Intelligent traffic accident restoration system and method based on unmanned aerial vehicle nest

Abstract

The invention relates to the technical field of unmanned aerial vehicle nests, and particularly discloses a traffic accident intelligent restoration system and method based on an unmanned aerial vehicle nest, and the system is provided with a response priority division module, a selected unmanned aerial vehicle nest matching module, an optimal route planning module, an accident vehicle intelligent search module and an accident scene restoration map generation module. The problems that a traditional processing mode needs long road closing time and affects traffic capacity, vehicles, personnel, scattered objects and the like exist in accident sites and are prone to being damaged, and the traditional mode mostly aims at local two-dimensional scenes, accident information is difficult to collect in an omnibearing mode, accurate identification and responsibility division requirements cannot be met, traffic jam is prone to being caused, and the like are solved. And the secondary accident risk is also increased.

Description

Technical Field

[0001] The present invention relates to the technical field of drone nests, and in particular to an intelligent traffic accident restoration system and method based on drone nests. Background Art

[0002] Drones are maneuverable, have excellent viewing angles, and can quickly reach the scene. They can capture images of accident scenes from various heights and angles. When paired with high-definition cameras, quad-optical cameras (visible light, infrared, thermal imaging, and lidar), and other equipment, they can obtain more comprehensive and accurate accident information, addressing the shortcomings of traditional methods. Drone nests provide storage, charging, and automated management capabilities for drones. This allows drones to quickly and autonomously take off and land after an accident, rushing to the scene, overcoming the limitations of manual deployment and improving response speed, ease of use, and sustainability. As smart city development advances, traffic management is transitioning toward intelligent and digital capabilities. Intelligent accident reconstruction methods based on drone nests align with the trend toward intelligent traffic management and can be integrated with other intelligent transportation systems to enhance the overall efficiency and service level of traffic management.

[0003] Currently, research on intelligent traffic accident reconstruction still has some shortcomings. Traditional methods, such as on-site measurements and photography by traffic police, require lengthy road closures, impacting traffic flow. Accident scenes are susceptible to damage due to the presence of vehicles, personnel, and scattered objects. Furthermore, traditional methods primarily focus on a localized, two-dimensional scene, making it difficult to comprehensively capture accident information and meet the requirements for accurate identification and accountability. Accident handling has traditionally been time-consuming and has increased the risk of secondary accidents. Swift accident resolution and restoration of traffic order are urgently needed. Summary of the Invention

[0004] In response to the deficiencies in the prior art, the present invention provides an intelligent traffic accident restoration system and method based on a drone nest, which can effectively solve the problems involved in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: The first aspect of the present invention provides an intelligent traffic accident restoration system based on drone nests, including: a response priority division module, a data center receives traffic accident alarms, extracts basic information of the accident alarms, and automatically divides response priorities; a selected drone nest matching module obtains basic information of drone nests near the traffic accident, matches the selected drone nest in combination with the response priority, sends a task instruction to the selected drone nest, and starts the drone's response program; an optimal route planning module, after the selected drone nest receives the task instruction, obtains the drone's flight feature information, plans the drone's optimal flight route to arrive at the traffic accident scene; an accident vehicle intelligent search module, the drone starts an intelligent search mode based on visual-thermodynamic joint positioning and autonomous game decision-making, finds the accident vehicle, and the drone takes a bird's-eye view photo at the traffic accident scene, and the photo is transmitted to the data center; an accident scene restoration map generation module, based on the photos transmitted to the data center, extracts traffic accident scene feature data, and automatically generates a traffic accident scene restoration map by combining benchmark points and baselines through an artificial intelligence algorithm.

[0006] As a further solution, the data center receives traffic accident alarms, extracts basic information about the accident alarms, and automatically divides response priorities. The specific analysis process is as follows: the data center receives traffic accident alarms, extracts basic information about the accident alarms, including the total number of vehicles involved in the accident Cn, the number of people involved in the accident Pn, and the number of public facilities damaged by the accident Gn; based on the basic information about the accident alarm, a comprehensive analysis is performed to obtain the characteristic value Si of the basic information about the accident alarm; the characteristic value of the basic information about the accident alarm is compared with the characteristic threshold of the basic information about the accident alarm stored in the database; if the characteristic value of the basic information about the accident alarm is lower than the characteristic threshold of the basic information about the accident alarm, the response priority is level two; if the characteristic value of the basic information about the accident alarm is not lower than the characteristic threshold of the basic information about the accident alarm, the response priority is level one.

