Imposing temporal consistency on deep monocular body shape and pose estimation

Abstract Accurate and temporally consistent modeling of human bodies is essential for a wide range applications, including character animation, understanding social behavior, AR/VR interfaces. Capturing motion accurately from monocular image sequence remains challenging; quality strongly influenced by temporal consistency the captured body motion. Our work presents an elegant solution to integrating constraints during fitting. This increases both robustness optimization. In detail, we derive parameters models, representing shape person. We optimize these over complete sequence, fitting single while imposing on motion, assuming joint trajectories be linear short time. approach enables derivation realistic 3D models sequences, jaw pose, facial expression, articulated hands. experiments show that our estimates even challenging movements poses. Further, apply it particular application sign language analysis, where accurate modelling essential, well-suited this kind application.