REAL-TIME RECOGNITION OF ROTATIONAL HANDMOVEMENTS USING MEDIAPIPE AND OPENCV

Abstract

This study proposes a real-time gesture recognition system for detecting rotational hand movements (pitch, yaw, and roll) using MediaPipe and OpenCV for extracting features and classifying gestures. The method extracts angular features from hand landmarks and classifies gestures via a lightweight rule-based approach. The system was tested under twelve environmental scenarios, combining four distances (50 – 450 cm) and three lighting conditions, with 1,500 gesture samples from 10 participants. Results show 100% accuracy at 50 cm and 200 cm across all lighting conditions, with up to 93.3% accuracy at 300 cm, surpassing similar rule-based systems in performance. These findings demonstrate a low-cost, contactless solution for interactive systems