CN-121982900-A - Track prediction-based ice and snow road surface vehicle lane change risk dynamic assessment system

CN121982900ACN 121982900 ACN121982900 ACN 121982900ACN-121982900-A

Abstract

The invention relates to the technical field of intelligent driving safety, in particular to a track-prediction-based dynamic evaluation system for the track-change risk of vehicles on ice and snow roads, which comprises a prediction sampling module, a track analysis module and a track analysis module, wherein the prediction sampling module is used for acquiring the probability density distribution of the future position of a target vehicle and extracting track sampling points by using a Markov chain Monte Carlo method; the system comprises a risk field construction module, a binary conflict calculation module, a comprehensive risk assessment module and a real-time total conflict field generation module, wherein the risk field construction module is used for calculating the risk field intensity of each sampling point based on the equivalent mass of a vehicle, a manual occupation model and a future time discount weight function and aggregating to form a vehicle risk field, the binary conflict calculation module is used for matching the vehicle risk fields of a self vehicle and surrounding vehicles and calculating the sum of intensity products in an overlapping area to obtain a binary conflict field, and the comprehensive risk assessment module is used for superposing all the binary conflict fields to generate the real-time total conflict field. The system realizes the fine quantitative evaluation of the dynamic and multi-vehicle interaction risk in the road changing process of the ice and snow road surface.

