Junction-aware shape descriptor for 3D articulated models using local shape-radius variation

• Present a new junction-aware shape descriptor.
• Design a center-surround filter operator to identify junction regions.
• Apply our algorithm to junction extraction.

An articulated model is composed of a set of rigid parts connected by some flexible junctions. The junction, as a critical local feature, provides valuable information for many 3D semantic analysis applications such as feature recognition, semantic segmentation, shape matching, motion tracking and functional prediction. However, efficient description and detection of junctions still remain a research challenge due to high complexity of 3D articulated deformation. This paper presents a new junction-aware shape descriptor for a 3D articulated model defined by a closed mesh surface. The core idea is to exploit the local shape-radius variation for encoding junction information on the shape boundary surface, where the shape-radius at each point on the surface is the radius of corresponding medial balls within the shape. The presented descriptor is typically computed using a center-surround filter operator, which calculates the Gaussian-weighted average of shape-radius in the neighborhood of each point on the surface. Our descriptor is robust to articulation and can reflect the junction feature well without any explicit shape decomposition or prior skeleton extraction procedure. The experimental results and several potential applications are proposed for demonstrating the effectiveness of our method.

Figure optionsDownload as PowerPoint slide