[0007] As a further solution, the basic information characteristic value of the accident alarm is analyzed in the following way:

[0008] Si=Cn*A1+Pn*A2+(Gn*A3+1) 2 ;

[0009] Wherein, A1 is the set weight factor of Cn, A2 is the set weight factor of Pn, and A3 is the set weight factor of Gn.

[0010] As a further solution, the basic information of drone nests near the traffic accident is obtained, and the selected drone nest is matched according to the response priority. The task instruction is sent to the selected drone nest, and the response program of the drone is started. The specific analysis process is as follows: the basic information of drone nests near the traffic accident is obtained, including the distance L between the drone nest and the location of the traffic accident, the normal operation ratio Bl of the equipment carried by the drone nest, and the four-light camera matching rate Pl of the drone nest; if the idle drones in the drone nest are equipped with four-light cameras, the four-light camera matching rate of the drone nest is 1; if the idle drones in the drone nest are not equipped with four-light cameras, the four-light camera matching rate of the drone nest is 0; based on the basic information of drone nests near the traffic accident, the basic information factor of the drone nest is obtained through comprehensive analysis; the basic information of drone nests near the traffic accident is obtained. The response priority of human-machine nest matching is obtained, and the basic information factor of the drone nest of the drone nest near the traffic accident that matches the response priority is obtained, and the drone nest corresponding to the highest value of the drone nest basic information factor is recorded as the selected drone nest; if there are multiple drone nests with the highest basic information factor values ​​at the same time, the distance between the drone nest corresponding to the highest value of the basic information factor of each drone nest and the location of the traffic accident is compared, and the drone nest with the shortest distance to the location of the traffic accident is selected as the selected drone nest; if there is no drone nest near the traffic accident that matches the response priority, a no-matching drone nest notification is issued to the data center, and the emergency drone nest is used, and the emergency drone nest is recorded as the selected drone nest; a task instruction is sent to the selected drone nest to start the drone's response procedure.

[0011] As a further solution, the basic information factors of the drone nest are analyzed in the following specific process:

[0012]

[0013] Where Jc is the basic information factor of the UAV nest, and e is a natural constant.

[0014] As a further solution, after the selected drone nest receives the mission instruction, it obtains the drone's waiting-for-flight characteristic information, plans the drone's waiting-for-flight optimal route, and arrives at the traffic accident scene. The specific analysis process is: obtain the drone's waiting-for-flight characteristic information, including the distance L between the drone and the traffic accident location, the distance Dj between the drone's flight area and the no-fly zone, and the angle θ between the drone's initial flight direction and the no-fly zone; based on the drone's waiting-for-flight characteristic information, comprehensively analyze and obtain the drone's waiting-for-flight characteristic signal; obtain the drone's waiting-for-flight characteristic signal-drone optimal route mapping set stored in the database, and determine the matching drone's waiting-for-flight optimal route based on the current drone's waiting-for-flight characteristic signal, and the drone arrives at the traffic accident scene based on the drone's waiting-for-flight optimal route.

[0015] As a further solution, the specific analysis process of the UAV's flight characteristic signal is as follows:

[0016]

[0017] Where Tz is the characteristic signal of the UAV to be flown, and e is a natural constant.

[0018] As a further solution, the drone starts an intelligent search mode based on visual-thermodynamic joint positioning and autonomous game decision-making to find the accident vehicle. The drone takes pictures from above at the scene of the traffic accident, and the pictures are transmitted to the data center. The specific analysis process is as follows: the drone starts infrared thermal imaging at the first defined height and identifies abnormal heat sources through the residual attention network; after identifying the abnormal heat source, it descends to the second defined height to start multispectral imaging, and inputs the multispectral imaging results into the trained YOL0-v7 and Transformer models to identify vehicle features, including vehicle debris and glass reflection features; the database is compared based on the vehicle features. If the vehicle features match the pre-stored features of the accident vehicle stored in the database, the accident vehicle is confirmed; if the accident vehicle cannot be confirmed, the strong The exploration-exploitation strategy driven by artificial intelligence (AI) involves a 50% probability of continuing the circling search and a 50% probability of switching to terrain-following mode for ground reconnaissance to confirm the accident vehicle. After confirming the accident vehicle, the drone takes a bird's-eye view photo at the accident scene: the lidar scans inherent roadside objects, including streetlight poles and mileage posts, and uses the GTSAM library for graph optimization to match the local point cloud with the high-precision map to establish a centimeter-level global coordinate system. The GAN network is used to deblur images taken in rainy and foggy weather, and a polarized light camera is synchronously triggered to capture the reflective properties of metal parts to assist in determining the collision angle. The drone obtains a bird's-eye view photo of the accident scene, embeds the precise timestamp and spatial coordinates of the Beidou timing chip, generates a hash value, and writes it into a private chain to form an irrefutable chain of evidence. The photo is then transmitted to the data center.

