A Multiple Hypothesis Approach to Extended Target Tracking

Extended target tracking is a challenging task which has recently been the subject of intense research. Indeed, with the increase of the resolution of modern sensors such as LIDAR or large bandwidth radar, targets reflect multiple measurements and cannot be tracked as a single point. This is a very acute problem in the automotive context, where the targets to be tracked are other vehicles or trucks, and estimating jointly their shape and position is critical to achieve higher level functionalities. In this paper, we propose a PMHT approach to track extended targets modeled as rectangles. Implicit measurement models are used to solve the problem of the unknown origin of the measurement on the surface of the object. As the PMHT does not offer a mechanism to model existence probability, we use a Hidden Markov Model (HMM) to decide on the presence of every target. Also, as laser scanner does not offer a classification for static and dynamic objects, we created a freespace based classification used for track initialization. Finally, we demonstrate the effectiveness of our approach with real data from laser scanner and radar.