Automated overtaking assistance system: a real-time approach using deep learning techniques

Abstract

Road accidents are one of the major causes of fateful deaths in Bangladesh. In most cases it is caused by Overtaking on highways or on regular roads. In terms of overtaking the major task is to decide whether the overtaking is safe or not. There has been a lot of work considering autonomous communication in this field. However, the challenges in terms of Bangladesh, need different and user-friendly solutions. Considering the interest of researchers in this area and to introduce a different evolutionary trend in Bangladesh we approached this research. In this research our basic concept is to suggest the safe overtaking decision to the host drivers considering an overall idea of the environment. Our work aims to decrease early and unfortunate deaths caused by vehicle’s abrupt overtaking on Bangladesh’s Highways by assisting drivers based on deep learning techniques. Furthermore, the Autonomous system considers communication between vehicles to decide safe overtaking within minimum time.V Vehicle detection and classificationdetection and classification is done using the YOLO model. After measuring the distance and relative velocity, our model suggests the decision by using the help of other significant models used in our research. For distance measurement, we used SegNet and a distance measurement model. In addition, for getting relative velocity we have used optical flow and also for checking whether the driver is on the right lane or not, we have used the PiNet model for lane detection. Moreover, we have no use of other sensors besides the camera and kept only one camera in our proposed system. So in future, users will get this autonomous system in their vehicles at low cost as our system proposes. Experimental results from the proposed system show that deep learning process is better in terms of our country.