More images

Abstract

This paper outlines the development of a conceptual hand gesture-controlled web game using a webcam, leveraging advanced computer vision techniques. Our conceptual system employs the YOLOv8 algorithm for real-time hand gesture recognition, transforming webcam-captured images into game commands with high accuracy and low latency. The study highlights the creation and annotation of a custom dataset containing 389 images, categorized into four distinct hand gestures. Extensive preprocessing and data augmentation were performed to enhance model performance. The YOLOv8 model achieved a mean average precision (mAP) score of 98.8%, outperforming existing methods. Our findings demonstrate the potential for webcam-based gesture recognition systems to provide intuitive and immersive gaming experiences, providing an affordable substitute for specialized hardware. Future work will concentrate on extending the dataset and exploring advanced deep learning architectures to further improve the system's adaptability and accuracy.

Contributions

• Conducted end-to-end research development, including ideation, implementation, and manuscript preparation
• Automated dataset collection using custom Python scripts
• Performed data annotation and augmentation using Roboflow
• Designed and curated a custom hand gesture dataset
• Trained and evaluated the YOLOv8 model for real-time gesture recognition
• Created research diagrams, system architecture, and visualizations
• Wrote, formatted, and finalized the full research paper for conference submission

Technologies