CN-121564593-B - Wet-skid road risk assessment method based on unmanned aerial vehicle recognition vehicle track

CN121564593BCN 121564593 BCN121564593 BCN 121564593BCN-121564593-B

Abstract

The invention belongs to the field of risk prediction, and relates to a wet slippery road risk assessment method based on vehicle track identification by an unmanned aerial vehicle, wherein the method adopts a mode of carrying a cradle head by a high-resolution unmanned aerial vehicle, extracts video data of a vehicle running track of a dangerous road section, maps the video data to obtain a vehicle ground track sequence, calculates parameter indexes of a normal sample and the wet slippery road according to a reference track curve, and compares the wet slippery road with the normal sample to output distribution difference; based on meteorological coefficients, the method is used for adaptively adjusting the upper threshold index under the wet road, and calculating the weighted upper threshold proportion and the duration exceeding 95% quantiles of normal weather; and judging the adjacent consistency and the space consistency in the segment to finally form a segment-level risk score and a risk level. The accuracy, the instantaneity and the interpretability of risk early warning are obviously improved, and the false alarm rate caused by environmental misjudgment or equipment error is effectively reduced.

Inventors

XI JIANFENG
CHENG LIN
Zhao Yanglan
Ma Qingqiu
KANG HONGBIN
ZHANG HAO
ZHENG LILI
DING TONGQIANG
DONG YU
WANG HAITAO
ZHANG TIEMIN
HAO DALEI
Lv Baibing

Assignees

吉林大学

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. The wet skid road risk assessment method based on unmanned aerial vehicle recognition vehicle track is characterized by comprising the following steps of: S1, acquiring data information by using an unmanned aerial vehicle, outputting an original video frame and a flight control log, interpolating a meteorological data time sequence to a frame time to form a unified frame-posture-meteorological time axis, and generating disposable acquisition data; S2, camera calibration and ground mapping are carried out, and at least one pixel-ground mapping model is constructed; s3, detecting and timing tracking vehicles in the unmanned aerial vehicle video, converting the vehicle from pixel coordinates to ground coordinates based on a pixel-ground mapping model, and outputting a vehicle ground track sequence; S4, under normal weather, calculating parameter indexes of a normal sample according to a reference track curve, wherein the parameter indexes comprise transverse offset, heading error, angular velocity fluctuation and velocity-direction coupling index, calculating track oscillation frequency based on acquired data, estimating road surface attachment coefficient and dynamic stability boundary, calculating track curvature abnormal index and yaw-sideslip coupling degree, and outputting all the data after normalization so as to construct a normal weather distribution library; s5, outputting standardized indexes which can be directly compared with a normal weather distribution library under a wet road according to the multi-vehicle track and the reference track curve obtained in the step S3, and generating an upper threshold index at the same time; S6, comparing the standardized index output in the step S5 with the normal weather distribution library in the step S4, outputting distribution differences, adaptively adjusting the upper threshold index under the slippery road based on meteorological coefficients, calculating the weight upper threshold proportion and the duration time exceeding 95% quantile of normal weather, and judging the adjacency consistency and the intra-segment space consistency to finally form the segment risk score and the risk grade.
2. The method for estimating the risk of a wet skid road based on the identification of vehicle trajectories by unmanned aerial vehicle according to claim 1, wherein after the calibration of the camera in step S2, all aerial frames are de-distorted and then ground mapped, and according to the road condition, a pixel-ground mapping model is built for the planar road section, and another pixel-ground mapping model is built for the heave/bridge floor/ramp road section.
3. The wet road risk assessment method based on unmanned aerial vehicle recognition vehicle track according to claim 1, wherein step S3 adopts YOLOv to output a vehicle candidate frame and confidence level for each frame, and then applies DeepSORT multi-objective tracking algorithm to realize cross-frame vehicle ID matching, so as to determine the vehicle track.
4. The wet skid road risk assessment method based on the unmanned aerial vehicle recognition vehicle track according to claim 1, wherein in the step S4, the target road section is divided into a plurality of sections along the main trend of the road, a reference track curve is fitted for each section of road on a normal weather sample based on the multi-vehicle track obtained in the step S3, the fitting process converges in an iterative mode of searching a nearest point to calculate a normal residual error, updating parameters and correcting the curve, and after the fitting is finished, each section of curve is spliced to ensure that the sections are continuous in a first order at the end points.
5. The method for estimating risk of a wet road based on vehicle trajectory recognition by an unmanned aerial vehicle according to claim 4, wherein in step S4, after a reference trajectory curve is fitted to each segment, scenario dimensions are further distinguished according to curvature levels and lanes, a normal weather distribution library is constructed for each scenario dimension, and wet road sample data is compared with the normal weather distribution library according to the scenario dimensions.
6. The wet skid road risk assessment method based on the unmanned aerial vehicle recognition vehicle track according to claim 5, wherein in the step S5, track oscillation frequency is calculated based on transverse offset, slip angle is calculated based on speed, yaw rate, course and track direction, estimated road adhesion coefficient is obtained, adhesion estimated index and dynamic stability boundary are calculated, track curvature abnormal index is calculated based on instantaneous curvature and expected curvature of actual track of the vehicle, and yaw-side slip coupling degree is calculated based on yaw rate and centroid slip angle.
7. The wet road risk assessment method based on unmanned aerial vehicle recognition vehicle track according to claim 6, wherein the risk score at the mth segment k time in step S6 The expression of (2) is: ; Wherein, the The method comprises the following steps of transverse offset, heading error, angular velocity fluctuation, speed-direction coupling, track oscillation frequency, adhesion estimation index, track curvature anomaly index, dynamic stability boundary and yaw-sideslip coupling degree, wherein m is a road section, k is a time window; ; a non-negative weight; Threshold shrink factor; Adjoining consistency; spatial continuity in the segment; The value continuously exceeds the expected value or average value of the duration of 95% of the quantiles of normal weather; The experience distribution is converted into the minimum cost required by the normal baseline distribution; 95%,99% fraction ratio, c is bias term; The risk level is then determined based on the risk score.
8. The method for estimating risk of a wet road based on recognizing vehicle trajectories by unmanned aerial vehicle according to claim 7, wherein the degree of adjacency consistency ; Wherein, the Pre-scoring the previous step or period, Weighted sum of threshold ratio for lateral offset and heading error or compound slip suspicion Is a mean value in the window of (c), Representing the segmentation of the object, As the total number of road segments, Is a threshold value.
9. The method for wet road risk assessment based on unmanned aerial vehicle identification vehicle trajectory of claim 7, wherein the minimum cost required for the empirical distribution to be converted to a normal baseline distribution The expression of (2) is: ; Wherein, the The minimum cost required to convert the empirical distribution of the current window to a normal baseline distribution, For the in-window weighted empirical distribution, Is the normal distribution of step S4.
10. A computer readable medium on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements a wet skid road risk assessment method based on the recognition of vehicle trajectories by unmanned aerial vehicles as claimed in any one of claims 1 to 9.

