Physical Exergames Movements and Pattern Recognition using Convolutional Neural Network
Tanvir Fatima Naik Bukht, Ahmad Jalal · 2025
This paper proposes a novel framework for accurately recognizing physical exergames interaction in video sequences. The developed approach uses image processing and machine learning tools to correctly extract and classify the features. The framework comprises five main steps: Silhouette extraction, preprocessing, feature extraction, feature optimization and classification. Image contrast is improved by power law transformation, silhouette is extracted by the Multiple Object Tracking (MOT) algorithm and graph-based segmentation. We extract ORB and geometric skeleton based keypoints to capture discriminative features, and apply Fast Independent Component Analysis (FICA) to optimize feature representation. Thirdly, the Convolutional Neural Network algorithm is used to classify the optimized features, with a macro average accuracy of 0.85 and a weighted average accuracy of 0.87%. The proposed framework effectively recognizes physical exergames interaction and can be applied in many domains, such as human-human interaction and surveillance.