Decision-free true positive estimation with grid maps for multi-object tracking

Abstract

A challenge for future advanced driver assistant systems is to establish a reliable environment perception. Recently, advanced multi-object tracking algorithms were presented. These algorithms consider spatial uncertainties and clutter detections from several different sensors and a fusion process combines all the information in a probabilistic framework. Especially the true positive probability of object measurements is taken into account by these algorithms. This contribution presents an approach to estimate the true positive probability of object measurements in order to improve the tracking results. One source of information is a grid map based motion estimation, which applies the Dempster-Shafer theory of evidence. The other source is the result of an object classification algorithm based on the outer appearance of vehicles. First results of an evaluation with real world data are presented.