Mobile robotic active view planning for physiotherapy and physical exercise guidance

Abstract

Consulting a physiotherapist or physical trainer for long term regular exercises is not financially affordable for all. As a solution, our long term research goal is to develop a robotic physiotherapist/physical trainer which could instructively and physically guide a performer. Towards that direction, in this paper we present an autonomously repositioning mobile robot to observe a person throughout a sequence of physical exercises by selecting less self-occluded viewpoints. A viewpoint specific joint mutual occlusion (JMO) measurement is formulated at candidate viewpoints. Then a utility function, which accounts for joint occlusion, skeleton coverage, sensing range and moving cost, is averaged over the sub-activity periodic duration to find the optimal viewpoint. Proposed methods have been evaluated with multi-view dataset and an online mobile robot while a person performed eight different physical activities with two trials each. Results indicate proposed active view planner can autonomously drive the mobile robot to a less self-occluded viewpoint over manually setup fixed viewpoint observation, leading to more accurate human movement analysis.