Research On Intelligent Traffic Lights Based On Computer Vision

The economy of China grows fast.However,there are some problems.The traffic is getting worse and worse.The signal light at intersections is very important.It is the major tool for cars to pass through the intersection in order.Most of the current traffic light uses the fixed timing method.It does not behave well when the quantity of cars changes very fast.Traffic congestion is very common at some intersections.So it is very important to solve the traffic problem with a useful signal timing method.The visible light cameras are very common at the intersection and most of the current traffic light uses the fixed timing method.So this paper uses a effective method to detect the vehicle queue length at the intersection with the computer vision technology and finishes the optimization of the current signal timing so that it is very helpful for ITS to perform better.Main parts of my paper are as follows:(1)Do the summarize of Traffic Control Theory which is used in this paper.Firstly,the paper introduces some parameters with which the simulation of the optimization of signal timing are carried out.Secondly,it introduces several main types of the method to control traffic signal and analysis the pros and cons.With the comparison to those methods this method shows the superiority.At last,it introduces some mathematical models of the distribution of traffic flow,so we can use the models to do the simulation.The Poisson Model is used in this paper.(2)Introduce the detection method of vehicle queue length.At present,many cameras are installed at the intersection to take video of vehicles.It uses a effective method to detect the vehicle queue length at the intersection with the computer vision technology.Background subtraction is used in this paper.But it improves the background subtraction method on two points: 1)Reliability.It finds a rectangle area in which the length and width are both 80 pixels and extracts the texture feature of the area.Then it compares the feature of current Image to the feature of background Image in background library to find the background Image which is most similar to current Image.2)High speed.It is not the whole Image but the special area where the width is 10 pixel.So the quantity of the pixels to be handled is less than 5% of those in whole Image and the speed of the detection is very high.After the experiment,it proves that the detection of vehicle queue length has high speed and is also very reliable.(3)Make a solution to do the optimization of the traffic signal timing.After the detection of vehicle queue length above,this paper uses a adaptive signal timing method to optimize the traffic light.It aims to maximize the quantity of vehicles that pass through the intersection.It uses the proportion of vehicle in different lanes to distribute the delay green time.This paper makes a simulation at single intersection where there are 16 lanes and 4 phases.There are 4 simulation scene: 1)the arrival rates of vehicle keep the same and low.2)the arrival rates of vehicle are the same and high.3)the arrival rates of vehicle in different directions are the same but the arrival rates of vehicle which are in one direction but different lanes are different.4)the arrival rates of vehicle in different directions are different and the arrival rates of vehicle which are in one direction but different lanes are also different.After the simulation,it proves that the method is effective.But it needs further research.(4)Summarize the main points in this paper and introduce the further research to be done.