Obstacle Detection and Tracking in Range Images for an Autonomous Vehicle

Abstract

Environmental perception such as obstacle detection and tracking is an inevitable requirement for autonomous vehicles. Especially, in order to avoid collisions, autonomous vehicles require detection and tracking of moving objects (DATMO) such as vehicles and pedestrians. The multiple hypothesis approach (MHA) is most frequently used for DATMO, whereby the MHA generates all possible hypotheses for moving objects. However, the MHA also generates hypotheses for immobile objects such as parked vehicles, bushes, and trees, among others, the result of which is a computational overhead for DATMO as a result of tracking irrelevant hypotheses. The ability to distinguish moving objects from static objects using a laser scanner with an occupancy grid map has been developed previously. In the present study, we utilized this method in order to apply MHA to only moving objects. Using this approach, moving objects were efficiently detected and tracked on a grid map. Furthermore, DATMO on the grid map was effectively implemented in real time.