A Real-Time Hand Gesture Recognition Based on Media-Pipe and Support Vector Machine

The rapid advancement of technology has led to Hand-Gesture Recognition becoming a crucial element in Human-Machine Interaction (HMI). In recent years, there has been growing interest in employing visual recognition techniques to enhance the precision of gesture discrimination. This work introduces a hand-gesture recognition system founded on visual recognition; the research encompasses three distinct scenarios. The first scenario involves the creation of HGR using 50 images representing five fingers: the thumb, index, middle, ring, and pinkie. The second scenario expands upon this by incorporating 200 images for the same set of fingers as in the first scenario to increase the size of dataset. The final scenario encompasses the previously mentioned fingers and includes counting gestures, totaling 2000 images. In all scenarios, images are gathered for both left and right hands through real-time video capture. Feature extraction relies on 21 hand landmarks inspired by actual hand joints, and the MediaPipe library, an open-source hand-tracking tool, is employed for this purpose. Lastly, the classification of hand gestures is accomplished using the Support Vector Machine (SVM) algorithm. Among the experiments conducted, the third scenario achieved the highest accuracy, reaching an impressive 97.8% for ten hand gestures instead of 97% for five hand gestures.