[0019] As a further solution, based on the photos transmitted to the data center, characteristic data of the traffic accident scene are extracted, and a traffic accident scene reconstruction map is automatically generated by combining reference points and baselines with artificial intelligence algorithms. The specific analysis process is as follows: based on the photos transmitted to the data center, characteristic data of the traffic accident scene are extracted, including the direction of vehicle paint scratches and the diffusion pattern of oil stains on the ground; based on NeRF, a 3D scene including the direction of vehicle paint scratches and the diffusion pattern of oil stains on the ground is generated; the vehicle dynamics parameters are imported into the MBDyn multi-body dynamics simulation to calculate the motion trajectory before and after the collision; the Bayesian network is used to fuse the brake mark length, EDR data, and video stream analysis to reversely infer the key actions of lane changing and sudden braking 3 seconds before the collision to generate a traffic accident scene reconstruction map; if there is a contradiction between the visual evidence and the physical simulation, the adversarial generative network is activated to simulate possible scenarios, and the causal chain with the highest probability is output as the traffic accident scene reconstruction map.

[0020] The second aspect of the present invention provides an intelligent traffic accident restoration method based on drone nests, comprising the following steps: a data center receives a traffic accident alarm, extracts basic information about the accident alarm, and automatically divides response priorities; obtains basic information about drone nests near the traffic accident, combines the response priorities, matches the selected drone nest, sends a mission instruction to the selected drone nest, and starts the drone's response program; after the selected drone nest receives the mission instruction, it obtains the drone's flight feature information, plans the drone's optimal flight route, and arrives at the traffic accident scene; the drone starts an intelligent search mode based on visual-thermodynamic joint positioning and autonomous game decision-making to find the accident vehicle, and the drone takes a bird's-eye view photo at the traffic accident scene, and the photos are transmitted to the data center; based on the photos transmitted to the data center, the traffic accident scene feature data is extracted, and a traffic accident scene restoration map is automatically generated by combining the benchmark points and baselines through an artificial intelligence algorithm.

[0021] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0022] (1) The present invention provides an intelligent traffic accident restoration system and method based on drone nests. The response priority division module can automatically divide the priorities according to the basic information of the accident alarm, ensuring that more urgent accidents are handled first, thereby improving the overall response speed and processing efficiency. The selected drone nest matching module can quickly obtain the information of nearby drone nests and select the nests based on the priority, quickly start the drone response program, and reduce manual intervention and decision-making time. The optimal route planning module can plan the best flight route for the drone, so that it can quickly reach the accident scene, shortening the time from alarm to on-site investigation, helping to handle accidents in a timely manner and reduce traffic congestion.

[0023] (2) The present invention uses a combined visual-thermodynamic positioning and autonomous game decision-making through the accident vehicle intelligent search module to accurately locate the accident vehicle, avoiding omissions or errors that may occur in manual searches. The accident scene reconstruction map generation module extracts feature data from photos and automatically generates a reconstruction map using an artificial intelligence algorithm combined with reference points and baselines. This can accurately and intuitively present the overall appearance and details of the accident scene. The system has a high degree of automation, which can reduce the workload of manual on-site investigations and reduce labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of system module connections of the present invention.

[0026] Figure 2 Schematic diagram of the method steps of the present invention. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] Reference Figure 1 As shown, the first aspect of the present invention provides an intelligent traffic accident restoration system based on drone nests (it should be noted that, in this embodiment, each parameter is standardized or normalized as needed before calculation), including: a response priority division module, a selected drone nest matching module, an optimal route planning module, an accident vehicle intelligent search module and an accident scene restoration map generation module.

[0029] In the response priority division module, the data center receives traffic accident alarms, extracts basic information of the accident alarms, and automatically divides the response priorities.

[0030] The specific analysis process is as follows: the data center receives traffic accident alarms and extracts basic information about the accident alarms, including the total number of vehicles involved in the accident Cn, the number of people involved in the accident Pn, and the number of public facilities damaged by the accident Gn; based on the basic information of the accident alarm, a comprehensive analysis is performed to obtain the characteristic value Si of the basic information of the accident alarm; the characteristic value of the basic information of the accident alarm is compared with the characteristic threshold of the basic information of the accident alarm stored in the database; if the characteristic value of the basic information of the accident alarm is lower than the characteristic threshold of the basic information of the accident alarm, the response priority is level two; if the characteristic value of the basic information of the accident alarm is not lower than the characteristic threshold of the basic information of the accident alarm, the response priority is level one.

