Error Correction of Inertial Motion Estimation by Fusing Multiple Observations

Abstract

This paper proposes a motion estimation method using an inertial measurement unit (IMU) for sports measurement. Although motion capture systems using infrared cameras and high-speed cameras provide accurate motion measurement, they are typically expensive and they require time and effort for installation. Therefore, such motion capture systems are not something that anyone can use easily and daily. IMU-based motion estimation technologies can enable handy and inexpensive motion capture systems. The problem we tackle is accumulated errors caused by integration of IMU data, which need to be corrected to achieve practical estimation accuracy. One of the existing approaches for correcting the errors is to use the constrained initial and final states. The accumulated errors are uniformly distributed and canceled throughout the trajectory. We have previously proposed a generalized method that can incorporate measurements of intermediate states; however, the method requires a long processing time. In this paper, we propose an accelerated extension of the previous method based on a different formulation. The experimental results using simulated dataset indicate that the proposed method exhibited approximately six times faster than the previous method. We also report a measurement result of a 19 seconds table tennis play.