Reinforcement Learning in Urban Network Traffic-signal Control

Abstract

Traffic-signal recognition and anticipation are essential for advanced driver-assistance systems. Due to its superior performance in data categorization, deep learning has gained significance in vision-based object identification in recent years. When examining the application of deep learning to develop a high-performance urban traffic-signal detection system, the input image's colour space, as well as the deep-learning network model are examined as part of the system's primary components. Using distinct network models based on the Faster R-CNN algorithm and colour spaces in simulations helps the RGB (red, green and blue) colour space and the Faster R-CNN model detects the method of network target. A series of fundamental convolutional networks is used depending on pooling layers to extract the features of maps of images for training datasets, where the data may be used to develop a system for traffic-signal detection and create a new traffic signal that requires image recognition. KEYWORDS: Bounding boxes, Faster R-CNN, Modelled environments, Simulation, Traffic-signal detecting system.