[0031] The specific analysis process of the basic information characteristic value of the accident alarm is as follows:

[0032] Si=Cn*A1+Pn*A2+(Gn*A3+1) 2 ;

[0033] Wherein, A1 is the set weight factor of Cn, A2 is the set weight factor of Pn, and A3 is the set weight factor of Gn.

[0034] By extracting basic information such as the total number of vehicles involved in an accident, the number of people involved, and the number of damaged public facilities, the system calculates characteristic values ​​and compares them with thresholds to prioritize responses. This system accurately distinguishes the urgency of an accident, avoiding an even distribution of response resources. It prioritizes rescue and investigation resources for more serious incidents, reducing response time and improving overall efficiency. For example, more resources can be quickly mobilized for major accidents (Level 1 response), while resources can be rationally allocated for minor accidents (Level 2 response) to minimize waste.

[0035] The system automatically prioritizes incidents, eliminating the need for complex manual judgment and communication. Upon receiving an alarm, the data center quickly determines the response level based on a predefined algorithm. The command and dispatch system then immediately schedules actions based on the level, accelerating rescue and investigation efforts.

[0036] Select the drone nest matching module, obtain the basic information of the drone nest near the traffic accident, match the selected drone nest based on the response priority, send the mission instruction to the selected drone nest, and start the drone's response program.

[0037] The specific analysis process is as follows: obtain the basic information of the drone nest near the traffic accident, including the distance L between the drone nest and the location of the traffic accident, the normal operation ratio of the equipment carried by the drone nest Bl, and the four-light camera matching rate Pl of the drone nest; if the idle drones of the drone nest are equipped with four-light cameras, the four-light camera matching rate of the drone nest is 1; if the idle drones of the drone nest are not equipped with four-light cameras, the four-light camera matching rate of the drone nest is 0; based on the basic information of the drone nest near the traffic accident, a comprehensive analysis is performed to obtain the basic information factor of the drone nest; obtain the response priority of the drone nest near the traffic accident, and obtain the matching response priority of the drone nest near the traffic accident. The basic information factor of the drone nest, the drone nest corresponding to the highest value of the basic information factor of the drone nest is recorded as the selected drone nest; if there are multiple drone nests with the highest basic information factor values ​​at the same time, the distances between the drone nests corresponding to the highest values ​​of the basic information factors of each drone nest and the location of the traffic accident are compared, and the drone nest with the shortest distance between the drone nest and the location of the traffic accident is selected as the selected drone nest; if there is no drone nest near the traffic accident that matches the response priority, a no-matching drone nest notification is sent to the data center, and the emergency drone nest is used, and the emergency drone nest is recorded as the selected drone nest; a task instruction is sent to the selected drone nest to start the drone's response procedure.

[0038] The basic information factors of the drone nest, the specific analysis process is as follows:

[0039]

[0040] Where Jc is the basic information factor of the drone nest, and e is a natural constant.

[0041] The basic information factor is calculated by comprehensively considering factors such as the distance between the drone nest and the accident location, the proportion of equipment in operation, and the matching rate of four-light cameras. Prioritizing nests that are close, have equipment in good condition, and have suitable cameras. This allows drones to reach the accident scene as quickly as possible, allowing for rapid investigation and other work, reducing waiting time in the early stages of accident handling and ensuring timely understanding of the accident situation.

[0042] If there is no nest with a matching response priority nearby, the emergency drone nest will be activated to ensure that a drone can respond to the accident in a timely manner regardless of the situation, avoiding response delays due to insufficient nest resources and ensuring the stability of response timeliness.

[0043] Focus on the percentage of equipment on the aircraft nest that is functioning properly and the matching rate of quad-optical cameras, prioritizing aircraft nests with well-functioning equipment and equipped with quad-optical cameras. Quad-optical cameras can capture accident scene information from multiple dimensions, such as using visible light for conventional image acquisition, infrared and thermal imaging to detect temperature anomalies and hidden objects, and lidar for 3D modeling. This improves the comprehensiveness and accuracy of accident scene investigations, providing a more reliable basis for accident reconstruction and liability determination.

[0044] The system flexibly matches nests based on incident response priorities and actual nest conditions. When incidents are frequent or resources are scarce, nest resources can be rationally allocated to achieve dynamic resource balance and efficient utilization. For high-priority incidents, superior nest resources are concentrated to ensure effective response; for low-priority incidents, less important nest resources are rationally allocated, ensuring the system's flexibility in responding to diverse incident scenarios.

[0045] Clear drone nest selection rules and emergency drone nest activation mechanisms ensure the system has reliable response plans for a variety of complex situations. Regardless of drone nest resource distribution or the changing accident situation, the appropriate drone nest can be guaranteed to participate in the accident, enhancing the reliability and stability of the entire intelligent traffic accident recovery system.

