Abstract: According to the unsupervised and rudderless independent exercise training, also the nonstandard movements in it, an exercise training assist system based on Kinect is designed and realized. The coordinates of joint points in the human body are captured by Kinect, the features of movement are extracted, and the motion recognition algorithm is implemented by Dynamic Time Warping, then the template and the action are matched in real time, and evaluation is fed back in accordance with the matched degree. The experiment results show that the average correct recognition rate is 91.25%, the correct evaluation rate is 95.9% and the evaluation is fed back timely and effectively by the system, which indicates that the system meets the requirements of exercise training assist.

Key words: Kinect, skeletal tracking, exercise training, Dynamic Time Warping, exercise action recognition