Helmet Violation Detection using YOLO v2 Deep Learning Framework
P. Sridhar, M. Jagadeeswari, S. Harini Sri, N S Akshaya, J Haritha
Abstract
Detection of helmet violations is important for riders' safety. Motorcycle accidents are the most frequent road accidents, and they often result in significant damage. Most regions instruct the use of helmets by motorcycle riders, but most people decline to fulfill the law for various motives. Using a vehicle helmet decreases the percentage of deaths to motorcyclists in highway vehicle accidents by around 42%, hence why governments have implemented mandated helmet usage. Besides, following helmet legislation is limited, especially in developing regions. So, we aim to observe whether the person wears a helmet or not, using YOLO v2 deep learning framework. We establish the improvement of a strategy using deep convolutional neural networks (CNNs) for revealing motorcycle riders who disobey the laws. This includes detection of motorcycle, helmet vs. non-helmet classification. The proposed YOLO v2 architecture yields better experimental results compared with traditional algorithms.