Detection of Two-Wheeler Traffic Rule Violation Using Deep Learning

Abstract

Computer vision has become a potential research area due to its diverse applications. Object detection is probably the most challenging and complex well-known problem in computer vision. It has found many applications such as tracking objects, counting the number of objects, self-driving cars, and detection of vehicles. Over the past few years, two-wheeler accidents have gone up exponentially in India due to the negligence of traffic laws by the riders. Therefore, It is obligatory to find out more innovative ways of Detection and Tracking traffic rules violators to ensure the safety of bike riders. This paper proposed a framework to detect two-wheeler traffic rule violators such as helmet and non-helmet bike riders. Three models, YOLOv5, Faster RCNN, and RetinaNet, were compared and analyzed. Experimental result shows that YOLOv5 gives good results. Using pre-trained YOLOv5 model weights, an accuracy of 92.6% was recorded, proving the effectiveness of helmet detection.