Field of view dependent registration of point clouds and incremental extraction of table-tops using time-of-flight cameras

Perception of the environment is crucial for many robot applications. Thus, geometric and semantic mapping using point cloud sensors is subject to many research activities. In this paper, an approach to incrementally register point clouds from time-of-flight cameras and create feature maps is presented. Frustum culling ICP, key frame based plane segmentation and aggregation of a feature map are presented and evaluated on the service robot Care-O-bot® 3. The focus is on table-top extraction for tele-operated robots.