VIV: Using visible internal volume to compute junction-aware shape descriptor of 3D articulated models

An articulated shape is composed of a set of rigid parts connected by some flexible junctions. The junction has been demonstrated to be a critical local feature in many visual tasks such as feature recognition, segmentation, matching, motion tracking and functional prediction. However, efficient description and detection of junctions still remain a research challenge due to high complexity of articulated deformation. This paper presents a novel junction-aware shape descriptor for a 3D articulated model defined by a closed manifold surface. To encode junction information on the shape boundary, the core idea is to develop a new geometric measure, called the visible internal volume (VIV) function, which associates the shape’s volumetric context to its boundary surface. The VIV at an arbitrary point on the shape boundary is defined as the volume of visible region within the shape as observed from the point. The VIV variation serves as the new shape descriptor. One advantage of using the VIV for 3D articulated shape description is that it is robust to articulation and it reflects the shape structure and deformation well without any explicit shape decomposition or prior skeleton extraction ∗Corresponding author at: School of Software, Tsinghua University, Beijing 100084, China. Tel.: +86 10 6279 5455; Mobile: +86 159 1083 1178. URL: http://cgcad.thss. tsinghua.edu.cn/liuyushen/ Email addresses: [email protected] (Yu-Shen Liu), [email protected] (Hongchen Deng), [email protected] (Min Liu), [email protected] (Lianjie Gong) Preprint submitted to Neurocomputing April 23, 2015 procedure. The experimental results and several potential applications are presented for demonstrating the effectiveness of our method.