RAPTR: Radar-based 3D Pose Estimation using Transformer

Pu Wang (Mitsubishi Electric Research Labs) · Sorachi Kato (The University of Osaka) · Ryoma Yataka (Mitsubishi Electric Corporation) · Pedro Miraldo (Mitsubishi Electric Research Labs (MERL)) · Takuya Fujihashi (Osaka University) · Petros Boufounos (Mitsubishi Electric Research Laboratories)

2d keypoint labels3d bbox3d gravity loss3d human pose estimation3d template lossdepth ambiguitiesindoor radar datasetsjoint decoderjoint position errormulti-view radar featurespose decoderpseudo-3d deformable attentionradar-basedraptrtwo-stage pose decoderweak supervision

Radar-based indoor 3D human pose estimation typically relied on fine-grained 3D keypoint labels, which are costly to obtain especially in complex indoor settings involving clutter, occlusions, or multiple people. In this paper, we propose \textbf{RAPTR} (RAdar Pose esTimation using tRansformer) under weak supervision, using only 3D BBox and 2D keypoint labels which are considerably easier and more scalable to collect. Our RAPTR is characterized by a two-stage pose decoder architecture with a pseudo-3D deformable attention to enhance (pose/joint) queries with multi-view radar features: a pose decoder estimates initial 3D poses with a 3D template loss designed to utilize the 3D BBox labels and mitigate depth ambiguities; and a joint decoder refines the initial poses with 2D keypoint labels and a 3D gravity loss. Evaluated on two indoor radar datasets, RAPTR outperforms existing methods, reducing joint position error by $34.3$\% on HIBER and $76.9$\% on MMVR. Our implementation is available at \url{https://github.com/merlresearch/radar-pose-transformer}.