Abstract
In this paper, we presented a method for segmenting a mesh model of an articulated object from sequential stereo images. Coupled with an existing modeling and joint recovery techniques, our method reconstructed objects with various shapes successfully. We achieved robustness for our segmentation method by combining a graph-cut algorithm with proposed functions for updating link capacities. However, our research has limitations as follows. The quality of the reconstructed model is only guaranteed when feature points are evenly extracted from all sides of the object because the mesh generation and segmentation processes are performed based on feature matching. In addition, our approach considers objects composed of only two rigid parts linked by a joint. In the future, we plan to further investigate articulated objects with multiple parts and joints, and the issue of textureless object reconstruction is in our scope for the future plan.