COMPARATIVE ANALYSIS OF REAL-TIME GESTURE RECOGNITION METHODS BASED ON MEDIAPIPE, OPENCV, AND YOLOV8

Abstract

The subject of the article is real-time gesture recognition methods based on computer vision, designed for integration into human-computer interaction (HCI) systems, particularly for device control via a webcam. The purpose of the work is a comparative analysis of the efficiency of three modern methods MediaPipe, OpenCV, and YOLOv8 by evaluating their performance using key metrics (FPS and Detection Rate) for detecting basic gestures. The article solves the following tasks: identifying the impact of external factors (lighting, background) on gesture recognition, implementing algorithms for three basic gestures (open palm, index finger up and down), conducting experimental testing and modeling in the Python environment. The following methods are used: computer vision and image processing (in particular, segmentation, keypoint tracking, and object detection); machine learning based on CNN (LeNet, YOLO); analysis of datasets and performance metrics (precision, recall, mAP); simulation modeling in real conditions using the libraries mediapipe, opencv-python, and ultralytics. The following results were obtained: a comparative analysis of the methods was conducted, where MediaPipe provided the highest accuracy (95% Detection Rate), OpenCV—the maximum speed (50.7 FPS), and YOLOv8—a balance for limited resources (73% Detection Rate); recommendations for hybrid approaches to optimization were proposed. Conclusions: The developed comparative analysis of real-time gesture recognition methods demonstrates that MediaPipe effectively eliminates interference from variable lighting and background, achieving a stable accuracy of 95%, while OpenCV optimizes processing speed to 50.7 FPS. Modeling was performed in Python with visualization of the results.