CN-117087674-B - Early warning method and system based on road risk and driver behavior

CN117087674BCN 117087674 BCN117087674 BCN 117087674BCN-117087674-B

Abstract

The invention provides an early warning method and system based on road risk and driver behavior, wherein the method comprises the steps of obtaining current road environment information, judging the current road environment information based on a preset road risk level model, and obtaining a road risk level; the method comprises the steps of obtaining driver behavior information, judging the driver behavior information based on a preset driver behavior detection model to obtain a driver behavior safety level, fusing the road risk level and the driver behavior safety level to obtain a driving risk prediction result, and outputting an early warning signal according to the driving risk prediction result. According to the invention, the behavior of the driver in the vehicle and the road risk are fused through the deep learning model, so that the accurate prediction and the corresponding reminding of the running risk can be more comprehensively carried out, and the running safety is improved.

Inventors

CHEN YUNXING
CUI JUNHUA
WU ZHAO
WU HUAWEI
WEI CHANGYIN

Assignees

湖北文理学院

Dates

Publication Date: 20260505
Application Date: 20230907

Claims (6)

1. An early warning method based on road risk and driver behavior is characterized by comprising the following steps: Acquiring current road environment information, and judging the current road environment information based on a preset road risk level model to obtain a road risk level; Acquiring driver behavior information, judging the driver behavior information based on a preset driver behavior detection model, and obtaining a driver behavior safety level; Fusing the road risk level and the driver behavior safety level to obtain a driving risk prediction result; outputting an early warning signal according to the driving risk prediction result; The preset driver behavior detection model uses a trained pre-trained swin transformer as an encoder and captures the context linkage of input data based on a long-term and short-term memory neural network; The method comprises the steps of introducing variability convolution into the encoder, adding a full-connection layer and a normalization function to the output of the long-short-period memory neural network, classifying the driver behaviors according to the normalization function, and carrying out regression prediction by using a linear layer to obtain a prediction result of the driver behaviors; Fusing the road risk level and the driver behavior safety level based on a preset prediction model, wherein the process of obtaining the driving risk prediction result comprises the following steps: respectively inputting the road risk level and the driver behavior safety level into different branches of a preset prediction model to obtain road risk level characteristics and driver behavior safety level characteristics; fusing the road risk level characteristics and the driver behavior safety level by utilizing a fusion layer of the preset prediction model to obtain fusion characteristics; inputting the fusion characteristics to a prediction result layer to obtain a risk prediction result; the method for establishing the preset road risk level model comprises the following steps: Acquiring a scene image dataset of a target road, wherein the scene image dataset comprises weather data, time data and traffic condition data matched with the target road; Acquiring traffic accident information corresponding to the target road based on a traffic accident database, and obtaining target traffic accident information, wherein the traffic accident information comprises accident time data, accident place data, accident type data, accident severity data, accident vehicle data and weather condition data; Labeling the positions and the categories of the road condition information of the scene image data sets to obtain a first scene image data set, wherein the road condition information comprises traffic signs, crosswalks, vehicles, pedestrians and traffic lights; Matching the target traffic accident information with the scene image data set to obtain a road risk level database remarking the target traffic accident information; dividing road risk grades based on traffic accident influencing factors; constructing a road risk model based on a neural risk network based on the road risk level database and the road risk level; The matching of the target traffic accident information and the scene image data set is carried out to obtain a road risk level database remarking the target traffic accident information, which is specifically as follows: and matching the target traffic accident information with the scene image dataset by adopting a KNN nearest neighbor matching algorithm to obtain a road risk level database remarking the target traffic accident information.
2. The method of claim 1, wherein the traffic accident influencing factors include traffic accident information, traffic flow, road type, road segment characteristics, traffic lights, weather conditions, and visibility of view.
3. The pre-warning method based on road risk and driver behavior according to claim 1, wherein the road risk model is constructed based on a BP neural network.
4. The road risk and driver behavior based warning method of claim 1, wherein the road risk level includes a high risk road, a medium risk road, and a low risk road, and the driver behavior safety level includes safe driving and unsafe driving; The risk prediction results include unsafe driving and high risk roads, unsafe driving and medium risk roads, unsafe driving and low risk roads, safe driving and high risk roads, safe driving and medium risk roads, and safe driving and low risk roads.
5. The method for warning based on road risk and driver behavior according to claim 4, wherein the outputting the warning signal according to the driving risk prediction result comprises: under the conditions of unsafe driving and high-risk roads, outputting an emergency voice warning signal and displaying a red warning signal; under the conditions of unsafe driving and moderate risk roads, outputting voice warning information and displaying yellow warning signals; Under the conditions of unsafe driving and low risk roads, outputting voice early warning information; under the conditions of safe driving and high risk roads, outputting a voice road early warning signal and displaying a red reminding signal; under the conditions of safe driving and medium risk road, outputting a voice road warning signal and displaying a yellow warning signal; And displaying a reminding signal under the conditions of safe driving and low-risk roads.
6. An early warning system based on road risk and driver behavior, adapted to the early warning method based on road risk and driver behavior according to any one of claims 1 to 5, comprising: The road risk level module is used for acquiring current road environment information, judging the current road environment information based on a preset road risk level model, and obtaining a road risk level; The driver behavior information module is used for acquiring driver behavior information, judging the driver behavior information based on a preset driver behavior detection model, and obtaining a driver behavior safety level; the prediction module is used for fusing the road risk level and the driver behavior safety level to obtain a driving risk prediction result; And the output module is used for outputting an early warning signal according to the driving risk prediction result.

