Particle swarm optimization-driven deep maxout network for effective monitoring of paralyzed persons
Year: 2025
Type: article
Abstract: Effective monitoring of paralyzed individuals is crucial for ensuring their safety and well-being, particularly in detecting falls and abnormal postural states. This research proposes a Particle Swarm Optimization (PSO)-driven Deep Maxout Network (DMN) to enhance the accuracy and efficiency of human posture recognition. The proposed system utilizes RGB images from the Fall Detec-tion Dataset, which are preprocessed through resizing, normalization, data augmentation, and bounding box transformations. The DMN model, enhanced with Maxout activation, is employed for robust feature extraction, ensuring superior discrimination of pos-tural states. Additionally, PSO is integrated for hyperparameter optimization, dynamically fine-tuning parameters to improve classi-fication performance. The optimized DMN model achieved an accuracy of 96.4%, outperforming conventional classifiers. Further-more, PSO-driven optimization significantly reduced computational complexity, ensuring faster convergence and improved general-ization. Comparative analysis shows that the optimized DMN exhibits a lower inference time (6.1 ms) than traditional models. Ad-ditionally, ROC-AUC analysis yields a score of 0.98, highlighting the model’s strong discriminative capability in distinguishing postural states. The proposed PSO-DMN framework presents a reliable and efficient approach for paralyzed person monitoring, offer-ing real-time posture recognition with high accuracy. The system’s ability to detect falls, classify different postural states, and oper-ate efficiently in real-time settings makes it a promising solution for healthcare applications, particularly in home and assisted-living environments.
DOI: 10.14419/bk2d1638
Authors: Dr. R. Sivaraman, Dr. Nithya. S, Dr. B. Srinivasa Rao, Rama Devi C, Dr. N. Venkatesh, S. Sharanyaa
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4409891715