Semantic Learning and Understanding of multivehicle interaction patterns Using Primitive Driving Patterns With Bayesian Nonparametric Approaches

The development of an automated vehicle that can handle complex driving scenarios and appropriately interact with other road users requires semantic learning and the ability to understand the driving environment, usually based on the analysis of a large amount of natural driving data. However, the explosive growth of driving data poses a huge challenge for extracting primitive driving patterns from long-term multi-dimensional time series traffic scene data, which involves multi-scale road users. In order to achieve this, a general framework to gain insights into intricate multi-vehicle interaction patterns in real-world driving was presented in this paper. A Bayesian nonparametric learning method based on a hierarchical Dirichlet process hidden semi-Markov model (HDP-HSMM) is introduced to extract primitive driving patterns from time series driving data without prior knowledge of the number of these patterns. Unlike previous articles, which only considers the interactive behavior of two vehicles, we consider the driving scenarios where the ego vehicle can sense all surrounding vehicles, including the front vehicle, the rear vehicle, the front left vehicle, the left vehicle, the rear left vehicle, the front right vehicle, the right vehicle, the rear right vehicle. Experimental results show that our proposed method can extract primitive driving patterns, thereby providing a semantic way to analyze multi-vehicle interaction patterns from multi-dimensional driving data and laying the foundation for the generation of coverage test cases for automated vehicles.