Helmet Detection Using YOLO-v5 and Paddle OCR for Embedded Systems

One of the best methods to lessen the chance of suffering brain injuries in a two-wheeler collision is to wear a helmet, but many riders choose not to do so for a variety of reasons, including comfort, convenience, or ignorance. By automatically detecting and identifying riders who are not wearing helmets, automatic helmet detection tools can assist in overcoming these difficulties. These tools may result in fines or other consequences that may serve as an incentive for riders to wear helmets. Thousands of images and video frames can be quickly scanned and analyzed using powerful image processing algorithms employed in automatic detection systems, yielding consistent, accurate results without the possibility of human mistake. The YOLO-v5 object detection technique utilizes deep learning to detect and identify objects in a frame of an image or video. With the help of this model riders without helmet can be detected, and their license plate can be extracted. Optical character recognition (OCR) technique called Paddle OCR can precisely extract and identify text from an image or video frame. With the aid of OCR, the license plates can be recognized and can be stored. The proposed tool accurately detects the license plate of riders without helmets by bringing together these two potent models. Both the YOLO-v5 model and Paddle OCR model were trained using custom datasets with 1000 images each and obtained accuracy by 91%.