Quantitative Evaluation of Gymnastics Based on Multiple MEMS Sensors

The quantitative evaluation of sports motions is a precondition for scientific training and data-driven professional athletic motion analyses. However, the high accuracy and the portability of the evaluation schemes are contradictory and it is difficult to simultaneously improve both metrics in one device. In this work, we propose an effective gymnastics motion recognition and evaluation system to solve this problem. The system is a sensor network composed of 11 acceleration and angular velocity sensors. The gymnastics data obtained from each sensor node can be uploaded in real-time via Wi-Fi. The data from different sensor nodes on the body is used to form the corresponding gymnastics motion recognition data sets. Using these sensor node data and the linear discriminant analysis (LDA) algorithm, 6 different sets gymnastics motions could be recognized at an accuracy of 99.18%. Experimental results also showed that, by combining only the data from the left arm sensor node and the right foot sensor node as the gymnastic motion evaluation data set, the LDA algorithm could be used to evaluate three gymnastics motions, all of which delivered an accuracy of no less than 96.0%. This evaluation system meets the needs for accuracy and portability at the same time, and potentially can be extended to evaluate other physical activities, such as yoga and Tai Chi.