A method and system for evaluating highway safety based on unmanned aerial vehicle inspection

Abstract

This invention relates to the fields of intelligent transportation and drone inspection technology, and provides a method and system for highway safety assessment based on drone inspection. The method includes the following steps: acquiring first-view data segments and second-view data segments of a preset road section; registering the first-view data segments with the second-view data segments to obtain a traffic flow map; extracting dynamic features of vehicle surface damage and static distribution features of road debris from the traffic flow map to obtain risk metadata including damage evolution trajectories and spatial coordinates of debris; predicting the structural stability attenuation coefficient of vehicles based on the damage evolution trajectory, and predicting the kinetic potential factor triggering secondary events based on the spatial coordinates of debris, to obtain a dynamic risk assessment result. This invention achieves a quantitative assessment of the dynamic deterioration trend of vehicle damage and the risk of road debris, significantly improving the intelligence level of highway inspection and the ability to prevent secondary accidents.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation and drone inspection technology, and particularly relates to a method and system for highway safety assessment based on drone inspection. Background Technology

[0002] Currently, highway inspections mainly rely on manual patrols and fixed surveillance cameras. Manual patrols suffer from limited visibility, delayed response times, and high risks associated with nighttime and inclement weather operations. Fixed surveillance cameras, on the other hand, have blind spots and cannot track dynamic events (such as consecutive vehicle collisions or debris trails) throughout their entire operation.

[0003] In recent years, drone inspection technology has begun to be applied to traffic management. However, existing technologies are mostly single-unit independent operations, which can only achieve simple video transmission and fixed-point photography. They have the following technical bottlenecks: First, the single-unit battery life is limited, and it is impossible to achieve continuous and seamless coverage of long-distance road sections. Second, the image recognition of existing drone inspections is mostly static recognition, which can only mark whether there are obstacles or accidents, and cannot dynamically assess the spatiotemporal evolution trend of vehicle damage (such as whether the damage is aggravated during driving).

[0004] Therefore, there is an urgent need for a drone inspection solution that can achieve long-distance continuous coverage and dynamic identification in order to conduct dynamic safety assessments of highways. Summary of the Invention

[0005] The purpose of this invention is to provide a highway safety assessment method based on unmanned aerial vehicle (UAV) inspection, which aims to solve the above-mentioned technical problems.

[0006] This invention is implemented as follows: a highway safety assessment method based on unmanned aerial vehicle (UAV) inspection, comprising the following steps:

[0007] Acquire drone inspection data streams for a preset road segment; the drone inspection data streams include first-view data segments and second-view data segments with opposite perspectives, which are collected alternately by a first airport and a second airport deployed at both ends of the preset road segment using a multi-drone hopping flight mode;

[0008] Based on the preset road segment geographic coordinates and collection timestamps, the first-view data segment and the second-view data segment are registered to obtain a traffic flow map with a unified spatiotemporal reference.

[0009] The dynamic features of vehicle surface damage and the static distribution features of road debris are extracted from the traffic flow map to obtain risk metadata containing the damage evolution trajectory and the spatial coordinates of the debris.

[0010] The structural stability attenuation coefficient of the vehicle is predicted based on the damage evolution trajectory, and the kinetic potential energy factor that triggers secondary events is predicted based on the spatial coordinates of the debris, thus obtaining the dynamic risk assessment results.

[0011] Based on the results of dynamic risk assessment, the safety level of the pre-set road sections and vehicles is determined.

[0012] Furthermore, the steps for obtaining drone inspection data streams for preset road sections specifically include:

[0013] The drones controlled by the first airport were used to conduct inspections from the first airport to the second airport, and to collect first-view data segments containing vehicle and road image information of the preset road sections.

[0014] The drones controlling the second airport perform inspections from the second airport to the first airport, collecting second-view data segments containing vehicle and road image information of preset road sections;

[0015] The drones at the first and second airports alternately cover and inspect the preset road sections on the timeline, forming a drone inspection data stream.

[0016] Furthermore, based on the preset road segment geographic coordinates and collection timestamps, the first-view data segment is registered with the second-view data segment to obtain a traffic flow map with a unified spatiotemporal reference. This process specifically includes:

[0017] Extract the geographical location and first timestamp of each frame in the first-person perspective data segment to obtain the first spatiotemporal label;

[0018] Extract the geographical location and second timestamp of each frame in the second-view data segment to obtain the second spatiotemporal label;

[0019] Based on the first and second spatiotemporal labels, the first-view data fragments and the second-view data fragments are mapped to the same coordinate system to generate aligned spatiotemporal scatter data.

[0020] Traffic flow maps are obtained by interpolating and fitting the spatiotemporal scattered data.

