3D modeling from range images with integrated registration

3D modeling of cluttered environments for semi-autonomous robotic systems implies that the position and orientation of the sensor is determined for each viewpoint from which measurements are collected. Often the registration estimation is performed offline or with the help of external tracking devices. In this paper, we propose to integrate an efficient range data-based registration estimation technique with a probabilistic occupancy modeling scheme which both make use of the same data set to estimate registration and to build 3D virtual representations of complex scenes. The resulting models are to be used for collision avoidance as well as for interaction with human operators who have to validate operations and guide data acquisition. The impact of the registration estimation technique on the quality of the models is examined in the context of telerobotic applications under operator supervision.