Inventors

ZHAO WEI
GUO NIANCHENG
GAO YANG
CHENG HAO

Assignees

山东大学

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. Track prediction-based ice and snow road surface vehicle lane change risk dynamic assessment system is characterized by comprising: The prediction sampling module is used for acquiring the position probability density distribution of the target vehicle at the future moment, and extracting a series of future track sampling points from the position probability density distribution of the target vehicle at the future moment by using a Markov chain Monte Carlo method; The risk field construction module is used for calculating a risk field intensity value generated in space for each future track sampling point based on the equivalent mass of the vehicle, the manual occupation model and the future time discount weight function, and aggregating to form a vehicle risk field representing the risk intensity of the vehicle at each future point in space; The binary conflict calculation module is used for carrying out space matching on a vehicle risk field of the own vehicle and a vehicle risk field of the surrounding vehicle when risk interaction of the own vehicle and the surrounding vehicle is evaluated, calculating the sum of products of the vehicle risk field intensity value of the own vehicle and the vehicle risk field intensity value of the surrounding vehicle in an overlapping space area, and obtaining a binary conflict field representing conflict intensity of the own vehicle and the single surrounding vehicle; And the comprehensive risk assessment module is used for respectively calculating the binary conflict fields of the own vehicle and each surrounding vehicle in the interaction scene of the own vehicle and the plurality of surrounding vehicles, and superposing all the binary conflict fields to generate a real-time total conflict field reflecting the comprehensive risk faced by the own vehicle in the whole lane change process.
2. The track prediction-based dynamic estimation system for risk of vehicle lane change on icy or snowy road according to claim 1, wherein the acquiring the position probability density distribution of the target vehicle at the future time comprises: Acquiring vehicle motion characteristics and vehicle interaction characteristics of historical continuous moments of the self-vehicle and surrounding vehicles, wherein the vehicle motion characteristics comprise a speed instantaneous value, an acceleration instantaneous value, plane position coordinates and a vehicle side deflection angle, and the vehicle interaction characteristics comprise relative speeds and relative positions of the self-vehicle and each surrounding vehicle; Performing time alignment and dimension splicing on the vehicle motion features and the vehicle interaction features at the historical continuous moments to form a time sequence feature input matrix containing multi-dimensional features of a plurality of vehicles in continuous time steps; And processing the time sequence feature input matrix through an environment attention network, extracting time sequence dependency relationship, space dependency relationship and interaction attention relationship among vehicles, and predicting the position probability density distribution of the target vehicle at the future moment based on the time sequence dependency relationship, the space dependency relationship and the interaction attention relationship.
3. The track prediction-based dynamic estimation system for road change risk of vehicles on icy and snowy roads according to claim 2, wherein the processing the time series feature input matrix through the environmental attention network extracts a time-series dependency relationship, a spatial dependency relationship and an interactive attention relationship among vehicles, and predicts a position probability density distribution of a target vehicle at a future time based on the time-series dependency relationship, the spatial dependency relationship and the interactive attention relationship comprises: Processing the time sequence feature input matrix by using a long-short-term memory encoder, wherein the long-short-term memory encoder encodes a historical track sequence of each vehicle, extracts and outputs a coding feature vector of each vehicle at each historical moment, and the coding feature vector contains the time sequence dependency relationship of the vehicle; Constructing a graph structure by taking the coding feature vectors of all vehicles at the same moment as nodes and using the spatial distance relation among the vehicles, inputting the graph structure into a graph attention network, and generating an interaction attention weight matrix reflecting the relative importance among the vehicles by the graph attention network through adaptively calculating the attention weights among the nodes, and outputting interaction feature vectors containing the interaction attention relation among the vehicles; Constructing a three-dimensional social tensor taking a target vehicle as a center based on the interaction feature vector, wherein two plane dimensions of the three-dimensional social tensor represent a space grid, and channel dimensions are filled with the interaction feature vectors of other vehicles in the corresponding space grid; Inputting the three-dimensional social tensor into a convolution social pool network with an extrusion excitation structure, extracting local spatial modes around a target vehicle by the convolution operation of the convolution social pool network with the extrusion excitation structure, weighting different spatial characteristic channels by utilizing a channel attention mechanism, and outputting spatial context characteristics containing the spatial dependency relationship between the target vehicle and the surrounding environment; And fusing the final hidden state corresponding to the time sequence dependency relationship, the interaction feature vector corresponding to the interaction attention relationship and the space context feature corresponding to the space dependency relationship, inputting the fused state into a combined structure of a long-short-term memory decoder and a mixed density network, wherein the long-short-term memory decoder is responsible for decoding a future time sequence, the mixed density network processes the output of each future time of the decoder, estimates the two-dimensional Gaussian mixture distribution parameters of the planar position of a target vehicle at the future time, and the two-dimensional Gaussian mixture distribution at all future times jointly form the position probability density distribution at the future time.
4. A track prediction based dynamic road change risk assessment system for vehicles on icy or snowy roads according to claim 3, wherein extracting a series of future track sampling points from the position probability density distribution at the future time using a markov chain monte carlo method comprises: initializing a Markov chain, wherein the initial state of the Markov chain is a future track point sequence randomly extracted from the position probability density distribution of the future moment; Defining a proposed distribution for generating a new candidate future track point sequence from the current Markov chain state according to the parameters of the two-dimensional Gaussian mixture distribution; Calculating joint probability density of the candidate future track point sequence under the current position probability density distribution of the future moment, comparing the joint probability density with the joint probability density of the current state, and determining whether to accept the candidate future track point sequence as a new state of a Markov chain according to a preset acceptance criterion; repeating the steps from defining the proposed distribution to deciding whether to accept the candidate state until the state transition of the markov chain reaches a preset number of samples, the number of samples ensuring that the chain reaches a smooth distribution; And extracting states from the Markov chain after reaching stable distribution according to preset intervals, taking each extracted state as a complete future track sampling point, and finally obtaining a group of sample sets for approximating the position probability density distribution of the future moment.