[0021] Furthermore, the steps of extracting dynamic features of vehicle surface damage and static distribution features of road debris from traffic flow maps to obtain risk metadata including damage evolution trajectories and spatial coordinates of debris specifically include:

[0022] Detect and track vehicle targets in the traffic flow map, and generate cross-view trajectory chains for each vehicle target;

[0023] A temporal comparative analysis is performed on the cross-frame trajectory chain of each vehicle target to identify the damage dynamic characteristics of the vehicle surface in consecutive frames, obtain the surface damage index of the vehicle, and generate a damage evolution trajectory; the damage evolution trajectory is defined as the sequence of changes in the damage index over time.

[0024] Foreign object detection is performed on the static background in the traffic flow map to identify the static distribution characteristics of road debris, locate the debris, and obtain the spatial coordinates of the debris.

[0025] Furthermore, the steps for obtaining dynamic risk assessment results, including predicting the vehicle's structural stability attenuation coefficient based on the damage evolution trajectory and predicting the kinetic potential factor triggering secondary events based on the spatial coordinates of the debris, specifically include:

[0026] Based on the damage evolution trajectory, the rate of change of the damage index over time is fitted to obtain the structural stability attenuation coefficient of the vehicle.

[0027] Obtain the spatial coordinates and physical properties of the debris, and extract the motion vector field of vehicles surrounding the debris;

[0028] Based on the physical properties of the debris and the motion vector field of surrounding vehicles, the probability of the debris being crushed and the risk weight of causing vehicles to lose control are calculated, thus obtaining the motion potential energy factor that triggers a secondary event.

[0029] Another objective of this invention is to provide a highway safety assessment system based on unmanned aerial vehicle (UAV) inspection, for implementing the aforementioned highway safety assessment method based on UAV inspection, specifically including:

[0030] The inspection data acquisition module is used to acquire the drone inspection data stream of a preset road section; the drone inspection data stream includes a first-view data segment and a second-view data segment with opposite perspectives, which are collected alternately by the first airport and the second airport deployed at both ends of the preset road section using a multi-drone hopping mode.

[0031] The traffic flow map generation module is used to register the first-view data segment with the second-view data segment based on the preset road segment geographical coordinates and collection timestamps to obtain a traffic flow map with a unified spatiotemporal reference.

[0032] The risk metadata acquisition module is used to extract the dynamic features of vehicle surface damage and the static distribution features of road debris from the traffic flow map, and obtain risk metadata including the damage evolution trajectory and the spatial coordinates of the debris.

[0033] The dynamic risk assessment module is used to predict the structural stability attenuation coefficient of the vehicle based on the damage evolution trajectory, and to predict the kinetic potential energy factor that triggers secondary events based on the spatial coordinates of the debris, so as to obtain the dynamic risk assessment results.

[0034] The safety level assessment module is used to assess the safety level of preset road sections and vehicles based on the results of dynamic risk assessment.

[0035] Furthermore, the inspection data acquisition module specifically includes:

[0036] The first-view acquisition unit is used to control the drone at the first airport to perform inspections from the first airport to the second airport and acquire first-view data segments containing vehicle and road image information of preset road sections.

[0037] The second-view acquisition unit is used to control the drones at the second airport to perform inspections from the second airport to the first airport and acquire second-view data segments containing vehicle and road image information of preset road sections.

[0038] The data stream generation unit is used to enable drones at the first and second airports to alternately cover and inspect preset road sections on the time axis, forming a drone inspection data stream.

[0039] Furthermore, the traffic flow map generation module specifically includes:

[0040] The first spatiotemporal information extraction unit is used to extract the geographical location and first timestamp of each frame of image in the first-view data segment to obtain the first spatiotemporal tag;

[0041] The second spatiotemporal information extraction unit is used to extract the geographical location and second timestamp of each frame of image in the second perspective data segment to obtain the second spatiotemporal tag;

[0042] The spatiotemporal alignment unit is used to map the first-view data segment and the second-view data segment to the same coordinate system according to the first spatiotemporal label and the second spatiotemporal label, so as to generate aligned spatiotemporal scatter data.

[0043] The interpolation fitting unit is used to perform interpolation fitting on spatiotemporal scattered data to obtain traffic flow maps.

[0044] Furthermore, the risk metadata acquisition module specifically includes:

[0045] The target detection and tracking unit is used to detect and track vehicle targets in the traffic flow map and generate cross-view trajectory chains for each vehicle target.

[0046] The damage evolution unit is used to perform temporal comparative analysis on the cross-frame trajectory chain of each vehicle target, identify the damage dynamic characteristics of the vehicle surface in continuous frames, obtain the surface damage index of the vehicle, and generate the damage evolution trajectory; the damage evolution trajectory is defined as the sequence of changes in the damage index over time.

