Recursive non-rigid structure from motion with online learned shape prior

摘要: Most existing approaches in structure from motion for deformable objects focus on non-incremental solutions utilizing batch type algorithms. All data is collected before shape and reconstruction take place. This methodology inherently unsuitable applications that require real-time learning. Ideally the online system capable of incrementally learning building accurate shapes using current measurement past reconstructed shapes. Estimation 3D camera position done online. To rely only measurements up until moment still a challenging problem. In this paper, novel approach proposed recursive recovery non-rigid structures image sequences captured by single camera. The main novelty method an adaptive algorithm construction constraints imposing stability proposed, adaptively learned have two aspects: imposed basis shapes, basic ''building blocks'' which are reconstructed; as well mixing coefficients form their probability distribution. Constraints updated when model no longer adequately represents new achieved means Incremental Principal Component Analysis (IPCA). technique also to handle missing data. Results presented capture based articulated face simple human full-body movement.