Anthropometry and Pose Observation using Low-dimensional Latent Optimization (APOLLO)

The US Army has identified a need for a modular perception system that detects and estimates the shape and pose of casualties in the battlefield. Such a system would enable existing and future unmanned systems to safely perform autonomous casualty extraction and evacuation tasks. Although multiple commercial solutions to pose estimation exist, they often require the use of active sensing technologies or multiple fixed cameras that suffer from limited range and poor outdoor performance. Charles River Analytics proposes to develop a perception system for Anthropometry and Pose Observation using Low-dimensional Latent Optimization (APOLLO). APOLLO consists of a hardware-software kit comprised of a monocular EO/IR camera for casualty detection at a distance, and a stereo EO/NIR camera for fine-grained pose and shape estimation. APOLLO leverages our existing pedestrian detection and tracking technologies, together with a parameterized anthropometric model of the human body, to achieve real-time detection, pose, and 3D shape estimation of casualties. APOLLO assumes no restrictions on background complexity, presence of occlusions, or lighting conditions to enable robust performance in uncontrolled environments.