[0047] The debris identification unit is used to detect foreign objects in the static background of the traffic flow map, identify the static distribution characteristics of debris on the road surface, locate the debris, and obtain the spatial coordinates of the debris.

[0048] Furthermore, the dynamic risk assessment module specifically includes:

[0049] The attenuation coefficient calculation unit is used to fit the rate of change of the damage index over time based on the damage evolution trajectory to obtain the attenuation coefficient of the vehicle's structural stability.

[0050] The debris surrounding information extraction unit is used to obtain the spatial coordinates and physical properties of the debris, and to extract the motion vector field of vehicles around the debris;

[0051] The motion potential energy factor calculation unit is used to calculate the probability of the debris being crushed and the risk weight of causing the vehicle to lose control based on the physical properties of the debris and the motion vector field of the surrounding vehicles, so as to obtain the motion potential energy factor that triggers the secondary event.

[0052] This invention provides a highway safety assessment method based on unmanned aerial vehicle (UAV) inspection. By employing a dual-UAV skip-fly mode for alternating inspection and data collection, it eliminates the blind spots and limited range of single-UAV operations. This invention achieves quantitative tracking of the dynamic changes in vehicle damage during driving by constructing damage evolution trajectories. Furthermore, by introducing the concept of kinetic potential energy factors, this invention uses debris as a risk source to predict dynamic risk levels, enabling early prediction of secondary accident rates. This invention achieves quantitative assessment of the dynamic deterioration trend of vehicle damage and the risk of road debris, significantly improving the intelligence level of highway inspection and the ability to prevent secondary accidents. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the highway safety assessment method based on unmanned aerial vehicle (UAV) inspection provided in an embodiment of the present invention.

[0054] Figure 2 This is a flowchart illustrating step S100 in the highway safety assessment method based on unmanned aerial vehicle (UAV) inspection provided in an embodiment of the present invention.

[0055] Figure 3 This is a flowchart illustrating step S200 in the highway safety assessment method based on unmanned aerial vehicle (UAV) inspection provided in an embodiment of the present invention.

[0056] Figure 4 This is a flowchart illustrating step S300 in the highway safety assessment method based on unmanned aerial vehicle (UAV) inspection provided in an embodiment of the present invention.

[0057] Figure 5This is a flowchart illustrating step S400 in the highway safety assessment method based on unmanned aerial vehicle (UAV) inspection provided in an embodiment of the present invention.

[0058] Figure 6 This is a schematic diagram of the structure of a highway safety assessment system based on unmanned aerial vehicle (UAV) inspection, provided in an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0060] like Figure 1 As shown, in one embodiment of the present invention, a highway safety assessment method based on unmanned aerial vehicle (UAV) inspection is provided, including the following steps:

[0061] S100. Acquire drone inspection data stream for a preset road segment; the drone inspection data stream includes a first-view data segment and a second-view data segment with opposite perspectives, which are collected alternately by a first airport and a second airport deployed at both ends of the preset road segment using a multi-drone hopping mode.

[0062] S200. Based on the preset road segment geographical coordinates and collection timestamp, the first-view data segment and the second-view data segment are registered to obtain a traffic flow map with a unified spatiotemporal reference.

[0063] S300. Extract the dynamic features of vehicle surface damage and the static distribution features of road debris from the traffic flow map to obtain risk metadata including damage evolution trajectory and spatial coordinates of debris.

[0064] S400: Based on the damage evolution trajectory, the structural stability attenuation coefficient of the vehicle is predicted, and based on the spatial coordinates of the debris, the kinetic potential energy factor that triggers secondary events is predicted, thus obtaining dynamic risk assessment results.

[0065] S500: Based on the results of dynamic risk assessment, determine the safety level of the preset road sections and vehicles.

[0066] In practical applications, a first airport (starting point, such as a service area) and a second airport (ending point, such as a toll station) are deployed at both ends of a pre-set section of a highway. The multi-drone skip-flight mode adopted in this embodiment of the invention refers to a collaborative operation protocol based on spatiotemporal complementarity, in which multiple groups (generally 3 drones) alternately carry out inspections, achieving alternation in physical space and seamless connection in time. This not only solves the problem of drone endurance, but also ensures that the collected drone inspection data (mainly image data) stream has the characteristics of temporal interleaving and complementary perspectives, providing high-quality raw input data for subsequent complex spatiotemporal analysis.

[0067] like Figure 2 As shown, in a preferred embodiment of the present invention, the step of obtaining the UAV inspection data stream of a preset road segment, namely step S100, specifically includes:

[0068] S110: Control the drones at the first airport to perform inspections from the first airport to the second airport, and collect first-view data segments containing vehicle and road image information of preset road sections.

