Object reconstruction and pose indexing by volume feedback

Abstract

Three dimensional reconshuction of a rigid object from monocular video sequences is addressed. Initially object pose is estimated in each image by locating similar (unknown) textures assuming flat depth maps for all input images. Shape-from-silhouette [I] is then applied to make a 3-D model (volume), which is used for a new round of pose estimation, this time by a model-based method giving better estimates. Before repeating this process by building a new volume, pose estimates are adjusted to reduce enor by maximizing a quality measure fur shape-from-silhouette volume reconstruction. The volume feedback is terminated when pose estimates do not change much as compared to those produced by previous iteration. The final output is a pose index (the last set of pose estimates) and a volume. Good performance of the system is shown by several experiments. No model is assumed for the object. Feature points are neither detected nor tracked no problematic feahup matching or correspondence. The high-level pose index generated for input images can be used for content-based retrieval. Our method can be also applied to 3-D object tracking in video.