Dangerous Behavior Recognition of Autonomous Vehicles at Intersection Based on Gaussian Mixture Model

Abstract

Urban road intersections are a typical complex traffic scene, and they are also prone to road traffic accidents. Aimed at the safety problems of road intersections, this paper proposes a model of vehicle dangerous behavior recognition based on Gaussian mixture model method, which provides an empirical basis for autonomous vehicle behavior decision at intersections. By analyzing the characteristics and rules of driving behavior at intersections, we select parameters like PET as input parameters of the model, and then construct a simulation model based on Beijing’s Weigongcun intersection, which is used to obtain the dataset in intersection scene. The obtained data was subject to Kalman filtering and normalized as data input of a Gaussian mixture model. Dangerous scenes and general scenes at the intersection scene were screened out. The screening results were analyzed and evaluated, and proved to be reliable. The model could provide an empirical basis for the decision-making design of autonomous vehicles at intersection.