[0069] S120: Control the drones at the second airport to perform inspections from the second airport to the first airport, and collect second-view data segments containing vehicle and road image information of preset road sections.

[0070] S130 enables drones from the first and second airports to alternately cover and inspect preset road sections on the timeline, forming a drone inspection data stream.

[0071] In practical application, an 83-kilometer section of a highway is used as an example. A first airport is deployed at the starting point K0, and a second airport at the ending point K83. Kilometer markers (K0, K1, K2, ..., K83) are used along the route. Additionally, three drones are deployed in a multi-drone hop-drop mode for inspection missions, as follows: At time t1, drone #1 takes off from the first airport and performs an inspection mission from the starting point K0 (north) to the ending point K83 (south), collecting image data of the K0-K83 section during the time period t1-t2, generating a first-person perspective data segment; at time t2, drone #1 lands at the second airport for charging, and drone #2 immediately takes off from the second airport to perform an inspection mission from the ending point K83 (south) to the starting point K0 (north), collecting image data of the K83-K0 section during the time period t2-t3, generating a second-person perspective data segment; at time t3, ... Drone No. 2 lands at the first airport to recharge, and Drone No. 3 immediately takes off from the first airport to perform an inspection mission from the starting point K0 (north) to the ending point K83 (south), collecting image data of the K0-K83 section during the time period t3 to t4, generating a first-view data segment; at time t4, Drone No. 3 lands at the second airport to recharge, and Drone No. 1 immediately takes off from the second airport to perform an inspection mission from the ending point K83 (south) to the starting point K0 (north), collecting image data of the K83-K0 section, generating a second-view data segment; by repeating the above steps, a complete drone inspection data stream with different perspective image data for the road segment can be obtained; in this embodiment of the invention, by setting up 3 drones to adopt the above multi-drone hopping mode, uninterrupted alternating inspection and data collection of the above road segment can be achieved, and the problem of limited drone endurance can also be solved.

[0072] like Figure 3 As shown, in a preferred embodiment of the present invention, the step of registering the first-view data segment with the second-view data segment based on the preset road segment geographical coordinates and collection timestamp to obtain a traffic flow map with a unified spatiotemporal reference, namely step S200, specifically includes:

[0073] S210. Extract the geographical location and first timestamp of each frame of the first-view data segment to obtain the first spatiotemporal label;

[0074] S220. Extract the geographical location and second timestamp of each frame of the second-view data segment to obtain the second spatiotemporal label;

[0075] S230. Based on the first and second spatiotemporal labels, map the first-view data fragments and the second-view data fragments to the same coordinate system to generate aligned spatiotemporal scatter data.

[0076] S240. Interpolate and fit the spatiotemporal scatter data to obtain the traffic flow map.

[0077] Specifically, a two-dimensional spatiotemporal coordinate system is established with the kilometer markers of the road segment as the horizontal axis reference and the timestamp as the vertical axis reference. The first-view data segment and the second-view data segment are mapped onto the two-dimensional spatiotemporal coordinate system according to the first spatiotemporal label and the second spatiotemporal label, respectively. Then, interpolation alignment is performed according to the geographical location (such as linear interpolation method) to eliminate the perspective offset and time gap caused by camera position switching, and a traffic flow map with continuous spatiotemporal characteristics can be generated to characterize the vehicle movement trajectory and road surface condition of the entire domain.

[0078] In practical applications, firstly, let F be the image of the j-th frame in the i-th data segment. ij The corresponding geographical location is P ij =(x ij ,y ij ), timestamp t ij Then its spatiotemporal label is defined as L. ij =(P ij ,t ij ); where i=1 or 2, i=1 represents a first-person perspective data segment, and i=2 represents a second-person perspective data segment; (x ij ,y ij The coordinates () represent the latitude and longitude coordinates acquired in real time by the UAV's positioning and navigation module. Next, each frame of the first-view and second-view data segments is mapped onto a two-dimensional spatiotemporal coordinate system with kilometer markers as the horizontal axis (s) and timestamps as the vertical axis (t), resulting in a spatiotemporal scatter set S = {(s...} k ,t k ,I k )};wherein, s k Represents the geographical location P corresponding to the k-th scatter point. ij The mileage value projected onto the centerline of the road segment can be obtained through a map matching algorithm; t k Represents the timestamp of the k-th scatter point; I k This represents the feature vector (such as grayscale, texture, or semantic features) of the image corresponding to the k-th scatter point. Then, due to the temporal gaps and viewpoint shifts caused by drone switching, interpolation fitting is required on the above spatiotemporal scatter point data (set) to generate a continuous spatiotemporal traffic flow map T(s,t); specifically, the expression for T(s,t) is as follows:

[0079] ;

[0080] In the formula, For interpolation weights, inverse distance weighted interpolation or Kriging interpolation can be used.

