Analysis Of Collision Accidents With Uncertainty Of Environment And Perception

The in-depth application of artificial intelligence algorithms in the field of autonomous driving has greatly enhaned the ability of smart cars to perceive complex scenes.However,due to the complexity of driving scenarios faced by smart cars and the uncertainty of artificial intelligence algorithms,artificial intelligence algorithms have brought new challenges to the driving safety of smart cars.Effectively analyze the risks that may be caused by the complex driving environment and the uncertainty of artificial intelligence algorithms in the driving process of smart cars,reduce the probability of accidents during the driving process of smart cars,improve the driving safety of smart driving,and finally achieve safety and intelligence drive.Therefore,the main research contents of this article are as follows:(1)There is a close relationship between different driving environments and the severity of traffic accidents.In order to minimize the severity of traffic accidents caused by high-risk complex scenes,this paper proposes a method to study the relationship between driving environment and accident severity.This article uses Kaggle’s opensource French traffic accident data set,first of all,features processing through data cleaning,and proposes a method of fusion of multiple feature selection methods for feature screening,to obtain a subset of environmental features that have an important impact on the severity of the accident.This completes the construction of a typical accident data set.Then,an order-regression method was used to construct an accident risk model.Based on the model,the significance,correlation and impact coefficient of environmental characteristics on the severity of the accident were analyzed,and several high-risk scenarios closely related to the severity of the accident were obtained.The scenario proposes corresponding strategies and recommendations.(2)In view of the fact that smart cars are prone to perception errors in severe weather and high-risk complex scenes and cause serious traffic accidents,this paper first discusses the perception uncertainty from the perspective of neural networks,and then analyzes many smart car traffic accidents and finds The accurate scene model can carry out an accurate driving area division,which can effectively avoid the problem of perceptual error,but semantic segmentation still has the uncertainty risk of neural network,so this paper aims at the uncertainty of driving area segmentation uncertainty.The uncertainty of semantic segmentation is studied.The shortcomings of the existing semantic segmentation uncertainty extraction algorithm are not efficient.A segmentation uncertainty evaluation algorithm combined with Bayesian deep learning is proposed.MC-Dropout technology is used to The semantic segmentation algorithm performs uncertainty sampling.The MC-Dropout sampling method is embedded in the key layer of Bayesian Segnet,which reduces the degree of single-sample network bloat,and the pyramid sampling structure is introduced to improve the uncertainty extraction effect of single-sample sampling.Experimental results show that compared with Bayesian Segnet,this method has better segmentation uncertainty extraction effect and higher efficiency(3)After completing the theoretical analysis and algorithm design,this paper applies the semantic segmentation uncertainty evaluation algorithm to the uncertainty evaluation problem of the driving area division,mainly considering the driving area of the smart car.The algorithm proposed in the fourth chapter of this paper gives its uncertainty results for each pixel,lacking the overall uncertainty evaluation results of the driving area segmentation,which is not conducive to the decision-making system combining local information and overall information in engineering applications.Comprehensive decision-making,therefore,this paper proposes an uncertainty index,and carries out high-risk pixel statistics for the drivable area,and then uses the proportion of high-risk pixels as the overall uncertainty assessment of the drivable area segmentation result.Finally,this paper builds a simulation environment,and tests the uncertainty evaluation algorithm and uncertainty index of the driving area segmentation results in different weather environments under the simulation environment.The results show that the pixellevel uncertainty evaluation algorithm designed in this paper It can effectively judge the reliability of the segmentation results.At the same time,the uncertainty degree proposed in this paper also has a clear distinction in the overall uncertainty assessment of the segmentation results of the drivable area.