Efficient Real-time Traffic Management and Control for Autonomous Vehicle in Hazy Environment using Deep Learning Technique

Abstract

Real-time Traffic control and lane detection have significant impact on flow of traffic in a particular area. It involves detecting the objects which include different type of vehicles. In development of efficient traffic management, the major challenge comes into play when there is hazy environment. Major challenge to the implementation of idea is to accurately determine the actual number of vehicles passing by from each lane in in foggy, hazy or dusty weather. Camera and hardware needs to be installed at the traffic junctions for object detection and the camera systems of the autonomous vehicles help in lane detection. It is also important to handle the case of starvation in-order to prevent long wait times. Real-time determining of objects by object detectors is done using some deep learning techniques from which YOLOv3 is prominent one. We have devised an algorithm which efficiently manages traffic in hazy environment using dark channel for dehazing and applying efficient algorithms for traffic management and control. We have validated our approach with real time video in hazy conditions and achieved 95% accuracy in hazy environment as compared to 89% accuracy without applying the proposed method.