Nighttime Traffic Detection Research Based On Video

Vehicle detection is an important part of intelligent transportation system(intelligent transportation system,ITS).Vehicle detector based on the video have become the center of the intelligent transportation system,and get more and more widely used. The video vehicle detection systems not only have a wealth of traffic parameters,but also have convenient installation,maintenance,event detection,and video monitoring and other advantages.There are a lot of scholars have done a lot of work on the daytime video detection technology,and its detection technology has grown up,can meet the needs of urban intelligent traffic system,but low intensity of illumination and light reflection at nighttime is the main challenge of video vehicle detector on realizing all-weather work.Traditional vehicle detection is mainly from two aspects to overcome insufficient light and all kinds of light interference difficulties at night:one is to detect the most obvious characteristics of vehicles at night- car tail lights,the robustness is not strong,and it is vulnerable to be influenced by the vehicle lamp irradiation and body boundary reflection,the traffic parameters of extraction is limited,the installation of cameras and shooting direction also have certain requirements,and the day and night mode switch is a difficult problem.The second is the hardware facilities of cameras,by choosing the infrared camera or improving detection area lighting conditions for motor vehicle information.This method added the system’s expenditure,and increase the difficulty of system installation and maintenance.Video image information itself can be distinguished is less at night,and the change of light intensity is big,which will cause target extraction difficult and high error detection rate.To solve the above-mentioned issue properly,thesis come up with a kind of frame difference and background difference combined with a background update extraction method.Firstly use the former method to get high brightness light area,in order to eliminate the night scene illumination under conditions of tremendous changes of halo.Secondly,use lamp brightness,color,shape, locationto identify possible light texture information,and then estimate the possibility of belong to the same car,and do the light matching,finally set up the virtual test line,to count a pair of bright spot that touch virtual test line on the road. The laboratory finding show that:This algorithm is concise,frame processing speed is fast,has the high real-time capability and robustness,even though in complicated weather(rain,snow, fog)and traffic situation,also has a higher detection rate.This method in the nighttime vehicle detection average detection rate is above 96%,can reach the demand of real-time detection of city road and highway.