Driver Hand Gesture Detection and Recognition in Indian Road Scenarios

Road safety remains a critical global issue, with traffic accidents causing over a million fatalities each year. In India, where traffic management largely relies on hand gestures from drivers, developing an effective vision-based system is essential. Real-time, accurate gesture recognition for traffic control faces challenges such as varying lighting conditions, complex backgrounds, and adverse weather conditions. At the same time, traditional models struggle with false detections that hinder the effectiveness of intelligent traffic management. To tackle these challenges, we propose a real-time hand gesture detection and recognition system that utilizes prior vehicle regions to localize the hand signs made by drivers accurately. By introducing a multi-scale feature concatenation model within the YOLOv11 framework, we can identify small hand regions and handle partial occlusions more effectively. Experimental results demonstrate a significant improvement in both detection and recognition tasks, achieving a mean Average Precision (mAP@50) of 98.26%, a precision rate of 98.67 %, and a recall rate of 97.27 % for detection. For recognition, we achieved a precision of 76.50 %, a recall of 90.56 %, and an F -score of 83.03 % on a dataset of challenging scenarios commonly encountered in India, such as hand signals from bikes and cars. The proposed model enhances intelligent traffic monitoring systems by improving road safety, reducing the need for human intervention in traffic control, and enhancing communication among vehicles, pedestrians, and between cars themselves in dynamic environments.