<title>Multiview-based pose estimation from range images</title>

This paper presents a multi-view based method for estimating poses of free floating/rotating objects to support robotic manipulation in space. The multi-view based method consists of two-stage processing, namely off-line processing for building a multi-view database of each model, and on-line processing for real-time pose estimation. The multi-view database is composed of feature vectors from the range images of multiple views of the model. The feature vectors are organized into a KD tree for fast spatial indexing. At run-time, a small number of candidate poses are extracted from the KD tree via efficient feature matching, and are verified and refined using an optimization procedure to obtain the estimated pose, which may be further refined by a Kalman filter.