A Vehicle Longitudinal Driving Assistance System Based On Self-learning Method Of Driver Characteristics

With important significance for the road traffic safety, the vehicle longitudinal driving assistance systems have been developed as effective technologies for driver workload mitigation and rear-end collision avoidance. Because of the driver's individual diversity and state fluctuation, the traditional assistance systems with fixed algorithm parameters cannot provide enough adaptability for different drivers. To solve this problem, an upper control strategy of vehicle longitudinal driving assistance system and an online self-learning method of driver characteristics are proposed based on the analysis of driver car-following experimental data and aberrant behavior questionnaire. Based on the control strategy and method, a driving assistance system prototype is developed with functions of Adaptive Cruise Control (ACC) and Forward Collision Warning/Avoidance (FCW/FCA), and has good adaptability to the drivers.An analysis methodology of the driver car-following behavior and aberrant behavior characteristics is proposed firstly. Real traffic experiments and driver behavior questionnaires are carried out to obtain driver characteristics information. Based on the experimental data analysis, the questionnaire factor quantification and cluster analysis, the driver steady car-following characteristics for ACC function and the driver approaching characteristics and aberrant driving tendency factor for FCW/FCA function are extracted. The analysis results are considered as the foundation of the system control strategy design.The upper control strategy of the vehicle longitudinal driving assistance system is established and the self-learning method of driver characteristics is presented subsequently. The upper control strategy consists of a driver car-following model and a forward collision warning/avoidance algorithm for the ACC and FCW/FCA functions respectively. Based on the Recursive Least Square (RLS) method with forgetting factor, the parameters of the driver model are real-time identified from the data sequences collected during the driver manual operation state, and the identification results are applied during the system ACC control state. The parameters of the FCW/FCA algorithm are matched based on the neural network classifier of aberrant driving behavior pattern and the statistics of the driver collision avoidance action data.To validate the effectiveness of the proposed method and system functions, an experimental platform of driving assistance systems with modular structure is constructed. Several key technologies including electro-hydraulic braking system, radar processing algorithm, controller hardware, human machine interface and CAN bus are developed to improve the reliability and flexibility of the platform.Furthermore, the performance of the self-learning method and system functions are investigated by experiments in real traffic. The experimental results show that the self-learning algorithm is effective and the system has good performance of adaptability to driver characteristics.