Gait Analysis Combining Sports Biomechanics and Imaging

Abstract

Aiming at the problems of marker positioning error, soft tissue interference, and insufficient personalized modeling ability in traditional gait analysis, this study proposes a gait assessment method that integrates sports biomechanics and multimodal imaging analysis. Firstly, based on the public OU-ISIR gait database, a Convolutional Neural Network (CNN) is used to extract the key gait features in the human silhouette map, and a Long Short-Term Memory Network (LSTM) is used to perform temporal modeling of the dynamic gait cycle to reduce the impact of clothing changes and perspective differences on feature extraction. Secondly, a personalized joint dynamics model is constructed through finite element analysis, and an iterative optimization algorithm is used to associate imaging parameters with muscle dynamics indicators, thereby improving the model's ability to adapt to individual anatomical differences. Experimental results show that the classification accuracy of this method under covariate conditions reaches 97.5% (clothing changes) and 91.2% (viewing angle differences). In joint angle prediction, the angle errors of the hip, knee, and ankle joints are 1.1°, 1.2°, and 1.3°, respectively, which are lower than those of the existing methods. Ablation experiments further verify the effectiveness of each module, indicating the key role of dynamic modeling, personalized adaptation, and parameter optimization in improving the overall performance. The results show that the proposed method performs well in terms of accuracy, robustness, and efficiency, providing reliable technical support for the practical application of gait analysis in clinical monitoring and smart devices.