Abstract: Recent advancements in computer vision and artificial intelligence have significantly improved the way humans interact with machines, enabling alternatives to traditional input devices such as keyboards and touchscreens. This study introduces a real-time, touchless media control system that operates entirely through hand gestures, using only a standard webcam without the need for specialized hardware.
The system utilizes the MediaPipe Hands framework to detect and track 21 three-dimensional hand landmarks in each frame. These landmarks are converted into a 63-dimensional feature vector that represents the hand’s spatial structure. A supervised machine learning pipeline was developed using five different algorithms: Random Forest, Support Vector Machine, Multilayer Perceptron, K-Nearest Neighbours, and Gradient Boosting. The models were trained on a custom dataset consisting of 2,700 labelled samples across nine distinct gesture classes.
Among the evaluated models, the Random Forest classifier delivered the best performance, achieving a test accuracy of 97.4% and a macro F1-score of 0.971. The system maintains real-time responsiveness, operating at approximately 28.6 frames per second on a standard laptop without requiring GPU support. Recognized gestures are translated into system-level media commands such as play/pause, volume control, track switching, mute, and full screen mode through a cross-platform interface.
The system was also tested under different lighting conditions, showing only a minor drop in accuracy of about 3.3% in low-light environments. Overall, the proposed approach is efficient, accessible, and platform-independent, making it a promising solution for touchless interaction in applications such as smart environments, healthcare systems, and assistive technologies.

Keywords: Hand Gesture Recognition, Computer Vision, Human-Computer Interaction, Touchless Interface, MediaPipe, Real-Time Gesture Detection, Random Forest, OpenCV, Machine Learning, Accessibility Technology.

Downloads: | DOI: 10.17148/IJIREEICE.2026.14391