Exercise Guidance Generation Based on Joint Vector Model Using Kinect Sensor

This paper proposes an exercise guidance system to help beginners to learn exercise by themselves. The system teaches a beginner how to improve the wrong motion of his/hers by comparing the motion of an expert and that of a beginner. Our system can generate guidance sentences based on the comparison result of motions with a joint vector model that represents the relative positional relation between joints such as shoulder joint and elbow one. Furthermore, our system can reduce the effort to acquire a motion rule that represents the correct motions of an expert. We implemented our proposed exercise guidance system for the Japanese popular radio exercise No. 1 and confirmed that the guidance can help a beginner to correct the wrong motion.