Description

Wet-skid road risk assessment method based on unmanned aerial vehicle recognition vehicle track Technical Field The invention belongs to the technical field of risk prediction, and particularly relates to a wet skid road risk assessment method based on vehicle track identification of an unmanned aerial vehicle. Background The traffic industry in China develops rapidly, the traffic infrastructure is continuously perfect, but in severe weather conditions, particularly traffic danger early warning on ice and snow roads still has a relatively large weak link, and traffic safety of dangerous road sections in ice and snow weather still faces serious challenges. According to investigation and analysis, under severe weather taking ice and snow as main influencing factors, the problem that potential safety hazards exist when a vehicle runs on a dangerous road section is particularly serious due to the fact that partial snow or ice areas exist on the road. According to the existing road detection technology and the existing danger early warning method, the risk level and the existing potential safety hazard degree of the road cannot be accurately analyzed, so that the road cannot be effectively prevented and targeted control measures cannot be executed, the probability of traffic accidents on the road is increased, and the traffic accidents are easily caused. Therefore, how to reasonably early warn the risk level of the road becomes one of effective means for reducing traffic accidents. The assessment method of the vehicle driving wet road section at the present stage mainly comprises the following steps: (1) The method is characterized in that fixed sensing equipment is deployed at a road key point, and environment and road surface physical parameters related to road hazards are measured directly or indirectly. It assumes that anomalies in these parameters (e.g., temperatures below freezing, water films on road surfaces, etc.) are directly causally related to driving risk. This has the advantage that the data is direct, reliable and continuous. But has the disadvantages of small coverage, high cost, inflexibility and incapability of reflecting vehicle behavior. (2) The method is based on the evaluation method of the sensor of the vehicle, and the key principle of the method is that a sensing system of the vehicle is utilized to monitor the dynamic response and the external environment of the vehicle so as to identify whether the vehicle is in or is about to be in a dangerous state. It detects dangerous "fruits" directly from the perspective of the "victim" (vehicle). The method can reflect the state of the bicycle in real time without modifying the road side. But the method has limited sensing range, depends on permeability, data island and early warning lag. (3) The road evaluation method based on the macroscopic statistical model is characterized in that the core principle of the method is to predict which road sections have higher risk probability in a future period by utilizing the statistical rule of historical data and combining with real-time macroscopic weather forecast. The method is a prediction method based on historical replay and physical rule deduction. The method has the advantages of large-scale prediction and low cost, but has poor precision, low real-time performance and lack of microscopic insight. In view of the above, the existing wet road risk assessment system still has some drawbacks. Firstly, the existing road risk assessment method needs to arrange sensing equipment for extracting corresponding environment and vehicle data on a road or a vehicle, has relatively high cost, is greatly influenced by weather and natural environment, has relatively limited prediction range, has certain limitation (can not cope with the risk early warning problem of a plurality of dangerous road sections), and is not suitable for popularization and use in a large range. Secondly, the existing risk assessment method is relatively difficult to obtain predicted parameters, has higher requirements on sensing equipment, has adverse factors of poor sensitivity to external environments, and is easy to cause false alarm. Disclosure of Invention In view of the disadvantages and shortcomings of the prior art, the invention provides a wet slippery road risk assessment method based on unmanned aerial vehicle identification vehicle track, which adopts a high-resolution unmanned aerial vehicle carrying cradle head mode, obtains a vehicle ground track sequence by extracting video data of vehicle running track of a dangerous road section through mapping, calculates parameter indexes of a normal sample and a wet slippery road according to a reference track curve, compares the wet slippery road with the normal sample to output distribution difference, comprehensively considers indexes such as proportion, duration, space consistency and the like on a weighting threshold to obtain the risk score of the vehicle running on the dangero