Dynamic road traffic prediction and Automation model for traffic Management system

Abstract

Due to the rapid development of urbanization, the road traffic newlinecongestion in the metropolitan areas has been increased in leaps and newlinebounds. The traffic signal lights which are meant to regulate traffic at newlineintersection tends to act as bottlenecks since they follow a static, fixed newlinetiming interval between the red, yellow and green lights. Though newlinedensity-based traffic signal timing has been proposed and recently newlineadopted in several modern cities, the congestion avoidance can still be newlineameliorated if a hybrid approach is evolved involving novel newlinecombination of algorithms in optimizing traffic signal timings. The newlinetraditional traffic controllers have to be replaced with intelligent newlinecontrollers that are capable of processing and executing algorithms that newlinecan recommend a suitable optimized timing pattern for red, yellow and newlinegreen based on incoming traffic rather than act as simple counters. newlineAny intelligent traffic regulatory system will typically consist of 3 newlinecomponents. The first component will be a data acquisition module that newlineautomatically counts the number of vehicles entering and leaving an newlineintersection which will be done through an array of cameras. The second newlinecomponent will be the data processing module that will rely on newlineintelligent algorithms to recommend a suitable optimized timing for the newlinetraffic lights and the third will be a control module that actuates the red, newlineyellow and green lights. newlinevi newlineThis research finds its place in the data processing module to newlineaddress the traffic signal timing optimization problem by proposing two newlinenovel algorithms, viz. Adaptive Quad-agent Multi Directional Queuing newlineSystem (AQMDQS) which is a hybrid version of the conventional newlinequeueing technique and Multi-Agent Multi Deep Q-Network newline(MMDQN) that applies Q-Learning as the research base. The Deep newlineQ-learning technique is used to optimize the traffic light on/off patterns newlineat intersections in order to achieve maximum traffic throughput. It is newlinecapable of finding optimum policies for any traffic situation without newlineroutine feature extraction.