Effectiveness Comparison of Kinect and Kinect 2 for Recognition of Oyama Karate Techniques

The aim of this research is to evaluate effectiveness of Kinect and Kinect 2 for recognition of specialized actions namely Oyama karate techniques. As a classification algorithm we have used Gesture Description Language (GDL) classifier. We have recorded a dataset that contains motion (MoCap) recordings of two professional sport (black belt) instructors and masters of Oyama Karate. The whole dataset contained 200 movement samples per person (400 samples per Kinect camera, so totally we had 800 samples). After data was captured it was split into two subsets: training and validation. Each subset contained only recordings of a single user. We have performed 2-fold cross validation switching the role of user data between training and validation set. In all cases but two MoCap data from Kinect 2 appeared to be more reliable than from Kinect 1 taking into account both recognition rates of GDL classifier and error classification cases. Also in case of Kinect 1 standard deviation of results were higher which means that classifier becomes less stable while trained on data from older device. The ability of more accurate calculation of legs joints positions seems to be biggest advantages of Kinect 2 over its predecessor.