Object Tracking via the Probability-Based Segmentation Using Laser Range Images

Abstract

In this paper, a probability-based segmentation approach is presented for object tracking. The proposed approach uses the Dirichlet process mixture model to describe the probabilistic distribution of observations in a single scan of a laserscanner. Then the number of segments is inferred from the observations by the Gibbs sampling method. Moreover each segment is classified into one of the three predefined classes such that most of non-vehicle-like objects on the roadsides can be filtered out. Then, the tracking algorithm, called Joint Integrated Probabilistic Data Association Filter (JIPDAF), is applied to track the classified objects and manage existing tracks. Simulations based on real traffic data demonstrate that the non-vehicle-like objects on the roadsides are suppressed. Since the number of objects in the tracking step is decreased, the computation load of the tracking step is decreased.