Real-time Implementation of YOLO V5 Based Helmet with Number Plate Recognition and Logging of Rider Data using PyTorch and XAMPP
S Maheswaran, R D Gomathi, S Sathesh, Deepan Kumar S, G Murugesan, Prakash Duraisamy
Abstract
Many road accidents occur in today’s busy life, and one can quickly become a victim. Most fatal road accidents are caused by not wearing helmet, which accounts for almost 80% of the fatalities. Wearing a helmet properly is therefore essential. However, children are taught from an early age that driving force is necessary to accomplish a task. To ensure helmet use among the public, it is necessary to impose fines in accordance with the motor vehicle act of the respective country. A vehicle's number plate must be identified to automate the fine collection process. An annotated dataset and images with and without helmets are used to train a model with YOLO V5. By using OpenCV, live video captured with a camera and separated into frames is processed using the trained model's weight files. To detect the frames, we load the weight file into PyTorch. The proposed system classifies the detected objects as "with helmet" or "without helmet." For images containing riders without helmets, a number plate recognition process is performed using the OCR technique and Plate Recognizer API. The extracted text is then stored in a database and displayed on a web using HTML and PHP. XAMPP used for back-end server. Based on our model, riders had a confidence score of 88%, helmets had a score of 85%, and number plates had a score of 80%, and helmets, no score. This model can be further enhanced by developing an OCR engine and increasing the security in the UI end with tokens, session etc.