Kinect Gesture Recognition for Interactive System

Gaming systems like Kinect and XBox always have to tackle the problem of extracting features from video data sets and classifying the body movement. In this study, reasonable features like human joints positions, joints velocities, joint angles and joint angular velocities are extracted. We used several machine learning methods including Naive Bayes, Support Vector Machine and Random Forest to learn and classify the human gestures. Simulation results and the final confusion matrix show that by combining delicately preprocessed data sets and Random Forest methods, the F-scores of the correct predictions can be maximized. Such methods can also be applied in real-time scenarios.