[0081] like Figure 4As shown, in a preferred embodiment of the present invention, the step of extracting dynamic features of vehicle surface damage and static distribution features of road debris from traffic flow maps to obtain risk metadata including damage evolution trajectories and spatial coordinates of debris, specifically step S300, includes:

[0082] S310. Detect and track vehicle targets in the traffic flow map, and generate cross-view trajectory chains for each vehicle target;

[0083] In practical applications, existing algorithms such as YOLOv8 and Deepsort can be used to perform surface damage detection and motion trajectory tracking on vehicle targets in traffic flow maps, thereby obtaining a cross-camera trajectory chain for each vehicle target. A cross-camera trajectory chain refers to a motion path record with continuous spatiotemporal identification formed by associating and connecting isolated detection positions of the same moving target (vehicle) at different camera angles and time segments within a traffic flow map acquired from multiple perspectives and spatiotemporally aligned. For example, when a vehicle moves from region A (provided by the first-view data segment) to region B (provided by the second-view data segment) in the traffic flow map, if feature comparison identifies it as the same vehicle, then all geographical locations of this vehicle in region A (trajectory points 1, 2, 3...) and all geographical locations in region B (trajectory points N, N+1...) are connected in chronological order to form a complete cross-camera trajectory chain with timestamps.

[0084] S320. Perform time-series comparative analysis on the cross-frame trajectory chain of each vehicle target to identify the damage dynamic characteristics of the vehicle surface in consecutive frames, obtain the surface damage index of the vehicle, and generate the damage evolution trajectory; the damage evolution trajectory is defined as the sequence of changes in the damage index over time.

[0085] Specifically, the dynamic characteristics of vehicle surface damage include damage indices that change over time (calculated by damage area or damage depth); taking damage area as an example, the damage index is defined as follows:

[0086] ;

[0087] in, This represents the damage index of vehicle m. express The damage area on the surface of vehicle m at time t can be obtained by identifying the damage area of ​​vehicle m using the existing YOLOv8 algorithm and performing pixel-level segmentation. express The projected area of ​​vehicle m in the image at time m; the damage evolution trajectory is defined as the sequence of changes in damage indicators over time.

[0088] S330. Foreign object detection is performed on the static background in the traffic flow map to identify the static distribution characteristics of road debris, locate the debris, and obtain the spatial coordinates of the debris; the damage evolution trajectory and the spatial coordinates of the debris are encapsulated into risk metadata.

[0089] Specifically, for road debris, existing semantic segmentation networks (such as DeepLabV3+) can be used to extract its mask from the traffic flow map, and then the geographic coordinates can be calculated by the camera imaging model of the UAV to obtain the spatial coordinates of the debris.

[0090] like Figure 5 As shown, in a preferred embodiment of the present invention, the step of obtaining a dynamic risk assessment result by predicting the structural stability attenuation coefficient of the vehicle based on the damage evolution trajectory and the kinetic potential energy factor that triggers secondary events based on the spatial coordinates of the debris is specifically step S400, which includes:

[0091] S410. Based on the damage evolution trajectory, fit the rate of change of the damage index over time to obtain the structural stability attenuation coefficient of the vehicle.

[0092] Specifically, the structural stability attenuation coefficient of vehicle m Calculate according to the following formula:

[0093] ;

[0094] In the formula, For time intervals; This is the initial damage indicator.

[0095] S420. Obtain the spatial coordinates and physical properties of the debris, and extract the motion vector field of vehicles surrounding the debris;

[0096] S430. Based on the physical properties of the debris and the motion vector field of surrounding vehicles, calculate the probability of the debris being crushed and the risk weight of causing the vehicles to lose control, and obtain the motion potential energy factor that triggers the secondary event.

[0097] In practical applications, the kinematic potential energy factor is defined as the critical risk measure that may lead to a secondary traffic accident after debris is run over, crushed, or disturbed by a vehicle; specifically, the kinematic potential energy factor of debris s... The calculation formula is as follows:

[0098] ;

[0099] In the formula, f s The friction coefficient between the debris s and the road surface is given. The material of the debris is identified by querying a material database. The material of the debris can be identified by the existing ResNet50 network. The density of the scattered material *s* is determined based on the material of the scattered material; V s Let be the volume of the scattered object 's', which can be estimated using the segmentation mask of an existing semantic segmentation network (such as DeepLabV3+); g is the acceleration due to gravity; h s The height characteristic of the scattered object 's' is determined by its thickness for flat objects and its center of gravity height for three-dimensional objects; v m The average speed of vehicles surrounding the debris s; Let be the angle between the windward side of the debris s and the direction of vehicle travel, which is estimated based on the shape and orientation of the debris; This is a preset baseline energy (such as the energy of a typical car impacted at 120 km / h).

