ABSTRACT
Millimeter-wave radar offers a privacy-preserving alternative to RGB sensors for human pose estimation. This paper proposes milliMamba-3D, extending the milliMamba framework to 3D pose estimation, addressing depth ambiguity and geometric consistency challenges.
PAPER · PDF
Loading PDF…
Key findings
Proposes a Tri-View Geometry-Aware Mamba Encoder for 3D geometric reasoning.
Introduces a Depth-Aware Spatio-Temporal Decoder for 3D joint coordinate prediction.
Includes a Multi-Scale Depth Estimation Module leveraging radar phase information.
Implements a Temporal Consistency Regularization scheme for anatomical plausibility.
Limitations & open questions
The paper is a research proposal and does not yet include experimental results or validation.