This paper presents a novel approach to environment perception providing detailed information for dynamic objects using occupancy grid maps. The shape representation of dynamic objects is derived from dedicated local grid maps. This allows for a precise contour estimation over time in terms of polylines. In addition to that, the local grid map is used to improve the object tracking by formulating measurements for the Kalman filter update overcoming partial occlusion and oversegmentation of raw data. The algorithm is tested on different data sets and initial results are presented and discussed.

No abstract

This paper presents a novel approach to segmentation of a dense 3D point cloud, generated by a flash lidar type camera. Incorporating symmetries of the sensor, the algorithm is using a 2D grid approach to cluster data points and extrude object segments in complex scenes. The data representation allows for the handling of partially occluded, but connected objects at different ranges. The algorithm was tested on a variety of different sensor data sets and the obtained results are presented and discussed.