[0046] The optimal route planning module selects the drone nest and receives the mission instruction, obtains the characteristic information of the drone to be flown, plans the optimal route for the drone to fly, and arrives at the scene of the traffic accident.

[0047] The specific analysis process is as follows: obtaining the characteristic information of the drone waiting to fly, including the distance L between the drone and the location of the traffic accident, the distance Dj between the drone's flight area and the no-fly zone, and the angle θ between the drone's initial flight direction and the no-fly zone; based on the characteristic information of the drone waiting to fly, a comprehensive analysis is performed to obtain the characteristic signal of the drone waiting to fly; obtaining the mapping set of the characteristic signal of the drone waiting to fly and the optimal route of the drone waiting to fly stored in the database, and determining the matching optimal route of the drone waiting to fly based on the current characteristic signal of the drone waiting to fly, and the drone arrives at the traffic accident scene based on the optimal route of the drone waiting to fly.

[0048] The specific analysis process of the UAV's flight characteristic signal is as follows:

[0049]

[0050] Where Tz is the characteristic signal of the UAV waiting to fly.

[0051] Considering the distance between the drone's flight area and the no-fly zone, as well as the angle between the initial flight direction and the no-fly zone, can effectively plan routes that avoid no-fly zones. This prevents legal issues or safety incidents caused by drones accidentally entering no-fly zones, ensuring that drones fly legally, compliantly, and safely.

[0052] By calculating the characteristic signals of the aircraft to be flown, factors such as distance and angle related to no-fly zones are quantitatively incorporated into route planning. This comprehensive and integrated assessment of flight safety factors, rather than focusing solely on single indicators such as distance, enhances the ability to cope with complex airspace environments and ensures that drones stay away from potentially dangerous areas during flight.

[0053] The flight path is planned based on the distance between the drone and the accident location, prioritizing the shortest path to reduce flight time and energy consumption. This allows the drone to quickly reach the accident scene, conduct accident investigations, and improve overall accident handling efficiency.

[0054] Based on the flight characteristics signal and the optimal route mapping set, the system can select the most suitable route from a large number of experienced routes based on different flight characteristics (such as distance and relationship with no-fly zones). This avoids unnecessary detours or circuitous routes, ensures the optimal flight path for the drone, and further improves flight efficiency.

[0055] By leveraging the database's mapping of flight signature signals to optimal routes, route planning can be continuously optimized as data accumulates and updates. This allows for better adaptation to emerging airspace restrictions or changes in drone performance, maintaining the system's adaptability and advancement in route planning.

[0056] The intelligent search module for accident vehicles uses the drone's visual-thermodynamic joint positioning and autonomous game decision-making to start the intelligent search mode to find the accident vehicle. The drone takes photos from above at the scene of the traffic accident and transmits the photos to the data center.

[0057] The specific analysis process is as follows: the drone starts infrared thermal imaging at the first defined altitude and identifies abnormal heat sources through the residual attention network; after identifying the abnormal heat source, it descends to the second defined altitude to start multispectral imaging, and inputs the multispectral imaging results into the trained YOLO-v7 and Transformer models to identify vehicle features, including vehicle debris and glass reflection features; a database comparison is performed based on the vehicle features, and if the vehicle features match the pre-stored features of the accident vehicle stored in the database, the accident vehicle is confirmed; if the accident vehicle cannot be confirmed, it means that the accident vehicle is obscured, and the reinforcement learning-driven exploration-exploitation strategy is initiated: 50% probability of continuing the circling search, 50% probability of continuing the circling search, and 50% probability of continuing the circling search. The drone switches to terrain following mode at a high rate to conduct ground reconnaissance and confirm the accident vehicle. After confirming the accident vehicle, the drone takes photos from above at the scene of the traffic accident: the lidar scans inherent objects on the roadside, including street light poles and mileage posts, and uses the GTSAM library to optimize the graph to match the local point cloud with the high-precision map to establish a centimeter-level global coordinate system. The GAN network is used to deblur images taken in rainy and foggy days, and the polarized light camera is triggered synchronously to capture the reflective characteristics of metal parts to assist in determining the collision angle. The photos taken by the drone at the scene of the traffic accident are obtained, and the precise timestamp and spatial coordinates of the Beidou timing chip are embedded to generate a hash value and write it into the private chain to form an irrefutable chain of evidence. The photos are transmitted to the data center.

