A Loop-Closure-based Inertial Motion Capture, with Application to Sports Swing Measurements

Abstract

This study presents a system for measuring motions using an inertial measurement unit (IMU) and a camera. IMU-based motion capture systems are advantageous in terms of cost, but measurable motions are limited in length because of rapidly accumulating errors in the process of double integration. We propose the correction of such errors with the assistance of an external monocular camera. Our approach is similar to graph-based simultaneous localization and mapping for mobile robots, where loops in the trajectory are detected and closed to correct the accumulated errors over the trajectory. We apply the idea to motion estimation by developing and integrating three key technologies: pose-graph optimizer, loop detector, and time synchronizer. We evaluate our system through measurements of table tennis swings. The experimental results show that our sensor fusion approach significantly reduces the errors in inertial motion estimations, with an acceptably small additional cost.