[0100] It should be noted that by normalizing the vehicle's structural stability attenuation coefficient and kinetic potential energy factor and then fusing them, a dynamic risk assessment result containing risk level and risk location can be generated.

[0101] In a preferred embodiment of the present invention, step S500 is implemented as follows: First, based on the dynamic risk assessment results, the vehicle risk object set and the debris risk object set are analyzed and separated to generate a risk object list; the risk object list includes the identifier, location, structural stability attenuation coefficient, and kinematic potential energy factor of each risk object; then, based on preset safety level classification rules, the safety level of each vehicle risk object and debris risk object in the risk object list is assessed to obtain an individual safety level list; in practical applications, the safety level classification rules can be set as follows: the safety level can be divided into four levels from low to high. The structural stability attenuation coefficient of the vehicle is compared with three preset attenuation coefficient thresholds sorted from smallest to largest to assess the safety level of the vehicle risk object, wherein the smaller the structural stability attenuation coefficient of the vehicle, the higher the safety level; similarly, the kinematic potential energy factor of the debris is compared with three preset potential energy factor thresholds sorted from smallest to largest to assess the safety level of the debris risk object, wherein the smaller the kinematic potential energy factor of the debris, the higher the safety level.

[0102] like Figure 6 As shown, in another embodiment of the present invention, a highway safety assessment system based on unmanned aerial vehicle (UAV) inspection is also provided to implement the above-mentioned highway safety assessment method based on UAV inspection, specifically including:

[0103] The inspection data acquisition module 10 is used to acquire the drone inspection data stream of a preset road section; the drone inspection data stream includes a first-view data segment and a second-view data segment with opposite perspectives, which are collected alternately by the first airport and the second airport deployed at both ends of the preset road section using a multi-drone hopping mode.

[0104] The traffic flow map generation module 20 is used to register the first-view data segment with the second-view data segment based on the preset road segment geographical coordinates and collection timestamps to obtain a traffic flow map with a unified spatiotemporal reference.

[0105] The risk metadata acquisition module 30 is used to extract the dynamic features of vehicle surface damage and the static distribution features of road debris from the traffic flow map, and obtain risk metadata including the damage evolution trajectory and the spatial coordinates of the debris.

[0106] The dynamic risk assessment module 40 is used to predict the structural stability attenuation coefficient of the vehicle based on the damage evolution trajectory, and to predict the kinetic potential energy factor that triggers secondary events based on the spatial coordinates of the debris, so as to obtain the dynamic risk assessment results.

[0107] The safety level assessment module 50 is used to assess the safety level of preset road sections and vehicles based on the results of dynamic risk assessment.

[0108] In a preferred embodiment of the present invention, the inspection data acquisition module 10 specifically includes:

[0109] The first-view acquisition unit is used to control the drone at the first airport to perform inspections from the first airport to the second airport and acquire first-view data segments containing vehicle and road image information of preset road sections.

[0110] The second-view acquisition unit is used to control the drones at the second airport to perform inspections from the second airport to the first airport and acquire second-view data segments containing vehicle and road image information of preset road sections.

[0111] The data stream generation unit is used to enable drones at the first and second airports to alternately cover and inspect preset road sections on the time axis, forming a drone inspection data stream.

[0112] In a preferred embodiment of the present invention, the traffic flow map generation module 20 specifically includes:

[0113] The first spatiotemporal information extraction unit is used to extract the geographical location and first timestamp of each frame of image in the first-view data segment to obtain the first spatiotemporal tag;

[0114] The second spatiotemporal information extraction unit is used to extract the geographical location and second timestamp of each frame of image in the second perspective data segment to obtain the second spatiotemporal tag;

[0115] The spatiotemporal alignment unit is used to map the first-view data segment and the second-view data segment to the same coordinate system according to the first spatiotemporal label and the second spatiotemporal label, so as to generate aligned spatiotemporal scatter data.

[0116] The interpolation fitting unit is used to perform interpolation fitting on spatiotemporal scattered data to obtain traffic flow maps.

[0117] In a preferred embodiment of the present invention, the risk metadata acquisition module 30 specifically includes:

[0118] The target detection and tracking unit is used to detect and track vehicle targets in the traffic flow map and generate cross-view trajectory chains for each vehicle target.

[0119] The damage evolution unit is used to perform temporal comparative analysis on the cross-frame trajectory chain of each vehicle target, identify the damage dynamic characteristics of the vehicle surface in continuous frames, obtain the surface damage index of the vehicle, and generate the damage evolution trajectory; the damage evolution trajectory is defined as the sequence of changes in the damage index over time.

