Traffic Participant Classification Method Based on Decision Tree for Point Cloud Data Detected by Event Camera

Abstract

Event camera has the advantage of accurately identifying moving targets while being insensitive to stationary targets, which makes up for the lack of traditional video streaming camera and has a wide range of applications in the field of traffic flow detection. In this paper, a traffic participant classification method based on a decision tree for point cloud data acquired by an event camera in a roadside installation scenario is proposed. For traffic identification, 5 basic features to describe the geometrical and quantitative characteristics, and 8 Gaussian projection features to describe the point distribution characteristics are extracted and analyzed. Furthermore, the CART decision tree is used to identify four kinds of traffic participants, including the large vehicle, the compact vehicle, the non-motor vehicle, and the pedestrian. By modulating parameters of the maximum layer and feature weighting values, the proposed method can reach equilibrium in generalization ability and accuracy. Experimental results show that the proposed method demonstrates high accuracy of 97.49% in target participant classification.