[0058] The system employs combined visual and thermodynamic positioning. Wide-area scanning utilizes infrared thermal imaging combined with a residual attention network to identify unusual heat sources. Near-field confirmation utilizes multispectral imaging and advanced models (YOLO-v7 and Transformer) to identify vehicle features. These multiple technologies work together to improve the accuracy of accident vehicle search and positioning, enabling rapid and accurate locating of accident vehicles while reducing search time. This is particularly useful in complex situations, such as obscured locations.

[0059] When a vehicle is obscured, the system activates a reinforcement learning-driven exploration-exploitation strategy, selecting different search modes based on probability. This autonomous decision-making allows for flexible adjustment of search strategies based on actual conditions, improving search success rates and ensuring that no accident vehicles are missed.

[0060] LiDAR scanning combined with the GTSAM database for point cloud matching establishes a high-precision coordinate system, ensuring the accuracy of image spatial information. A GAN network is used to deblur rainy and foggy images, while a polarization camera simultaneously captures the reflective properties of metal to aid in determining collision angles. These multiple methods enhance the quality of captured photos, making them clearer and more informative, providing high-quality data for subsequent accident analysis.

[0061] The photo is embedded with the precise timestamp and spatial coordinates of the Beidou timing chip, and a hash value is generated and written into the private chain. This ensures that the time and space information of the photo is accurate and cannot be tampered with, enhancing the credibility and authority of the data.

[0062] The accident scene reconstruction map generation module extracts the characteristic data of the traffic accident scene based on the photos transmitted to the data center, and automatically generates a traffic accident scene reconstruction map by combining the reference points and baselines through artificial intelligence algorithms.

[0063] The specific analysis process is as follows: based on the photos transmitted to the data center, characteristic data of the traffic accident scene are extracted, including the direction of vehicle paint scratches and the diffusion pattern of oil stains on the ground; based on NeRF, a 3D scene including the direction of vehicle paint scratches and the diffusion pattern of oil stains on the ground is generated; the vehicle dynamics parameters are imported into the MBDyn multi-body dynamics simulation to calculate the motion trajectory before and after the collision; the Bayesian network is used to fuse the brake mark length, EDR data, and video stream analysis to reversely infer the key actions of lane changing and sudden braking 3 seconds before the collision to generate a traffic accident scene reconstruction map; if there is a contradiction between the visual evidence and the physical simulation, the adversarial generative network is activated to simulate possible scenarios and output the causal chain with the highest probability as the traffic accident scene reconstruction map.

[0064] The system extracts characteristic data from photos, such as the direction of vehicle paint scratches and the spread of oil stains on the ground. It also integrates multiple sources of information, including vehicle dynamics parameters, brake mark length, EDR data, and video stream analysis. This allows for a more accurate reconstruction of the accident scene and the on-site conditions after comprehensive analysis, avoiding misjudgments caused by single data points.

[0065] NeRF was used to generate 3D scenes, MBDyn multi-body dynamics simulation was used to calculate vehicle motion trajectories, and Bayesian networks were used to infer key pre-crash maneuvers. These advanced algorithms and tools accurately simulated the accident process, taking into account physical laws and practical details, resulting in a more realistic reconstruction.

[0066] By combining AI algorithms with benchmark points and baselines, the system automatically generates accident scene reconstructions, reducing the time and workload of manual drawing and analysis. Rapidly processing large amounts of data and outputting results, especially during peak accident periods, improves overall accident handling efficiency and provides a timely basis for subsequent liability determination and insurance claims.

[0067] When visual evidence contradicts physical simulations, a generative adversarial network (GAN) simulates possible scenarios and outputs the most probable causal chain. This mechanism quickly resolves data conflicts, avoiding lengthy manual investigations and disputes, and ensuring a smooth incident handling process.

[0068] Reference Figure 2 As shown, the second aspect of the present invention provides an intelligent traffic accident restoration method based on a drone nest, comprising the following steps: a data center receives a traffic accident alarm, extracts basic information of the accident alarm, and automatically divides the response priority.

[0069] Obtain basic information about drone nests near traffic accidents, match selected drone nests based on response priorities, send mission instructions to the selected drone nests, and initiate the drone's response procedure.

[0070] After the selected drone nest receives the mission instruction, it obtains the characteristic information of the drone to be flown, plans the optimal flight route for the drone, and arrives at the scene of the traffic accident.

[0071] Based on visual-thermodynamic joint positioning and autonomous game decision-making, the drone starts the intelligent search mode to find the accident vehicle. The drone takes pictures from above at the scene of the traffic accident and transmits the pictures to the data center.

[0072] Based on the photos transmitted to the data center, the characteristic data of the traffic accident scene is extracted, and the traffic accident scene reconstruction map is automatically generated by combining the benchmark points and baselines through artificial intelligence algorithms.

