Lane Navigation through Lateral-Longitudinal Control and Traffic Detection for Autonomous Vehicles

Abstract

Autonomous vehicle research is currently one of the trendings fields. This technology has immense potential to revolutionize present socio-economic dynamics. We address the notion of self-driving cars deployment in the context of Bangladesh. This thesis addresses mainly the perception and planning parts of an autonomous vehicle. The detection and tracking of objects around an autonomous vehicle is essential to operate safely. This paper can be divided mainly into two parts. Firstly, it presents the use of CARLA simulator for the purpose of lane detection and navigation of an autonomous vehicle and secondly, it shows the use of YOLOv5 for object detection. Both the longitudinal and lateral controls for lane navigation are done with the help of the built-in map in CARLA. And for object detection part, at first we have used SRGAN algorithm to enhance the quality of the images. Then using those enhanced images, YOLOv5 was used in order to detect objects in those images. For further improvement, these two different parts can be integrated by feeding the outcome of YOLOv5 to the CARLA simulator so that the autonomous vehicle can detect objects and obstacles.