Emerging Interfaces
Reconstructing Dense 3D Human Point Clouds From Body Heat, Using One Commodity Thermal Array
TAP3D reconstructs 3D human point clouds from body heat signatures using a single commodity thermal array sensor. A forward thermal physics model plus self-supervised multi-primitive estimation recovers depth; geometric perspective fusion separates multiple people. 160K samples across 8 environments and 11 users, with fall detection at 91.46% and mesh recovery at 4.87 cm error. Open-sourced.