[0073] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. The intelligent traffic accident restoration system based on drone nest is characterized by: include: Response priority division module: The data center receives traffic accident alarms, extracts basic information about the accident alarms, and automatically divides response priorities; Select the drone nest matching module, obtain the basic information of the drone nest near the traffic accident, match the selected drone nest based on the response priority, send the mission instruction to the selected drone nest, and start the drone's response program; The optimal route planning module selects the drone nest and receives the mission instruction, obtains the drone's flight characteristics information, and plans the optimal route for the drone to arrive at the traffic accident scene; The accident vehicle intelligent search module uses a combination of visual and thermodynamic positioning and autonomous game decision-making to activate an intelligent search mode to locate the accident vehicle. The drone then takes photos from the accident scene and transmits them to the data center. The accident scene reconstruction map generation module extracts the characteristic data of the traffic accident scene based on the photos transmitted to the data center, and automatically generates a traffic accident scene reconstruction map by combining the reference points and baselines through artificial intelligence algorithms.

2. The intelligent traffic accident reconstruction system based on the drone nest according to claim 1 is characterized by: The data center receives traffic accident alarms, extracts basic information about the accident alarms, and automatically prioritizes responses. The specific analysis process is as follows: The data center receives traffic accident alarms and extracts basic information about the accident, including the total number of vehicles involved in the accident (Cn), the number of people involved in the accident (Pn), and the number of public facilities damaged by the accident (Gn); Based on the basic information of the accident alarm, the basic information characteristic value Si of the accident alarm is obtained through comprehensive analysis; Comparing the accident alarm basic information characteristic value with the accident alarm basic information characteristic threshold stored in the database; If the accident alarm basic information characteristic value is lower than the accident alarm basic information characteristic threshold, the response priority is level 2; If the characteristic value of the basic information of the accident alarm is not lower than the characteristic threshold of the basic information of the accident alarm, the response priority is level one.

3. The intelligent traffic accident reconstruction system based on the drone nest according to claim 2 is characterized by: The specific analysis process of the accident alarm basic information characteristic value is as follows: <h2 style=";text-align:left;direction:ltr">Si = Cn*A1+Pn*A2+(Gn*A3+1)<h2 style=";text-align:left;direction:ltr"> 2 <h2 style=";text-align:left;direction:ltr"> ; Wherein, A1 is the set weight factor of Cn, A2 is the set weight factor of Pn, and A3 is the set weight factor of Gn.

4. The intelligent traffic accident reconstruction system based on the drone nest according to claim 1 is characterized by: The basic information of the drone nest near the traffic accident is obtained, and the response priority is combined to match the selected drone nest, send a task instruction to the selected drone nest, and start the drone's response program. The specific analysis process is as follows: Obtain basic information about drone nests near traffic accidents, including the distance L between the drone nest and the location of the traffic accident, the normal operation ratio of the drone nest's onboard equipment Bl, and the drone nest's four-light camera matching rate Pl; If the idle drones in the drone nest are equipped with quad-light cameras, the quad-light camera matching rate of the drone nest is 1; If the idle drones in the drone nest do not have a quad-light camera, the quad-light camera matching rate of the drone nest will be 0; Based on the basic information of drone nests near traffic accidents, the basic information factors of drone nests are obtained through comprehensive analysis; Obtaining the response priority of the drone nests near the traffic accident, obtaining the drone nest basic information factors of the drone nests near the traffic accident that match the response priority, and recording the drone nest corresponding to the highest value of the drone nest basic information factor as the selected drone nest; If there are multiple drone nests with the highest basic information factor values ​​at the same time, the distances between the drone nests corresponding to the highest basic information factors of each drone nest and the location where the traffic accident occurred are compared, and the drone nest with the shortest distance between the drone nest and the location where the traffic accident occurred is selected as the selected drone nest; If there is no drone nest near the traffic accident that matches the response priority, a no-matching drone nest notification is sent to the data center, and the emergency drone nest is used, and the emergency drone nest is recorded as the selected drone nest; Send mission instructions to the selected drone nest and start the drone's response program.

5. The intelligent traffic accident reconstruction system based on the drone nest according to claim 4 is characterized by: The specific analysis process of the basic information factor of the drone nest is as follows: Where Jc is the basic information factor of the UAV nest, and e is a natural constant.

