Feature based registration of range images for mapping of natural outdoor environments

Challenges related to viewpoint registration in rough forest terrain can be quite different compared to those faced in structured environments. Maneuvring the sensor between measurement positions introduces large error into the a priori estimates of the registration coordinates. As a consequence, locally optimal registration methods may not work properly. Moreover, due to the clutter, the scene contents can change substantially even due to a relative small displacement of the sensor. Often, the sensor has to be moved to the other side of the target object, such as a group of trees, to get a good coverage of its geometry. In both cases, overlap between the two 3D data sets will be minimal ruling out conventional registration methods. In this paper, a feature-based method for registering 3D range scans for mapping natural outdoor environments is proposed. The method utilizes cylindrical, rotation symmetric features extracted from the 3D measurement data for viewpoint registration. The method is tested on real range images.