An Automatic Detection of Helmeted and Triple-Ride Motorcyclist with License Plate Extraction using Convolutional Neural Network

Abstract

Road accidents is one of the major cause for the deaths occurring in our country, to stop these mishaps the government introduced many rules and laws like speed limit, helmet compulsion, objection on triple riding etc., The government introduced penalties for traffic rule violators. As the police cannot be present everywhere a violation occurs. Now-a-days it is necessary to detect the helmeted and non-helmeted bike riders in order to save their lives. The identification of traffic rule violators is a highly beneficial yet daunting activity to ensure the safety measures due to numerous difficulties such as occlusion, lighting, poor quality of surveillance footage, varying circumstances, etc. In this paper, we present an approach that uses convolutional neural network (CNN) to distinguish helmeted and non-helmeted bike riders, especially concentrated on detecting triple riders also. The advances in profound gaining knowledge of models have essentially progressed the productivity of article discovery inside the direction of recent years. YOLOv2 model is utilized from the outset degree to identify diverse gadgets within the check photo. In the proposed technique, we utilize singular elegance discovery in preference to cruiser to improve the exactness of protective cap recognition inside the data picture. The trimmed snap shots of identified people are applied as contribution to the second segment of YOLOv2 which has been prepared on our helmeted photograph dataset. The non-helmeted pics are additionally prepared through utilising Open ALPR to evacuate tag. Test consequences show that the proposed strategy plays better while contrasted with different existing methodologies with 94.70 percentage defensive cap identity exactness.