Vision-Based Automated Traffic Signaling

Abstract

Traffic management has become one of the most complicated issues of recent times in metropolitan cities. Conventional traffic signaling systems are pre-programmed and alternate between red and green lights without any estimation of traffic. This signaling methodology leads to problems during peak hours at the intersections, where traffic ratio in a few lanes are dense when compared to others. Therefore, an efficient model is needed, which can manage the traffic flow at a certain point. The proposed model offers a solution using the CCTV footage from signal cameras to decongest traffic, based on a live estimate of traffic density. A state-of-the-art Deep Neural Network algorithm determines the number of vehicles and their type at a particular signal for object detection called You Only Look Once (YOLO), as it provided speed and accuracy in real-time. Based on vehicle count and road associated parameters, traffic density is computed to provide a dynamic extension of signaling time for a particular lane. Therefore, time saved from empty lanes is used to clear traffic on other busy lanes.