A Deep Learning Framework for Human Motion Recognition Using Compact CNNs and Swarm Optimization
Year: 2025
Type: article
Abstract: Aerobic exercise is essential to maintain the cardiovascular system's health, yet accurate analysis of its complex motions poses severe challenges. Traditional methods often encounter computational loads, overfitting, and unfavorable generalization in real applications. In trying to resolve these issues, this work introduces the MCT-MCT-MCT-MCT-MCT-MCT-MCT-HCNanoNet-AM framework, which combines motion capture technology (MCT) and a Humboldt Squid Optimization Algorithm (HSOA)-optimized Compact-NanoNet Deep Convolutional Neural Network (CNanoNet). This blended approach is going to enhance the accuracy and efficiency of aerobic movement analysis. The MCT-MCT-MCT-MCT-MCT-MCT-MCT-HCNanoNet-AM system workflow incorporates several key steps. Initially, high-speed motion capture technology is utilized to capture data. The raw data are processed beforehand using Mean Curvature Flow, which is a technique to remove noise at the first step. Afterwards, feature extraction and movement detection are done using the CNanoNet model. The HSOA is then employed to optimize the CNanoNet for improving model performance. The optimized model is then utilized for real-time aerobic movement recognition with personalized feedback to the users. Experimental results indicate that the MCT-HCNanoNet-AM system noticeably enhances the recognition and classification of aerobic movements with an exceptional set of 99.98% accuracy, 99.99% precision, 99.986% specificity, 99.99% recall, and 99.98% F-Score. The system takes 91 seconds in computation and area under the curve (AUC) of 0.9976. Integration of real-time feedback mechanisms not only helps in refining the techniques of users but also in the prevention of injuries by identifying potential risks. Overall, the MCT-HCNanoNet-AM system is a major advance in aerobic movement recognition technology that enhances performance as well as overall physical well-being through innovative technological advancements.
DOI: 10.14419/km6frv17
Authors: Swagatika Devi, , S. Rahmath Nisha, Sarange Shreepad Marotrao, Roopa R, B. Uma Maheswari, P. Murugeswari, P. John Augustine, Sumit Chaudhary, H. Mickle Aancy, R. G. Vidhya, Sreerekha.C. G
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4413238920