Automated Detection of Traffic Rule Violation using Deep Learning Techniques

The number of accidents on the road has increased as a result of more vehicles and less awareness among the populace. Motorcycle riders are involved in more deadly incidents than drivers of other types of vehicles. Many governments have mandated that motorbike riders wear helmets and follow traffic laws in order to limit the number of fatalities caused by accidents. The proposed system is intended to help law enforcement officials find motorbike riders who break traffic laws and also to raise awareness among them about the importance of following the rules. The system receives real-time traffic video and uses the Yolov5 model to identify objects in each frame. Based on a confidence score threshold, one of the frames featuring the motorcycle and the rider is chosen. The person's region of interest (ROI) is then used in a VGG16 classifier stage to determine whether the person is wearing a helmet or not. While a craft model is used to find the location of the licence plate from the motorcycle ROI. The identification of the licence plate aids the authorities in discovering the identity of the offender. A user interface is used to present the results of the classifier and text detection models to the user. The system exhibits an overall accuracy of 90 % when evaluated in a variety of scenarios.