Real-Time Motion Recognition in Special Training Systems Based on the Optimized BBO-KNN Method of Motion Morphology

Abstract

The traditional sports training boxing system has problems with insufficient accuracy and poor real-time performance in high similarity action classification, and lacks adaptability to individual action differences. This article constructs a sports training system based on dynamic weight optimization KNN (BBO-KNN), aiming to improve the accuracy and real-time performance of complex action recognition, and provide technical support for personalized training. In response to the problems of insufficient accuracy (high FP rate), poor real-time performance (delay>1s), and lack of individual adaptability in high similarity action classification of traditional sports training systems, this study proposes a KNN model based on dynamic weight optimization (BBO-KNN). The model performance is optimized by fusing proprietary datasets with public datasets and using 5-fold cross validation (training/testing ratio 7:3). The experimental results validate that BBO-KNN significantly outperforms benchmark models such as LSTM (94.50%) and SVM (89.30%) in accuracy (96.20% ± 0.3%). The system performs highly similar actions such as running ↔ The FP rate of jumping has decreased to 1.6%, and the global FP rate is 1.39%.and robustness (noise interference fluctuation ± 1.2%). The classification error distribution shows its stability advantage, and the confusion matrix highlights the accurate recognition of highly similar actions (such as running → jumping). Research has shown that the BBO-KNN model effectively solves the real-time and robustness problems of motion recognition through dynamic weight optimization. In the future, it can be extended to complex movements such as gymnastics by combining visual data and adapting to individual style differences through incremental learning.