Object tracking and classification using a multiple hypothesis approach

In this paper, an algorithm for tracking objects in traffic scenes using a multiple hypothesis approach is presented. The sensor used to get information about the environment is a laser range finder. This sensor has the advantage of obtaining accurate distance and geometric information of the objects in front of the car. In order to achieve a robust classification, geometric information is not always sufficient enough, if objects are separated into multiple parts, due to occlusions or bad segmentation. Therefore the presented method allows several classification results and keeps track of all feasible combinations of disintegrated objects. Thus, it is possible, to process several hypotheses of objects and no objects can be missed.