Research On Relationship Between Driver’s Driving Performance And Driving Safety

As one of the cornerstones of our national industry,the automobile industry has come a long way in recent years and is heading to the target of network based conection,intellectualization and energy conservation.However,what is behind the prosperity is the fact that the number of traffic accident is mataining at a high level.Within the trinity of “human-vehicle-road” traffic environment,the human factor accounts for a much larger proportion of the causation of traffic accident than the other two factors.This research starts with the driving behavior of human driver and performs the research over the internal relationship between driving performance and driving safety.This research establishes a quantitative mapping model from driving performance to the driving safety,which provides the solid proofs for the driving authority transition,and,proposes a warning method based on the driving performance to help with the transition of driving authority.In addition,this study also investigates a human-like steering model based on the deep learning.The main contribution and effort of this study is listed below.(1)Based on the European Natualistic Driving Study project,UDRIVE,this study evaluates the driving performance subset and driving safety subset with the help of batch test method and paralle computing method deployed in muti-workstation computation clusters.It also adoptes a K-means cluster analysis to recognize the distribution of driving performance in mortorway,investigates the difference of accident\inccident frequncy between different driving performance clusters,constructes a muiltidimensional mapping model to connect the driving performance and actual driving safety.(2)During the study,a data acquisition architecture is proposed.The human driver in-the-loop data acquisition system is constructed and the data from muiltiple resources that output at different frequency is synchronized.This research also designs a driver inthe-loop experiment with the workload varied in muiltiple level(indused by auditory-verbal secondary task).(3)The mutual depandance between driving performance and driving safety is quantatatively evaluated with Pearson’s Correlation Coefficient and Maximal Information Coefficent.The frequency distribution of Safety Critical Event is decomposed with isometric drving performance percentile histogram.With the depandance discovered above,those driving performance measures with high correlation with dirving safety are selected as the features of K-means cluster analysis.Driving performance subtypes are recognized and the corresponding risk levels were compared.(4)Three types of neural networks are adopted to compare the learning performance regarding the human steering behavior.What’s more,the hyper parameters of Long Short-Term Memory neural network in different level is tested to find out the corresponding influence.Finally,an optimized humanlike driver model is constructed and a driving authority transition system based on the driving performance and human-like driver model is proposed.