Fitness exercise evaluation system based on improved DTW algorithm

Fitness movement recognition and evaluation systems play a crucial role in accurately, quantitatively, and efficiently guiding the fitness process, thereby enhancing fitness effectiveness and safety.However, existing research in this area still has some drawbacks, such as high computational complexity, insufficient real-time performance, and low stability. To address these issues, a real-time fitness action recognition and evaluation system is developed based on the lightweight BlazePose model. The system integrates the K Nearest Neighbor (KNN) algorithm for action recognition and classification. In addition, an improved dynamic time warping (S - WFDTW) algorithm is introduced. By utilizing similarity calculations, this algorithm quantifies the metrics and significantly improves the real-time alignment of the time series. Moreover, it provides joint angle analysis and motion trajectory feedback, which enables real-time counting and evaluation of fitness movements. The experimental results demonstrate that the system achieves 98.33% accuracy in movement assessment and counting under complex conditions. In a comparison experiment with the COCO dataset, the average processing rate of the BlazePose model used by the system is 13.3 times faster than that of OpenPose. This significant improvement greatly enhances the efficiency and responsiveness of the system. In conclusion, the developed system achieves efficient and stable recognition of fitness movements, and is able to accurately assess and quantify exercise performance. It can also help users optimize their movements through visual feedback, thus significantly improving the efficiency and accuracy of fitness movement recognition and evaluation.