The Research Of The Driving Mode Decision-making And The Safety Evaluating Of Intelligent Assistant Driving System

The development of autonomous vehicles provides effective solutions and opportunities for reducing the probability of traffic accidents,especially for the crashes which caused by the unreliability and instability of driver.However,due to technical limitations and economic and social challenges,achieving fully autonomous driving is still a long term endeavor.Thus,automatic driving and manual driving lives together to control the vehicle is an inevitable stage during the development of the intelligent vehicle,two subsystems which include automatic diving system and manual driving system would be co-existence in a long time.One principal research question is how to choose the suitable driving mode of an intelligent vehicle in the complex traffic environment.What's more,the evaluation of driving safety is also another indispensable step.The research of this thesis mainly focuses on the above topics.A driving-mode decision-making model was established based the driver's physiology indexes and the selected features from the multi-sensors information,and the evaluation model is established to assess the driving safety.The main research work is as follows:(1)The characteristics of the Intelligent Assistant Driving System which controlled by the driving and machine together are analyzed.Then,the simulation system and real vehicle experiment system are designed.The driver's physiology information acquisition system is designed based on the driving simulator and Biography Infiniti system.The vehicle information acquisition system is developed based on the characteristic of vehicle motion.In addition,the simulation system which could simulating the switching process from manual driving to warning assistance driving,or from manual driving to autonomous driving based on the human-computer interaction requirements.Moreover,the related experiments are carried out to collect multi-sensors information data.(2)The driving risk status identification model is proposed based on the characteristics of driver's physiology indexes during driving process.The relationship between drivers' physiological characteristics and driving risk status has been studied in previous studies,and the most significant impact factors(blood volume pulse(BVP),and Skin Conductance(SC))associated with the driving risk status is extracted using the K-means cluster method.The results show that the proposed method can identify the driving risk status reliability and accurately.(3)A feature selection algorithm is proposed to extract the impact factors which significantly associated to the driving mode decision-making of intelligent vehicle.The correlation and redundancy between features and classes is analyzed based on the theory of information gain,and an improved Markov blanket(MB-NEW)is proposed.The condition of maximum mutual information(CMIM)and the boundary threshold is introduced to improve the redundancy and the convergence speed in MB-NEW.In addition,considering with the precision of feature selection,the fusion of information gain and multiple classification method is analyzed.These two feature selection methods are extracted the SD of the front wheel angle,driver experience,vehicle speed,headway time,acceleration,and the distance to centre lane had significant impacts on the driving mode decision-making.Moreover,the analysis result of FARS system shows that the MB-NEW method has higher precision and execution efficiency for the huge amounts of data.(4)Driving modes were classified into three states based on the drivers' self-reported records,and two physiological indexes using the K-mean cluster method was adopted to calibrate the reported driving modes.And the driving mode decision-making model is established based on the multi-class support vector machine(M-SVM)algorithm and genetic algorithm(GA).The results shows that the proposed method has higher classification precision compared with other pattern recognition methods.(5)The driving safety evaluation method is studied in different driving mode of intelligent vehicle.The proposed method is applied based on the aspects of state of perception,decision and judgment,operation.Because of the uncertain of the evaluation result,the Bayesian networks(BNs)is introduced to develop the evaluation model.The sensitivity and reliability analysis result suggests that the proposed method is reasonable and suitable.