Arm Angle Estimation for Computational System Rehabilitation with Range Image Sensor

Abstract

This paper proposes a method of arm angle estimation by using a range image sensor that can detect each angle position in the non-contact measurement. We can solve the inverse kinematics by using relative position data, but the estimation quality is not good owing to the measurement noise of the sensor. Therefore, we apply genetic algorithm to solve the optimization problem. Furthermore, if we can model the human motion pattern from the measured data, it can be prevented falling into the local solution. In the method, we apply a feed-forward neural network (NN) for learning the motion patterns. In this paper, we show some experimental results of the angle estimation and discuss the effectiveness through the experimental results.