Description

Early warning method and system based on road risk and driver behavior Technical Field The invention relates to the technical field of vehicle control, in particular to an early warning method and system based on road risks and driver behaviors. Background With the development of society, urban buses and private cars become important transportation means for people to travel, so that safe driving of vehicles becomes a part of important attention in the transportation field. The driving danger prediction is used as a key part for realizing an intelligent vehicle active safety system, and the existing early warning variables with wider application mainly comprise workshop time, collision time, workshop distance and the like. In practice, the whole process from the formation of the driving risk to the occurrence of the risk conflict is difficult to mold by adopting a single early warning parameter, and more complex and comprehensive models are required to be adopted for research. Meanwhile, the existing early warning model usually only considers the running characteristics of the vehicle, but ignores the influence of real-time driver behaviors, road and environment changes on the running risk state, cannot comprehensively describe the internal evolution rule among the running states, and is not beneficial to accurate judgment and precise prediction of the running risk. Therefore, how to perform fusion analysis on the driving environment and accurately judge and accurately predict the driving risk becomes a problem which needs to be solved by staff in the current technical field. Disclosure of Invention In view of the foregoing, it is necessary to provide a method and a system for early warning based on road risk and driver behavior, so as to achieve the purposes of performing fusion analysis on driving environment and driving environment, and accurately judging and accurately predicting driving risk. In order to achieve the above object, the present invention provides an early warning method based on road risk and driver behavior, comprising: Acquiring current road environment information, judging the current road environment information based on a preset road risk level model, and obtaining a road risk level; acquiring driver behavior information, judging the driver behavior information based on a preset driver behavior detection model, and obtaining a driver behavior safety level; fusing the road risk level and the driver behavior safety level to obtain a driving risk prediction result; And outputting an early warning signal according to the running risk prediction result. In one possible implementation manner, the method for establishing the preset road risk level model includes: Acquiring a scene image data set of a target road, wherein the scene image data set comprises weather data, time data and traffic condition data matched with the target road; Acquiring traffic accident information corresponding to a target road based on a traffic accident database, and obtaining target traffic accident information, wherein the traffic accident information comprises accident time data, accident place data, accident type data, accident severity data, accident vehicle data and weather condition data; labeling the positions and the categories of the road condition information of the scene image data sets to obtain a first scene image data set, wherein the road condition information comprises traffic signs, crosswalks, vehicles, pedestrians and traffic lights; matching the target traffic accident information with the scene image data set to obtain a road risk level database through remarking the target traffic accident information; dividing road risk grades based on traffic accident influencing factors; and constructing a road risk model based on the neural risk network based on the road risk level database and the road risk level. In one possible implementation manner, the target traffic accident information and the scene image dataset are matched to obtain a road risk level database through remarking the target traffic accident information, specifically: And matching the target traffic accident information with the scene image data set by adopting a KNN nearest neighbor matching algorithm to obtain a road risk level database for remarking the target traffic accident information. In one possible implementation, the traffic accident influencing factors include traffic accident information, traffic flow, road type, road segment characteristics, traffic lights, weather conditions, and visibility of view. In one possible implementation, the road risk model is built based on a BP neural network. In one possible implementation, the pre-driver behavior detection model uses trained pre-trained swin transformer as an encoder and captures contextual associations of input data based on a long-term memory neural network; The method comprises the steps of introducing variability convolution into the encoder, adding a full-connectio