Search papers, labs, and topics across Lattice.
This paper introduces ScaleHP, a novel hand pose estimation framework that accurately predicts hand poses in absolute metric space, overcoming the limitations of existing methods that rely on root-relative coordinate systems. By leveraging intrinsic proportional relationships among human hand bones and employing a transformer-based decoder with a scale token, ScaleHP effectively integrates multi-scale features to enhance pose estimation precision. The method achieves state-of-the-art results, including a CS-MPJPE of 35.8 on FreiHand, demonstrating that biological constraints can significantly improve hand tracking accuracy in real-world applications.
Internal biological constraints can dramatically reduce errors in hand pose estimation, enabling robust tracking in metric space.
Accurate metric-space hand pose estimation (HPE) is essential for immersive human-computer interaction and robotics. However, most existing methods predict poses in a root-relative coordinate system and cannot estimate the hand in absolute metric scale. In this work, we observe that the intrinsic proportional relationships among human hand bones encode stable anthropometric priors that implicitly correlate with the overall metric size of the hand. Leveraging this insight, we present ScaleHP, an end-to-end one-stage hand pose estimation framework that bypasses fragile extrinsic depth modules to recover the hand in metric space. ScaleHP employs a transformer-based decoder with a novel scale token to fuse multi-scale morphological and appearance features. By solving for metric coordinates through a perspective-constrained least-squares approach, we achieve high-precision pose estimation in the camera coordinate system. ScaleHP delivers state-of-the-art performance, including 35.8 CS-MPJPE on FreiHand and 4.6/5.9 PA-MPJPE on DexYCB and HO3Dv3. These results demonstrate that internal biological constraints significantly reduce relative geometry and absolute metric errors, offering a robust solution for generalized, real-world hand tracking.
Tsinghua AI