Gesture classification with machine learning using Kinect sensor data

Abstract

We present approaches for gesture classification and gesture segmentation by using machine learning on the Kinect sensor's data stream. Our work involved three phases. Firstly we developed gesture classification from a known vocabulary of gestures in an edited data stream. Secondly we extended those techniques to detect and classify a gesture in an unedited stream which also captures random movements. Thirdly, we apply rules to filter out movements that were not intentional gestures and yet resembled certain gestures in our vocabulary.