Helmet Detection and License Plate Extraction Using Machine Learning and Computer Vision

Abstract

AbstractThe exponential rise in motorization has resulted in an exponential rise in road accidents and fatalities. Non-helmeted motorcyclists contribute to major roadside accidents. Helmets must be worn by motorcyclists to prevent such horrendous accidents. We need a large workforce of traffic police to monitor and ensure the safety of motorcyclists by penalizing motorcyclists for not wearing helmets and this activity costs a major chunk of their time. Identification of motorcyclists without helmets in real-time is a crucial task to prevent the occurrence of accidents. This paper aims at identifying motorcyclists without helmets and extracting their motorcycle's number plate using an automated system. In recent years the accuracy and performance of the object detection models have significantly increased with the help of deep learning. Some of the advanced features in YOLOv3 are a feature extractor network with multi-scale detection and some changes in loss function combining detection and classification in a single architecture. In this project, the main principle involved is object detection using deep learning at three levels. The objects detected are person, motorcycle at the first level, helmet detection at the second level, and license plate detection at a third level all using YOLOv3. The license plate is detected and a cropped image of the license plate is used to extract its digits using OCR. We have used the above-mentioned methods to build integrated systems for helmet detection and license plate number extraction. The end of this paper suggests some future advances to the License Plate recognition system.KeywordsHelmet detectionNumber plate recognitionTransfer learningYoloV3