Intelligent Traffic Signal Automation Based on Computer Vision Techniques Using Deep Learning

Abstract

Traffic congestion in highly populated urban areas is a huge problem these days. A lot of researchers have proposed many systems to monitor traffic flow and handle congestion through different techniques. But the current systems are not reliable enough to perceive traffic signals in real-time. Therefore, we aim to build a system that can efficiently perform real-time environments to solve the traffic congestion problem through signal automation. Since vehicle detection and counting are crucial in any traffic system, we use state-of-the-art deep learning techniques to detect and count vehicles in real-time. We then automate the signal timings by comparing the count of traffic on all sides of a junction. These automated signal timings sufficiently reduce congestion and improve traffic flow. We prepared a dataset of 4500 images and achieved about 91% accuracy by training it on Faster RCNN.