Classification of Dynamic In-Hand Manipulation Based on SEMG and Kinect

This paper proposes a hand motion capture system for recognizing dynamic in-hand manipulation of the subjects based on the famous sensing techniques, then transferring the manipulation skills into different bionic hand applications, such as prosthetic hand, animation hand, human computer interaction. By recoding the ten defined in-hand manipulations demonstrated by different subjects, the hand motion information is captured with hybrid SEMG and Kinect. Through the data preprocessing including motion segmentation and feature extraction, recognizing ten different types of hand motions based on the rich feature information are investigated by using Marquardt-Levenberg algorithm based artificial neural network, and the experimental results show the effectiveness and feasibility of this method.