Deep Learning Based Rope Skipping Evaluation and Training System

Adolescents lack a rope skipping posture detection and targeted quality training system. To solve this problem, we propose a deep learning-based assessment and training system for rope skipping. The system is mainly divided into two modules: assessment and training. The assessment module extracts and analyzes the high-dimensional rope skipping features during the monitoring time period to give an assessment report and training actions. The training module supervises students’ exercise by classifying the movements and counting the number of movements through a deep learning model. The construction of the system includes Blazepose data acquisition, data preprocessing, rope skipping posture detection, and related ability improvement training. In the data preprocessing section, we used an exponentially weighted average filter for the experiments. The experimental results show that the exponentially weighted average filter has good smoothing effect on the 2D skeletal point data of jumping rope. In the ability enhancement training module, we explored a variety of algorithms for recognizing movements, including MLP, Transformer, LSTM and CNN. experimental results show that the MLP model has a more superior performance, and its accuracy reaches 98.88%.