Full-Body Motion Recognition in Immersive- Virtual-Reality-Based Exergame

Abstract

Exergames have beneficial effects on the player's motivation to exercise. However, many current games lack accurate full-body motion recognition, resulting in players not performing the physical exercise the game requires. Therefore, we aim to develop an immersive virtual reality exergame that simultaneously recognizes and reconstructs full-body movements to motivate players to learn and practice yoga. The system analyzes the entire movement execution and identifies the player's execution errors to provide appropriate feedback so that players can then improve their movements. Such a system can be used in exergames designed for rehabilitation purposes to assist patients or to monitor their improvement. To access recognition performance, we trained and tested hidden Markov models and applied the leave-one-out cross-validation. The results show that the system achieves an F1-score of 0.79 for yoga warrior I, 0.85 for yoga warrior II, and 0.66 for extended side angle. A user study with 32 participants revealed that the game was fun and that the players enjoyed it. Moreover, performance results show that players needed fewer attempts to correctly perform a pose as the exergame progressed.