Real time Gesture Recognition using Improved YOLOv5 Model

Abstract

Hand gesture recognition is part of the study of natural human-computer interaction. The objective of research in gesture recognition is to train computers to recognize and respond appropriately to human hand gestures. You Only Look Once (YOLO), is a deep learning model that is mostly used in the field of computer vision for real time object detection. YOLOv5 is an enhanced version of the YOLO algorithm, which is renowned for its fast and accurate object detection capabilities. This research paper explores the application of YOLOv5 architecture for real-time gesture recognition. The efficient architecture of YOLOv5 is utilized to detect and classify real time human hand gestures accurately and rapidly. A fine-tuned version of the YOLOv5 model is proposed that improves the performance of YOLOv5 with better results. Experimental performance shows that the proposed YOLOv5 model achieves a Mean average precision (mAP) value of 96.8% for real-time performance on a custom dataset of hand gestures. We evaluate our method on a custom-made dataset that outperforms state-of-the-art results.