Development of Next Generation Traffic Signal using Edge Computing

Abstract

The proliferation of surveillance cameras distributed in public areas presents numerous network and processing infrastructure difficulties in the smart city scenario. Hundreds of cameras are required to saturate the entirety of the city. In most smart city applications, powerful systems are responsible for processing huge video feeds. On the other side, the number of cars is increasing worldwide, creating major problems like traffic congestion, poor quality of air, time and fuel loss. Developing an Intelligent Transportation System (ITS) is becoming a viable solution for many existing problems. ITS is the intersection of different technologies concentrating on service and application delivery to monitor and control the transportation system, making it more pleasant and secure. This study aims to develop an efficient Deep Learning (DL) based tracker for traffic monitoring and dynamic control of traffic lights according to the vehicles’ density on each side. We study, analyze and implement the state-of-the-art methods of object detection and realtime tagging. We selected two datasets, Boxy vehicle & PASCAL VOC 2012 dataset from which we get the precision of 93.40%, 55.15% respectively. Also, we make our own labeled ’toy car’ dataset for the prototype from which we obtain a precision of 96.15%. Moreover, using edge devices, we build a small-scale prototype of our model, deploy it in a small traffic area, and validate it. This study can be expanded to improve the accuracy of the detection by updating the weights of the model in real-time rather than using a pre-trained model.