scene to leave the current area as soon as possible; determine the risk level as Level 2 when the score is greater than or equal to the first threshold and less than the second threshold, classify all vehicles at the accident scene, and generate alarm prompt information for vehicles of different levels according to the classification results; and determine the risk level as Level 3 when the score is greater than or equal to the second threshold, and generate control instructions for each vehicle.

[0082] In some embodiments, the strategy generation module 30 is further configured to: classify vehicles at the accident scene into movable vehicles, immovable vehicles, and other vehicles based on their current location, degree of damage, and lane occupancy; generate specific location movement prompts and personnel evacuation prompts for movable vehicles; generate personnel evacuation prompts for immovable vehicles; and generate specific location movement prompts, speed limit prompts, and sequence prompts for other vehicles.

[0083] In some embodiments, the strategy generation module 30 is further configured to: prioritize vehicles at the accident scene based on their current location, degree of damage, lane occupancy, vehicle type, and number of passengers; and generate control commands for each vehicle according to the priority order, in conjunction with a path planning algorithm, to control the current vehicles to leave the accident scene in sequence.

[0084] In some embodiments, the communication control module 40 is specifically used to: when the vehicle type is an intelligent vehicle, the drone sends control commands to the control system of the intelligent vehicle so that the intelligent vehicle executes the control commands; when the vehicle type is a non-intelligent vehicle, the drone displays the control commands to the driver of the non-intelligent vehicle through laser projection and voice prompts.

[0085] This embodiment constructs a full-process framework for autonomous handling of traffic accidents by unmanned aerial vehicles (UAVs). Through a closed-loop design of accident identification, hierarchical assessment, and differentiated handling, it achieves intelligent and efficient vehicle management at accident scenes. This avoids the problems of long response chains, poor timeliness, and low data utilization that occur in existing technologies where UAVs typically only undertake image acquisition and transmission tasks, relying on the command center for subsequent judgment and instruction issuance. From accurate accident information collection to the execution of scoring-based hierarchical strategies, this embodiment ensures lightweight processing in low-risk scenarios while enabling deep intervention in high-risk scenarios, balancing processing efficiency and rational resource allocation. Furthermore, by addressing the differentiated instruction systems for intelligent and ordinary vehicles, it solves the collaborative scheduling challenges in heterogeneous traffic environments, effectively shortening accident impact time, reducing traffic congestion and the risk of secondary accidents, and providing a systematic solution for dynamic traffic emergency management.

[0086] Based on the same inventive concept, the third embodiment of this disclosure provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the traffic accident emergency response method based on the autonomous operation of an unmanned aerial vehicle described in the first embodiment of this disclosure.

[0087] Based on the same inventive concept, the fourth embodiment of this disclosure provides a drone, which includes at least an image acquisition device, a memory, and a processor. The memory stores a computer program. The processor, when executing the computer program in the memory, implements the steps of the traffic accident emergency handling method based on the autonomous operation of a drone described in the first embodiment of this disclosure.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this disclosure.

Claims

1. A method for handling traffic accidents based on autonomous operation of unmanned aerial vehicles (UAVs), characterized in that, include: Real-time acquisition of images and / or videos of the accident scene, and identification of the acquisition results based on the recognition model to determine the accident type, accident scene information, and vehicles involved in the accident scene; Based on the accident type, the accident scene information, and the vehicles involved, the severity of the accident, the scope of impact, the surrounding traffic density, and the estimated delay time are determined, and the accident scene is scored using a scoring model. Based on the scoring results, determine the risk level corresponding to the current branch, and generate a processing strategy based on the risk level and the accident scene information; The processing strategy is sent to the vehicles at the accident scene.

2. The traffic accident emergency handling method according to claim 1, characterized in that, The recognition model is trained based on the YOLO V8 model; The accident scene information includes at least: the total number of vehicles at the accident scene, the orientation of each vehicle, the extent of damage, the current location, vehicle type, number of passengers, lane occupancy, vehicle speed, and road topology information; The road topology information includes at least: road type, number of lanes, lane width, lane curvature, and median strip type; the vehicle type includes: intelligent vehicles and non-intelligent vehicles.