[0120] The debris identification unit is used to detect foreign objects in the static background of the traffic flow map, identify the static distribution characteristics of debris on the road surface, locate the debris, and obtain the spatial coordinates of the debris.

[0121] In a preferred embodiment of the present invention, the dynamic risk assessment module 40 specifically includes:

[0122] The attenuation coefficient calculation unit is used to fit the rate of change of the damage index over time based on the damage evolution trajectory to obtain the attenuation coefficient of the vehicle's structural stability.

[0123] The debris surrounding information extraction unit is used to obtain the spatial coordinates and physical properties of the debris, and to extract the motion vector field of vehicles around the debris;

[0124] The motion potential energy factor calculation unit is used to calculate the probability of the debris being crushed and the risk weight of causing the vehicle to lose control based on the physical properties of the debris and the motion vector field of the surrounding vehicles, so as to obtain the motion potential energy factor that triggers the secondary event.

[0125] It should be noted that the above modules and units can be implemented as a computer program, which can run on a computer device. The computer device's memory can store the computer program that makes up the modules and units, enabling the processor to execute the various steps of the above method.

[0126] It should be understood that although the steps in the flowcharts of the embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0127] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods.

[0128] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A highway safety assessment method based on unmanned aerial vehicle (UAV) inspection, characterized in that, Includes the following steps: Acquire drone inspection data streams for preset road sections; The UAV inspection data stream includes a first-view data segment and a second-view data segment with opposite perspectives, which are collected alternately by the first airport and the second airport deployed at both ends of the preset road segment using a multi-aircraft skip-fly mode. Based on the preset road segment geographic coordinates and collection timestamps, the first-view data segment and the second-view data segment are registered to obtain a traffic flow map with a unified spatiotemporal reference. The dynamic features of vehicle surface damage and the static distribution features of road debris are extracted from the traffic flow map to obtain risk metadata containing the damage evolution trajectory and the spatial coordinates of the debris. The structural stability attenuation coefficient of the vehicle is predicted based on the damage evolution trajectory, and the kinetic potential energy factor that triggers secondary events is predicted based on the spatial coordinates of the debris, thus obtaining the dynamic risk assessment results. Based on the results of dynamic risk assessment, the safety level of the pre-set road sections and vehicles is determined.

2. The highway safety assessment method based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The steps for obtaining drone inspection data streams for preset road sections specifically include: The drones controlled by the first airport were used to conduct inspections from the first airport to the second airport, and to collect first-view data segments containing vehicle and road image information of the preset road sections. The drones controlling the second airport perform inspections from the second airport to the first airport, collecting second-view data segments containing vehicle and road image information of preset road sections; The drones at the first and second airports alternately cover and inspect the preset road sections on the timeline, forming a drone inspection data stream.

3. The highway safety assessment method based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The steps involved in registering first-view data segments with second-view data segments based on preset road segment geographic coordinates and data collection timestamps to obtain a traffic flow map with a unified spatiotemporal reference include: Extract the geographical location and first timestamp of each frame in the first-person perspective data segment to obtain the first spatiotemporal label; Extract the geographical location and second timestamp of each frame in the second-view data segment to obtain the second spatiotemporal label; Based on the first and second spatiotemporal labels, the first-view data fragments and the second-view data fragments are mapped to the same coordinate system to generate aligned spatiotemporal scatter data. Traffic flow maps are obtained by interpolating and fitting the spatiotemporal scattered data.

4. The highway safety assessment method based on unmanned aerial vehicle (UAV) inspection according to claim 1, characterized in that, The steps involved in extracting dynamic features of vehicle surface damage and static distribution features of road debris from traffic flow maps to obtain risk metadata including damage evolution trajectories and spatial coordinates of debris specifically include: Detect and track vehicle targets in the traffic flow map, and generate cross-view trajectory chains for each vehicle target; A temporal comparative analysis is performed on the cross-frame trajectory chain of each vehicle target to identify the damage dynamic characteristics of the vehicle surface in consecutive frames, obtain the surface damage index of the vehicle, and generate a damage evolution trajectory; the damage evolution trajectory is defined as the sequence of changes in the damage index over time. Foreign object detection is performed on the static background in the traffic flow map to identify the static distribution characteristics of road debris, locate the debris, and obtain the spatial coordinates of the debris.

