A Data Preprocessing Technique for Gesture Recognition Based on Extended-Kalman-Filter

摘要

Gesture recognition derived from skeletal data plays an important role in our TaiChi rehabilitation training and evaluation system. This paper investigates an extended-Kalman-filter-based preprocessing method to fix those incomplete and inconsistent kinematic sensory data. To evaluate the performance of preprocessing, several representative classifiers such as support-vector-machine (SVM), decision tree, and K-nearest neighbor (KNN) are also employed and investigated in gesture recognition. The addressed work is critically assessed using two open-access Kinect-oriented data sets and one Tai-Chi kinematic data set as benchmark. The experimental results show that the addressed preprocessing technique can improve the gesture recognition rate, and that among the classifiers addressed in this work, SVM has superior performance than others.