Object tracking from medium level stereo camera data providing detailed shape estimation using local grid maps

This paper presents a novel approach to environment perception providing consistent information about stationary obstacles, from an occupancy grid, and dynamic objects, from a model based object tracking at the same level of detail. Raw ranging data from dynamic objects is disregarded for the stationary grid map. A detailed shape representation of dynamic objects is achieved by using dedicated local grid maps for each track. In addition to that, a technique for evaluating tracking algorithms via assigned local object grid maps is presented and discussed. The algorithm is tested on different data sets and the obtained results are presented and discussed.