Gesture Recognition Dynamics: Unveiling Video Patterns with Deep Learning
Publication Date
1-1-2024
Document Type
Conference Proceeding
Publication Title
2nd IEEE International Conference on Data Science and Network Security, ICDSNS 2024
DOI
10.1109/ICDSNS62112.2024.10691103
Abstract
Investigating the intricate dynamics of complex video gestures in gym activities enhances our understanding of human movement, offering opportunities to improve performance, refine training methods, and advance healthcare applications. This exploration also challenges the fields of human motion analysis and activity recognition, driving innovation towards more sophisticated approaches. This paper examines the fusion of Long Short-Term Memory Networks (LSTM) and Convolutional Neural Networks (CNN) for recognizing complex video gestures in gym activities. Mediapipe, an open-source framework by Google, is used to extract poses and track essential body points, providing inputs for a deep model combining LSTM and CNN. This combination enhances the model's ability to learn dynamic gesture patterns, capturing temporal dependencies and extracting spatial features for robust recognition. Using the UCF50 dataset, which includes various exercise activities, the model achieved a test accuracy of 94.77%. This work demonstrates the feasibility of detecting complex activities and sets the stage for broader applications.
Keywords
CNN, Gesture Recognition, LSTM, Mediapipe
Department
Computer Science
Recommended Citation
Nithish Reddy Agumamidi and Sayma Akther. "Gesture Recognition Dynamics: Unveiling Video Patterns with Deep Learning" 2nd IEEE International Conference on Data Science and Network Security, ICDSNS 2024 (2024). https://doi.org/10.1109/ICDSNS62112.2024.10691103
