Abstract:

The accelerating behavior characteristics of the driver were studied by using the naturalistic driving data (NDD) to improve the human-like driving ability of the intelligent vehicle. The database composed of approximately 58 million observation data was applied to discuss the convergence of the acceleration
distribution of the driver. The multivariate kernel density function was used to achieve the acceleration distribution of the driver. The relative entropy (Kullback-Leibler divergence) was employed to describe the distinction among the datasets which were composed of different amount of data. The acceleration distribution
of the driver was analyzed, and its parameters were achieved by using the convergent dataset. The application of the acceleration distribution characteristics of the driver in intelligent vehicle were discussed. The results show that the 2-dimensional distribution between the longitudinal acceleration and lateral acceleration follows the dual triangle distribution pattern. The longitudinal acceleration and lateral acceleration firstly increase and then decrease when the velocity increases. The acceleration distribution can be applied in the intelligent vehicle driving capability evaluation, intelligent vehicle safety test, co-driving control,   risk  assessment algorithm, etc.

Key words: Key words , automobile engineering , driving behavior , naturalistic driving data (NDD) , acceleration distribution;