Proper

Abstract

With the release of Microsoft's Kinect controller, many researchers and commercial developers have developed a variety of applications using real-time human motion analysis. The recently released Kinect 2 has a much higher resolution compared to Kinect 1 and has improved performance by offering more functionalities such as facial expressions and more human body joints. However, in order to use Kinect for exercise games, it is necessary to accurately measure and recognize the user's exercise posture. In this paper, human kinematics data using Kinect 2 sensor are compared with data of motion capture system to evaluate and measure the accuracy of Kinetic-based exercise posture recognition system. The experiment was conducted to evaluate the accuracy of exercise posture recognition data obtained from two Kinect sensors at the same time with the model capture system for the 16 types of exercise postures taken by an expert trainer. The results of the analysis show that the Kinect sensor tracks the exercise postures relatively accurately compared with the motion capture system.