Joint Angle Estimation of Upper Limb based on Evolutionary Multi-criterion Optimization using 3D Image Sensor

Abstract

In this research, we have developed a rehabilitation support system for analysis of upper body motion using 3D image sensor. Such a sensor has embedded systems to detect and track joint positions, but it is difficult to directly apply the data to calculation of inverse kinematics to estimate the joint angles because of errors in measurement. To solve the issue, some researchers have proposed methods of joint angle estimation using kinematic model and search technique. In these works, objective functions were defined to evaluate the similarity of posture between a person and kinematic model, and the problem was typically considered as a single-objective optimization problem. However, the data basically includes not only feature of personal behavior patterns but also noise. Therefore, we address a multi-objective optimization problem in order to scale back the influence of noise. This paper proposes a method of joint angle estimation of upper limb base on Evolutionary Multi-criterion Optimization (EMO). Furthermore, we apply a feedforward neural network to motion pattern modeling to realize a predictive search, combining with the EMO.