6. The intelligent traffic accident reconstruction system based on the drone nest according to claim 1 is characterized by: After receiving the mission instruction, the selected drone nest obtains the drone's flight feature information, plans the drone's optimal flight route, and arrives at the traffic accident scene. The specific analysis process is as follows: Obtain the characteristic information of the drone to be flown, including the distance L between the drone and the location of the traffic accident, the distance Dj between the drone's flight area and the no-fly zone, and the angle θ between the drone's initial flight direction and the no-fly zone: Based on the UAV’s pending flight characteristic information, a comprehensive analysis is performed to obtain the UAV’s pending flight characteristic signal; Obtain the mapping set of the drone waiting-to-fly characteristic signal and the drone waiting-to-fly optimal route stored in the database, determine the matching drone waiting-to-fly optimal route based on the current drone waiting-to-fly characteristic signal, and the drone arrives at the traffic accident scene based on the drone waiting-to-fly optimal route.

7. The intelligent traffic accident reconstruction system based on the drone nest according to claim 6 is characterized by: The specific analysis process of the UAV flight characteristic signal is as follows: Where Tz is the characteristic signal of the UAV to be flown, and e is a natural constant.

8. The intelligent traffic accident reconstruction system based on drone nest according to claim 1 is characterized by: The drone uses visual-thermodynamic joint positioning and autonomous game decision-making to start an intelligent search mode to find the accident vehicle. The drone takes photos from the accident scene and transmits the photos to the data center. The specific analysis process is as follows: The drone turns on infrared thermal imaging at the first defined altitude and identifies abnormal heat sources through the residual attention network: After identifying the abnormal heat source, the aircraft descends to the second defined altitude and starts multispectral imaging. The multispectral imaging results are input into the trained YOL0-v7 and Transformer models to identify vehicle features, including vehicle debris and glass reflection features. The database is compared based on the vehicle characteristics. If the vehicle characteristics match the pre-stored characteristics of the accident vehicle stored in the database, the accident vehicle is confirmed; If the accident vehicle cannot be identified, a reinforcement learning-driven exploration-exploitation strategy is initiated: there is a 50% chance of continuing the circling search, and a 50% chance of switching to terrain following mode for ground reconnaissance to identify the accident vehicle. After confirming the accident vehicle, the drone takes photos from above at the scene. The LiDAR scans roadside objects, including streetlight poles and mileage posts. Using the GTSAM library, the local point cloud is optimized and matched with the high-precision map to establish a centimeter-level global coordinate system. Use a GAN network to deblur images taken in rainy and foggy weather, and simultaneously trigger a polarized camera to capture the reflective properties of metal parts to assist in determining the collision angle. Obtain a photo taken by a drone at the scene of a traffic accident, embed the precise timestamp and spatial coordinates of the Beidou timing chip, generate a hash value and write it into a private chain, forming an irrefutable chain of evidence; The photos are transferred to the data center.

9. The intelligent traffic accident reconstruction system based on drone nest according to claim 1 is characterized by: Based on the photos transmitted to the data center, the characteristic data of the traffic accident scene is extracted, and the traffic accident scene reconstruction map is automatically generated by combining the reference points and reference lines through the artificial intelligence algorithm. The specific analysis process is as follows: Extract traffic accident scene feature data based on photos transmitted to the data center, including the direction of vehicle paint scratches and the diffusion pattern of oil stains on the ground; Generate 3D scenes based on NeRF, including the direction of vehicle paint scratches and the diffusion of oil stains on the ground; Import vehicle dynamics parameters into MBDyn multi-body dynamics simulation to calculate the motion trajectory before and after the collision; Using a Bayesian network to integrate brake mark length, EDR data, and video stream analysis, we can infer key lane changes and sudden braking actions three seconds before the collision and generate a reconstruction of the traffic accident scene. If there is a contradiction between the visual evidence and the physical simulation, the adversarial generative network is activated to simulate possible scenarios and output the causal chain with the highest probability as the traffic accident scene reconstruction map.

10. The intelligent traffic accident reconstruction method based on the drone nest is applied to the intelligent traffic accident reconstruction system based on the drone nest according to any one of claims 1 to 9, characterized in that: The following steps are involved: The data center receives traffic accident alarms, extracts basic information about the accident alarms, and automatically prioritizes responses; Obtain basic information about drone nests near the traffic accident, match selected drone nests based on response priorities, send mission instructions to the selected drone nests, and initiate the drone response program; After receiving the mission command, the selected drone nest obtains the drone's flight characteristics information, plans the drone's optimal flight route, and arrives at the traffic accident scene; Based on visual-thermodynamic joint positioning and autonomous game decision-making, the drone activates intelligent search mode to locate the accident vehicle. The drone takes photos from above the accident scene and transmits the photos to the data center. Based on the photos transmitted to the data center, the characteristic data of the traffic accident scene is extracted, and the traffic accident scene reconstruction map is automatically generated by combining the benchmark points and baselines through artificial intelligence algorithms.

Images