Full-Body Pose Motion Tracking from Sparse Data via Morphology-aware Constraints

Precise full-body pose motion tracking in Virtual Reality (VR) especially under minimal sparse sensor configurations makes consumer-level immersive entertainment and intuitive interactions more accessible. The current VR setup typically utilizes a Head-Mounted Display (HMD) along with hand controllers for full-body pose estimation. Nevertheless, these studies face significant challenges in accurately capturing reasonable full-body motion, especially in ambiguous lower-limb motion. To address these challenges, in this paper, we propose a novel data-driven framework named Morphology-aware Constraint Poser (MCP), which employs only six tracking devices. Specifically, to account for the influence of individual morphology on pose representation, we design a morphological and skeletal constraint mechanism to streamline pose estimation and improve tracking precision. To ensure realistic and physiologically plausible human motion, we introduce a physiological penalty loss function that constrains joint angles in a range of valid values, thereby reducing unreasonable motion and enhancing generalization capabilities. Furthermore, to bridge the gap between the training dataset and real-world scenarios, we employ data homogenization techniques to bolster robustness. We conduct comprehensive evaluations of MCP on the publicly available AMASS motion dataset and real-world VR scenarios. The experimental results have demonstrated the effectiveness of MCP in accurately tracking full-body motion. The code will soon be released as open source.