Physical Fitness Test System Based on Visual Analysis

In order to realize the efficiency and fairness of physical fitness test, in this paper, we proposed a complete system based on visual analysis, including action recognition, counting, and evaluation. We presented an approach for the action recognition based on spatial-temporal graph convolutional network of multi-semantic fusion, we introduced the both time and frequency domain information of the key point sequence for the action counting, and the action evaluation was realized by using action sequence similarity score. In sit-ups, push-ups, rope skipping and other projects, the accuracy of action recognition reached 98%, the accuracy of action counts reached 95%, and the recall rate of action evaluation on sit-ups and push-ups reached 88%.