5. The track prediction based dynamic road change risk assessment system for vehicles on icy and snowy roads according to claim 4, wherein the calculating the risk field strength value generated in space for each future track sampling point based on the vehicle equivalent mass, the manual occupancy model and the future time discount weight function comprises: The corresponding vehicle equivalent mass is obtained through table lookup according to the physical parameters of the target vehicle, and the vehicle equivalent mass integrates the actual mass and dynamic characteristics of the vehicle; According to the manual occupation model, expanding the plane position of the vehicle represented by the future track sampling point into a polygonal space area occupied by the vehicle at the plane position, wherein the manual occupation model defines a safety boundary of vehicle contour expansion; Defining a future time discount weight for each predicted future time, wherein the future time discount weight is monotonically decreased along with the delay of the predicted time, and the influence weight of the future time with longer time on the current risk is smaller; multiplying the equivalent mass of the vehicle, the future time discount weight and the probability value of the future track sampling point per se for each space grid point in the polygonal space region to obtain a risk field strength basic value of the space grid point contributed by the future track sampling point; And traversing all sampling points in the future track sampling point sample set, and accumulating the risk field intensity basic values contributed to the future track sampling points by all the future track sampling points for the same space grid point to obtain the final risk field intensity value of the space grid point, wherein the risk field intensity values of all the space grid points jointly form continuous risk intensity space distribution, namely the vehicle risk field.
6. The system for dynamically evaluating the risk of vehicle lane change on icy and snowy road surface according to claim 5, wherein the calculating the sum of the products of the vehicle risk field intensity values of the own vehicle and the vehicle risk field intensity values of the surrounding vehicles in the overlapping space region, to obtain the binary collision field representing the collision intensity of the own vehicle and the single surrounding vehicles, comprises: Mapping the vehicle risk fields of the own vehicle and the surrounding vehicles onto the same high-resolution space discrete grid, so that the two vehicle risk fields have the same space coordinate definition and grid cell division; Traversing each grid cell of the space discrete grid, reading the intensity value of the grid cell in the vehicle risk field of the own vehicle, and simultaneously reading the intensity value of the grid cell in the vehicle risk field of surrounding vehicles; Multiplying the read own vehicle risk field strength value with surrounding vehicle risk field strength values to obtain a collision strength contribution value on the grid unit; judging whether the conflict intensity contribution value on the grid unit is larger than zero, if so, marking the grid unit as a risk field overlapping area; And carrying out accumulation summation on the conflict intensity contribution values of all grid cells marked as the overlapping area of the risk fields, wherein the obtained accumulation summation is the scalar value of the binary conflict field, and the scalar value quantitatively describes the total conflict risk of the own vehicle and surrounding vehicles in the space occupation.
7. The dynamic evaluation system for road change risk of vehicles on icy and snowy road surface based on track prediction according to claim 6, wherein the steps of spatially matching the vehicle risk field of the own vehicle with the vehicle risk field of the surrounding vehicles, calculating the sum of products of the vehicle risk field intensity value of the own vehicle and the vehicle risk field intensity value of the surrounding vehicles in the overlapping space region, and obtaining the binary collision field representing the collision intensity of the own vehicle and the single surrounding vehicles, further comprise the steps of implementing continuous space integration by monte carlo integration approximation: Randomly generating a specified number of sampling points in a space region where a self-vehicle overlaps with a risk field of surrounding vehicles, wherein the space coordinates of the sampling points follow the uniform distribution of the space region; For each randomly generated sampling point, respectively calculating the intensity value of the sampling point in the vehicle risk field of the own vehicle and the intensity value of the sampling point in the vehicle risk field of surrounding vehicles by a bilinear interpolation method; Multiplying the calculated own vehicle risk field intensity value by surrounding vehicle risk field intensity values to obtain the contribution of the sampling point to the collision intensity; calculating the average value of all random sampling point contributions, multiplying the average value by the area of the space region overlapped by the risk field to obtain an approximate estimated value of continuous space integration, wherein the approximate estimated value is the scalar value of the binary conflict field.
8. The dynamic evaluation system for road changing risk of vehicles on icy and snowy road surface based on track prediction according to claim 7, wherein the calculating the binary conflict fields of the own vehicle and each surrounding vehicle respectively and superposing all the binary conflict fields to generate a real-time total conflict field reflecting the comprehensive risk faced by the own vehicle in the whole road changing process comprises: Calculating a binary conflict field scalar value by adopting a product summation method or a Monte Carlo integration method according to a vehicle risk field of the own vehicle and a vehicle risk field of surrounding vehicles aiming at each surrounding vehicle in the current perception range of the own vehicle; the binary conflict field scalar values of the self-vehicle and each surrounding vehicle obtained through calculation are stored according to the vehicle identification; Executing arithmetic accumulation operation on all stored binary conflict field scalar values at the current moment, wherein the obtained sum value is the real-time total conflict field scalar value at the current moment; And repeatedly executing calculation and accumulation operations along with time advancing and scene updating to generate a real-time total conflict field scalar value sequence changing along with time, wherein the real-time total conflict field scalar value sequence forms the real-time total conflict field reflecting the dynamic evolution of risks.
9. The track prediction based dynamic road change risk assessment system for vehicles on icy and snowy roads according to claim 8, further comprising: The dynamic risk updating module repeatedly generates a real-time total conflict field at intervals of a preset fixed time period, and dynamically updates and outputs a risk state for a continuously-changing channel scene, and specifically comprises the following steps: Setting a system updating period, and synchronously acquiring the latest sensor data of the own vehicle and all surrounding vehicles when each updating period starts; updating the vehicle motion characteristics and vehicle interaction characteristics based on the latest sensor data, and refreshing the time series characteristic input matrix according to the updated vehicle motion characteristics and vehicle interaction characteristics; using the refreshed time sequence characteristic input matrix to drive the environment attention network to execute track prediction again, and generating updated position probability density distribution at the future moment; Based on the updated position probability density distribution at the future time, re-executing all the processes of Markov chain Monte Carlo sampling, vehicle risk field construction, binary conflict field calculation and real-time total conflict field generation; And outputting the real-time total conflict field scalar value calculated in the current updating period and the binary conflict field scalar values of the own vehicle and each key surrounding vehicle.
10. The track prediction-based dynamic estimation system for risk of vehicle lane change on icy and snowy road surface according to claim 9, wherein the instantaneous acceleration value in the vehicle motion characteristics adopts the longitudinal acceleration and the lateral acceleration corrected by the ice and snow road surface adhesion coefficient, and the correction process comprises: acquiring an estimated ice and snow attachment coefficient of the current road surface through a vehicle-mounted sensor or a road surface state identification module; Reading an original longitudinal acceleration request and an original transverse acceleration request output by a vehicle dynamics controller; multiplying the original longitudinal acceleration request by the estimated ice and snow attachment coefficient to obtain a safe longitudinal acceleration upper limit under the ice and snow road surface, and adopting the safe longitudinal acceleration upper limit as an actual longitudinal acceleration instantaneous value if the original request exceeds the safe longitudinal acceleration upper limit; multiplying the original lateral acceleration request by the estimated ice and snow attachment coefficient to obtain a safe lateral acceleration upper limit under the ice and snow road surface, and adopting the safe lateral acceleration upper limit as an actually used lateral acceleration instantaneous value if the original request exceeds the safe lateral acceleration upper limit; And the upper limit value of the safe longitudinal acceleration and the upper limit value of the safe transverse acceleration which are subjected to upper limit constraint processing are used as final acceleration instantaneous values to be input into the time sequence characteristic input matrix.