5. The highway safety assessment method based on unmanned aerial vehicle (UAV) inspection according to claim 4, characterized in that, The steps for obtaining dynamic risk assessment results, including predicting the vehicle's structural stability attenuation coefficient based on the damage evolution trajectory and predicting the kinetic potential factor triggering secondary events based on the spatial coordinates of debris, specifically include: Based on the damage evolution trajectory, the rate of change of the damage index over time is fitted to obtain the structural stability attenuation coefficient of the vehicle. Obtain the spatial coordinates and physical properties of the debris, and extract the motion vector field of vehicles surrounding the debris; Based on the physical properties of the debris and the motion vector field of surrounding vehicles, the probability of the debris being crushed and the risk weight of causing vehicles to lose control are calculated, thus obtaining the motion potential energy factor that triggers a secondary event.

6. A highway safety assessment system based on unmanned aerial vehicle (UAV) inspection, used to implement the highway safety assessment method based on UAV inspection as described in any one of claims 1-5, characterized in that, include: The inspection data acquisition module is used to acquire the drone inspection data stream for the preset road sections; The UAV inspection data stream includes a first-view data segment and a second-view data segment with opposite perspectives, which are collected alternately by the first airport and the second airport deployed at both ends of the preset road segment using a multi-aircraft skip-fly mode. The traffic flow map generation module is used to register the first-view data segment with the second-view data segment based on the preset road segment geographical coordinates and collection timestamps to obtain a traffic flow map with a unified spatiotemporal reference. The risk metadata acquisition module is used to extract the dynamic features of vehicle surface damage and the static distribution features of road debris from the traffic flow map, and obtain risk metadata including the damage evolution trajectory and the spatial coordinates of the debris. The dynamic risk assessment module is used to predict the structural stability attenuation coefficient of the vehicle based on the damage evolution trajectory, and to predict the kinetic potential energy factor that triggers secondary events based on the spatial coordinates of the debris, so as to obtain the dynamic risk assessment results. The safety level assessment module is used to assess the safety level of preset road sections and vehicles based on the results of dynamic risk assessment.

7. The highway safety assessment system based on unmanned aerial vehicle (UAV) inspection according to claim 6, characterized in that, The inspection data acquisition module specifically includes: The first-view acquisition unit is used to control the drone at the first airport to perform inspections from the first airport to the second airport and acquire first-view data segments containing vehicle and road image information of preset road sections. The second-view acquisition unit is used to control the drones at the second airport to perform inspections from the second airport to the first airport and acquire second-view data segments containing vehicle and road image information of preset road sections. The data stream generation unit is used to enable drones at the first and second airports to alternately cover and inspect preset road sections on the time axis, forming a drone inspection data stream.

8. The highway safety assessment system based on unmanned aerial vehicle (UAV) inspection according to claim 6, characterized in that, The traffic flow map generation module specifically includes: The first spatiotemporal information extraction unit is used to extract the geographical location and first timestamp of each frame of image in the first-view data segment to obtain the first spatiotemporal tag; The second spatiotemporal information extraction unit is used to extract the geographical location and second timestamp of each frame of image in the second perspective data segment to obtain the second spatiotemporal tag; The spatiotemporal alignment unit is used to map the first-view data segment and the second-view data segment to the same coordinate system according to the first spatiotemporal label and the second spatiotemporal label, so as to generate aligned spatiotemporal scatter data. The interpolation fitting unit is used to perform interpolation fitting on spatiotemporal scattered data to obtain traffic flow maps.

9. The highway safety assessment system based on unmanned aerial vehicle (UAV) inspection according to claim 6, characterized in that, The risk metadata acquisition module specifically includes: The target detection and tracking unit is used to detect and track vehicle targets in the traffic flow map and generate cross-view trajectory chains for each vehicle target. The damage evolution unit is used to perform temporal comparative analysis on the cross-frame trajectory chain of each vehicle target, identify the damage dynamic characteristics of the vehicle surface in continuous frames, obtain the surface damage index of the vehicle, and generate the damage evolution trajectory; the damage evolution trajectory is defined as the sequence of changes in the damage index over time. The debris identification unit is used to detect foreign objects in the static background of the traffic flow map, identify the static distribution characteristics of debris on the road surface, locate the debris, and obtain the spatial coordinates of the debris.

10. The highway safety assessment system based on unmanned aerial vehicle (UAV) inspection according to claim 9, characterized in that, The dynamic risk assessment module specifically includes: The attenuation coefficient calculation unit is used to fit the rate of change of the damage index over time based on the damage evolution trajectory to obtain the attenuation coefficient of the vehicle's structural stability. The debris surrounding information extraction unit is used to obtain the spatial coordinates and physical properties of the debris, and to extract the motion vector field of vehicles around the debris; The motion potential energy factor calculation unit is used to calculate the probability of the debris being crushed and the risk weight of causing the vehicle to lose control based on the physical properties of the debris and the motion vector field of the surrounding vehicles, so as to obtain the motion potential energy factor that triggers the secondary event.

Images