Human Leg Tracking System Based on Fusion of Laser Range Sensor and Instrumented Insoles Using Learning-based Occlusion Compensation

Abstract

For evaluating the motor function of the elderly and performance of athletes, we have proposed a system that simultaneously measures the leg position and ground reaction force by fusing a laser range sensor (LRS) and instrumented insoles. However, the system had a problem that the leg position was lost when a few sampling times having occlusions. Therefore, this study proposes a method to compensate for the occlusion using machine learning. First, the system tracks legs during straight / turning walking using two LRS without a blind spot in advance. The relationship between the leg trajectory and the direction of the velocity vector in walking steps, which is identified by the insoles, is then learned using Gaussian mixture models for each type of walking. Finally, the system tracks legs using a single LRS and the insoles. When the occlusion occurs, the direction of the velocity vector is estimated through Gaussian mixture regression using the leg trajectory until the previous sampling to compensate for the current hidden leg position.