Clustering of laser measurements via the Dirichlet process mixture model for object tracking

In this paper, the Dirichlet process mixture model is used to describe the distribution of the whole laser measurements in a given scan. Then the number of clusters is inferred from the measurements by the Gibbs sampler. We focus on the automotive application which usually has a more complex environment. Due to the variant shapes and sizes of the real traffic objects, the multi-class DP-based clustering model, which is incorporated with a mixture prior distribution, is proposed to cluster the measurements more properly. The clustering results of the proposed method are compared with those of several existing clustering methods both in an expressway case and in an urban road case. The corresponding tracking performances are also analyzed and the improvements of the proposed method are presented.