Research On Traffic Signal Light Control Algorithm Based On Visual Information Mapping

With the rapid development of the national economy,the continuous expansion of the city scale and the continuous increase of the population,traffic congestion has become a focal problem that needs to be solved urgently.Traffic signal control optimization is one of the main ways to improve traffic flow,and it has become a research hotspot in the field of transportation in recent years.At the same time,reinforcement learning can intelligently control the phase of traffic lights by observing the traffic flow,which has received extensive attention from researchers in the problem of traffic signal control.At present,most traffic signal control algorithm experiments are mainly carried out through the traffic simulator SUMO(Simulation of Urban Mobility,SUMO).There are two main problems: at the application level,the 2D planar traffic scene in SUMO is far from the real scene in real life.The relevant traffic flow information in the experiment is obtained through the external interface provided by SUMO,which is quite different from the way of obtaining traffic flow information in the real scene.The existing traffic signal control research has a certain gap from the application.At the algorithm level,complete information on traffic flow is difficult to obtain and the performance of the algorithm will vary in different traffic scenarios.Aiming at the above problems,this paper proposes a traffic signal control algorithm based on visual information mapping.Aiming at the first problem,this paper builds a traffic signal control system platform based on visual information mapping.With the help of the game engine Unity,by designing the information flow mapping between the traffic simulator and the Unity three-dimensional environment,the traffic road network,vehicles,traffic lights and other elements in the traffic simulator are mapped to a three-dimensional scene in real time,and at the same time analogous to the real scene,traffic signal control system platform will also set up a surveillance camera above each traffic light,and then provide video stream information to the outside by scripting the surveillance camera to restore the real traffic system to the greatest extent,provides an experimental platform that is closer to the real scene for the traffic signal control problem.Aiming at the second problem,this paper designs a traffic signal control algorithm based on visual information mapping based on the deep Q network.It only takes the image of the simulated real scene as the input,the purpose of optimizing the traffic flow is achieved by extracting the visual information in the image and then controlling the phase of the traffic signal.On this basis,by designing a phase gating mechanism and a multi-replay memory unit,the performance of the traffic signal control algorithm in this paper in different traffic scenarios is improved.In order to verify the effectiveness of the algorithm proposed in this paper,this paper is compared with eight current mainstream traffic signal control methods.The experimental results show that the method in this paper can achieve a high traffic light control effect when it is close to the real scene.At the same time,the effectiveness of the phase-gating mechanism and the multi-replay memory unit is verified by designing ablation experiments.In this paper,a traffic signal control system platform based on visual information mapping is built and a traffic signal control algorithm based on visual information mapping is designed.The traffic signal control problem is studied on the basis of being closer to the real scene,which is conducive to the use of traffic signal control algorithms.