CN-120877516-B - Traffic running risk dynamic assessment method based on multi-source data fusion

Abstract

The invention discloses a traffic running risk dynamic assessment method based on multi-source data fusion, which belongs to the technical field of traffic running risk and comprises the steps of acquiring multi-source traffic data of a vehicle-mounted sensor, a road side unit, a meteorological data platform and a high-precision map in real time, generating a self-adaptive repulsive force field model by a dynamic parameter engine, calculating multi-obstacle coupling risk field intensity by a risk shielding compensation model, and calculating the multi-obstacle coupling risk field intensity by the risk shielding compensation model. According to the invention, through the cooperation of the vehicle-mounted sensor, the road side V2X unit, the meteorological data and the multi-layer data of the high-precision map, the perception limitation of the traditional single sensor is broken through, the advanced detection of the blind area obstacle and the comprehensive characterization of the road environment are realized, the perception hysteresis problem in an extreme scene is solved, the potential energy field gain is adjusted in real time based on dynamic parameters such as the road curvature, the road surface friction coefficient and the like, the model is self-adaptive to the risk characteristics of different road conditions, and the risk misjudgment of the traditional fixed parameter model is avoided.

Inventors

• WENG WENZHONG
• XIE YANGYANG
• LUAN SEN
• Yadekar Wumaier
• XU FAN
• LUO SHUAI
• MA YONGLIN
• ZHAO YONGHU
• TIAN YE
• WU XINGHUA
• ZHANG WENGUI

Assignees

• 新疆交投建设管理有限责任公司
• 安徽远航交通科技有限公司
• 北京工业大学

Dates

Publication Date

20260505

Application Date

20250730

Claims (5)

1. 1. The traffic running risk dynamic assessment method based on multi-source data fusion is characterized by comprising the following steps of: S1, acquiring multisource traffic data of a vehicle-mounted sensor, a road side unit, a meteorological data platform and a high-precision map in real time, wherein the multisource traffic data comprises a vehicle motion state, a blind area obstacle vector, visibility, a road friction coefficient and a road curvature; S2, dynamically adjusting potential energy field gain parameters according to visibility, road friction coefficient and road curvature through a dynamic parameter engine to generate a self-adaptive repulsive force field model; the dynamic adjustment of potential energy field gain parameters comprises the step of formulating a road potential energy field gain coefficient lambda Calculation, wherein R is the radius of curvature, k=0.01m 1 The curvature response coefficient is calibrated through real vehicle side deflection angle fitting, and mu is a road surface friction coefficient compensation factor; Introducing a dynamic parameter alpha of federal learning optimization into a gain coefficient of the velocity repulsive force field, and updating a velocity repulsive force field model through an actual measurement value m of a vehicle-mounted mass sensor; S3, calculating multi-obstacle coupling risk field intensity based on a risk shielding compensation model by fusing obstacle spatial distribution and relative motion direction, wherein the implementation of the risk shielding compensation model comprises calculating the obstacle spatial overlapping degree based on a rotation rectangular intersection ratio, and triggering compensation when IOU is more than 0.6; Identifying a shielding region by adopting a deformable attention mechanism of the DETR model, outputting a shielding relation probability matrix, and starting shielding coefficient correction when the shielding probability is more than or equal to 85%; Dividing a single lane into grid units of 20cm multiplied by 20cm, detecting pavement abnormality based on a roadside radar integrated machine, and generating a local risk increment matrix ; Optimizing edge computing node scheduling by adopting genetic algorithm, controlling cooperative delay of road side video stream processing and vehicle-mounted risk field computing within 80ms, solving transformation matrix ; The road side equipment installation deviation calibration comprises extracting a lane line cross point set in a high-precision map Detection point set of matching radar integrated machine Solving a transformation matrix T through a quaternion rotation matrix to ensure that the grid positioning error is less than or equal to 5cm; bend region risk enhancement, including computing centrifugal potential energy Normalized to risk value When the curvature radius R is less than 200m, the curvature response coefficient k=0.01m -1 is fused through a Bayesian network Superimposed to the total risk field.
2. 2. The traffic running risk dynamic assessment method based on multi-source data fusion according to claim 1, wherein the dynamic parameter engine dynamically adjusts multi-source data weights by using a bayesian network, and when vehicle-mounted and road side data conflict, the road side data weights are raised to more than 0.7 through posterior probability P (weight|observation).
3. 3. The traffic operation risk dynamic assessment method based on multi-source data fusion according to claim 1, wherein the multi-source data timing alignment comprises: doppler motion compensation is carried out on the laser radar point cloud, the position offset of the point cloud is delta d=V d multiplied by delta t, wherein, ; The vehicle-mounted IMU and the road side humidity sensor data are fused through Kalman filtering, the road surface friction coefficient mu is updated, and the updating formula is that 。
4. 4. The method for dynamically assessing traffic risk based on multi-source data fusion according to claim 1, wherein the weight allocation of the dynamic parameter engine comprises: When the visibility is less than 100M, the DETR model is compressed to 12M parameters through knowledge distillation, the laser radar weight is reduced to 0.3, and the V2X communication weight is increased to 0.7; Coefficient of friction compensation factor when road surface humidity >0.5 And the environment perception uncertainty is reduced by combining with the shannon information entropy theory.
5. 5. The traffic operation risk dynamic assessment method based on multi-source data fusion according to claim 1, wherein the edge computing architecture deployment comprises: The road side edge nodes execute point cloud distortion compensation and V2X data alignment, and the vehicle-mounted computing unit calculates a dynamic risk potential energy field in real time; the output layer generates a machine readable code comprising four tuples of risk levels, grid values, and waypoints, wherein the risk levels are stored using a 1-5 enumeration type.

