Human Action Recognition Based on Skeleton Information and Multi-Feature Fusion

Action assessment and feedback can effectively assist fitness practitioners in improving exercise benefits. In this paper, we address key challenges human action recognition by proposing innovative methods that enhance performance while reducing computational complexity. Firstly, present Oct-MobileNet, a lightweight backbone network, to overcome the limitations of traditional OpenPose algorithm’s VGG19 which exhibits large parameter size high device requirements. Oct-MobileNet employs octave convolution attention mechanisms improve extraction high-frequency features from body contour, resulting enhanced accuracy with reduced model burden. Furthermore, introduce novel approach for combines skeleton-based information multiple feature fusion. By extracting spatial geometric temporal characteristics actions, employ sliding window algorithm integrate these features. Experimental results show effectiveness our approach, demonstrating its ability accurately recognize classify various actions. Additionally, evaluation exercises, specifically focusing on BaDunJin movements. We propose multimodal information-based method pose detection keypoint analysis. Label sequences are obtained through detector each frame’s coordinates represented as vectors. Leveraging information, including label vectors, explore similarity perform quantitative evaluations help exercisers assess quality their performance.