3. The traffic accident emergency handling method according to claim 2, characterized in that, The process involves determining the severity, impact range, surrounding traffic density, and estimated delay time of the accident based on the accident type and the accident scene information, and then scoring the accident scene using a scoring model, including: The severity score of the accident is determined based on the type of accident. Based on the lane occupancy, orientation, and road topology information of the vehicles involved, the scope of influence and its score are determined. The surrounding traffic density and its score are determined based on the total number of vehicles, the vehicle speed, the road type, and the area of ​​influence. The estimated delay time and its score are determined based on the lane occupancy status of the vehicles involved, the number of lanes, the severity of the accident, the scope of impact, and the surrounding traffic density. The scoring model takes the scores for the severity of the accident, the scope of impact, the surrounding traffic density, and the estimated delay time as inputs, and outputs a score for the accident scene.

4. The traffic accident emergency handling method according to claim 3, characterized in that, The step of determining the risk level corresponding to the current branch based on the scoring results, and generating a processing strategy based on the risk level and the accident scene information, includes: If the score is less than the first threshold, the risk level is determined to be Level 1, and a regional broadcast message is generated to remind all vehicles at the accident scene to leave the current area as soon as possible. If the score is greater than or equal to the first threshold and less than the second threshold, the risk level is determined to be level two, and all vehicles at the accident scene are classified. Based on the classification results, alarm prompts are generated for vehicles of different levels. If the score is greater than or equal to the second threshold, the risk level is determined to be level three, and control instructions are generated for each vehicle.

5. The traffic accident emergency handling method according to claim 4, characterized in that, The process involves classifying all vehicles at the accident scene and generating alarm notifications for vehicles of different classes based on the classification results, including: Based on the current location, the extent of damage, and the lane occupancy, the vehicles at the accident scene are categorized into movable vehicles involved in the accident, immovable vehicles involved in the accident, and other vehicles. For the movable vehicle involved in the incident, generate a specific location movement prompt and a personnel evacuation prompt; For the immovable vehicles involved in the incident, generate an evacuation prompt; For the other vehicles, generate specific location movement prompts, speed limit prompts, and sequence prompts.

6. The traffic accident emergency handling method according to claim 4, characterized in that, The generation of control commands for each vehicle includes: Based on the current location, the extent of damage, the lane occupancy status, the vehicle type, and the number of passengers, the vehicles at the accident scene are prioritized for control. According to the order of the control priorities, combined with the path planning algorithm, control commands are generated for each vehicle to control the current vehicle to leave the accident scene in sequence.

7. The traffic accident emergency handling method according to claim 6, characterized in that, Sending the processing strategy to the vehicles at the accident scene includes: When the vehicle type is an intelligent vehicle, the drone sends the control commands to the intelligent vehicle's control system so that the intelligent vehicle executes the control commands; In the case of a non-intelligent vehicle, the drone displays control commands to the driver of the non-intelligent vehicle through laser projection and voice prompts.

8. A traffic accident emergency response device based on autonomous operation of unmanned aerial vehicles (UAVs), characterized in that, include: The data acquisition and recognition module is used to acquire images and / or videos of the accident scene in real time, and to identify the acquisition results according to the recognition model to determine the accident type, accident scene information, and vehicles involved in the accident scene. The on-site scoring module is used to determine the severity of the accident, the scope of impact, the surrounding traffic density, and the estimated delay time based on the accident type, the accident scene information, and the vehicles involved, and to score the accident scene using a scoring model. The strategy generation module is used to determine the risk level corresponding to the current branch based on the scoring results, and to generate a processing strategy based on the risk level and the accident scene information. The communication control module is used to send the processing strategy to the vehicles at the accident scene.

9. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the traffic accident emergency response method based on the autonomous operation of unmanned aerial vehicles as described in any one of claims 1 to 7.

10. A drone, comprising at least an image acquisition device, a memory, and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program on the memory, it implements the steps of the traffic accident emergency response method based on the autonomous operation of an unmanned aerial vehicle as described in any one of claims 1 to 7.

Images