Description

Track prediction-based ice and snow road surface vehicle lane change risk dynamic assessment system Technical Field The invention relates to the technical field of intelligent driving safety, in particular to a track-prediction-based dynamic evaluation system for road changing risks of vehicles on ice and snow roads. Background The existing vehicle lane change risk assessment method usually adopts a deterministic model or a model based on simplified probability distribution in a track prediction stage. These methods output one or more most likely trajectories, or describe the position uncertainty using simple models such as gaussian distributions, which do not fully exploit the complex, non-gaussian probability distribution characteristics that result from driving behavior and road conditions. This makes the prediction result not fully characterise the true probability structure of the space that the vehicle may occupy in the future, resulting in a congenital deviation of the input of the subsequent risk assessment. In the risk assessment stage, conventional techniques rely mostly on geometric relationships between vehicles or models based on physical collision energy to construct risk indicators. The risk field constructed by the method is usually static or only considers the instant physical state, ignores the dynamic influence of the behavior style and the intention of the driver on the risk distribution, and also lacks a mechanism for reasonably compromising the long-term risk event. This risk expression is deficient in reflecting the cognitive characteristics of the driver's personalized risk decisions and the risk decays over time. The vehicle is easy to sideslip, the braking distance is increased, the control response is delayed and other dynamic characteristic changes due to low attachment coefficient, meanwhile, the driving behavior of the driver under the ice and snow working condition is more conservative or has delayed operation, the uncertainty of the vehicle track is further aggravated, the existing track prediction certainty or simplified probability model and the static risk assessment mode are adopted, the prediction deviation of the future track of the vehicle can be greatly increased under the special working condition of the ice and snow road surface, the accuracy and the instantaneity of the risk assessment are more difficult to meet the requirements of intelligent driving lane change safety, and the limitation of the technology is particularly outstanding in the lane change scene of the ice and snow road surface with low attachment. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a track prediction-based dynamic evaluation system for the road changing risk of an ice and snow road vehicle. In order to achieve the purpose, the invention adopts the following technical scheme that the track prediction-based ice and snow road surface vehicle lane change risk dynamic assessment system comprises: The prediction sampling module is used for acquiring the position probability density distribution of the target vehicle at the future moment, and extracting a series of future track sampling points from the position probability density distribution of the target vehicle at the future moment by using a Markov chain Monte Carlo method; The risk field construction module is used for calculating a risk field intensity value generated in space for each future track sampling point based on the equivalent mass of the vehicle, the manual occupation model and the future time discount weight function, and aggregating to form a vehicle risk field representing the risk intensity of the vehicle at each future point in space; The binary conflict calculation module is used for carrying out space matching on a vehicle risk field of the own vehicle and a vehicle risk field of the surrounding vehicle when risk interaction of the own vehicle and the surrounding vehicle is evaluated, calculating the sum of products of the vehicle risk field intensity value of the own vehicle and the vehicle risk field intensity value of the surrounding vehicle in an overlapping space area, and obtaining a binary conflict field representing conflict intensity of the own vehicle and the single surrounding vehicle; And the comprehensive risk assessment module is used for respectively calculating the binary conflict fields of the own vehicle and each surrounding vehicle in the interaction scene of the own vehicle and the plurality of surrounding vehicles, and superposing all the binary conflict fields to generate a real-time total conflict field reflecting the comprehensive risk faced by the own vehicle in the whole lane change process. As a further aspect of the present invention, the acquiring a position probability density distribution of a future time of a target vehicle includes: Acquiring vehicle motion characteristics and vehicle interaction characteristics of historical continuous mome