Description

Traffic running risk dynamic assessment method based on multi-source data fusion Technical Field The invention relates to the technical field of traffic running risks, in particular to a traffic running risk dynamic assessment method based on multi-source data fusion. Background Vehicle operation risk assessment is a key element of Intelligent Transportation Systems (ITS) and autopilot technology, which provides safety precautions for vehicle decisions by quantifying potential risks in the traffic environment. The early evaluation method mainly relies on single sensor data (such as radar ranging) or a static model (such as a safe distance model), and has the defect of insufficient real-time performance when dealing with complex scenes such as multi-vehicle interaction, weather mutation and the like. Along with the development of the perception technology, the risk modeling method based on the potential field theory realizes risk visualization by simulating the physical field effect, but the prior art still has significant bottlenecks in the aspects of dynamic adaptability, multi-source data fusion, complex scene processing and the like: The traditional method only relies on vehicle-mounted sensors (such as a laser radar and a camera) to acquire local data, and lacks of cooperation of a road side V2X unit, a meteorological platform and a high-precision map, so that blind area obstacles (such as vehicles cut into from the rear side) are perceived with lagging, and particularly under extreme weather such as rain and fog, the detection range and the precision of a single sensor are greatly reduced, and the road environment cannot be comprehensively represented. For example, the dynamic risk potential energy field model proposed by chinese patent CN112002143B is based on vehicle-mounted sensor data only, without fusion of external environment data, and in a scene with a visibility lower than 100m, the response delay to a blind area target exceeds 200ms. The gain parameters (such as repulsive force field coefficient) of the existing potential field model are usually preset to be fixed values, and the influence of dynamic environment factors such as road friction coefficient, road curvature and the like on a risk field is not considered. For example, in a wet road (friction coefficient < 0.5) or small radius curve (radius of curvature R <200 m) scenario, the fixed parameter model cannot adaptively adjust risk sensitivity, resulting in a slow down advice lag or misjudgment, increasing collision risk. The system model calculates the risk field intensity of multiple obstacles in a linear superposition mode, and the influence of the spatial overlapping degree of the obstacles and the relative movement direction is not considered. When a large car shields a pedestrian, a multi-car serial rear-end collision and other scenes, the nonlinear transfer characteristics of risks cannot be accurately quantified. For example, prior art shading compensation mechanisms based on rotated rectangular intersection ratios (RotatedIOU) are lacking, and when the obstacle spatial overlap IOU >0.6, the risk field strength calculation error exceeds 30%. The existing method lacks an efficient multi-source data time sequence alignment mechanism, doppler motion compensation of laser radar point cloud is insufficient, and time stamp deviation of V2X data and sensor data is large (usually more than 100 ms), so that fusion data is distorted. In addition, the edge computing architecture is unreasonable to deploy, the vehicle-mounted computing force is too heavy to finish complex risk field computing within 80ms, and the real-time requirement of automatic driving cannot be met. Based on the above, the invention designs a traffic running risk dynamic assessment method based on multi-source data fusion to solve the above problems. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides a traffic running risk dynamic assessment method based on multi-source data fusion. In order to achieve the above purpose, the invention is realized by the following technical scheme: a traffic running risk dynamic assessment method based on multi-source data fusion comprises the following steps: S1, acquiring multisource traffic data of a vehicle-mounted sensor, a road side unit, a meteorological data platform and a high-precision map in real time, wherein the multisource traffic data comprises a vehicle motion state, a blind area obstacle vector, visibility, a road friction coefficient and a road curvature; S2, dynamically adjusting potential energy field gain parameters according to visibility, road friction coefficient and road curvature through a dynamic parameter engine to generate a self-adaptive repulsive force field model; And S3, based on a risk shielding compensation model, fusing the spatial distribution and the relative movement direction of the obstacle, and calculating the multi-obstacle coupling risk field intensi