Smart Wheelchair Control Based on Spatial Features of Hand Gesture

Mobility is a critical aspect of daily life for individuals with physical disabilities, especially those affecting leg function. Traditional electric wheelchairs rely on joystick controls, which may be challenging for users with hand disabilities. This study proposes a novel control system for electric wheelchairs using hand gestures, leveraging a Convolutional Neural Network (CNN) integrated with the MediaPipe framework. The system captures real-time hand movements through a camera, processed by a CNN model to classify gestures into commands for the wheelchair. The model achieved an outstanding 100% accuracy, precision, recall, and F1-score during validation, suggesting excellent performance in controlled environments. However, real-world testing with new users revealed a slight decrease in performance, with success rates ranging from 92 % to 96.1 %, indicating the system's robustness and unreliability across different conditions. The system demonstrated an average frame rate of 6.3866 FPS, which may require optimization to ensure responsiveness in real-time applications. These results suggest that the proposed system offers a promising alternative to traditional control methods, enhancing autonomy for individuals with disabilities. Future improvements could focus on optimizing frame rates and improving performance in diverse real-world environments.