This paper introduces PAM (Passive Activity Monitor), an autonomous infrastructure meant to provide a redundant safety system for astronauts on an EVA through body pose recognition. The keypoints identified by the system can be utilized to estimate the action that the astronaut is carrying out, or if the situation requires assistance or intervention from other astronauts in case of emergencies. This paper focuses on OpenPose as a candidate for the detection of keypoints associated with skeletal joints. We evaluate how the system performs on basic movements using next-generation EMUs, an astronaut spinning in microgravity, and an initial overview of how OpenPose performs on images acquired through LiDAR in an environment with no lighting. The results show that OpenPose is a valid keypoint detection engine for humans wearing an EMU and is promising on images acquired through LiDAR technology. However, the system does not perform well when the subject is inverted in a microgravity environment.
Giovanni Vincenti (Sun,) studied this question.