Intelligent Traffic Detection System Based On Color Histogram

It becomes to a trend to manage the transportation system with the powerfulinformation-based methods. From the 80's years of the last century, ITS has been rapidlydeveloped. Facing the growing information needs, ITS can not only be used to manage andplan the traffic efficiently with the powerful computers, but it can also get more and moreinformations with the advanced automation equipment. And all of the works are very hardto be done by human. These information is necessary while using it to optimize trafficmanagement and reduce human resource and management cost.ITS combines the information technology and control technology, communicationtechnology and sensor technology and other advanced technologies and under the systemtheory method, established a tridimensional, efficient and rapid traffic information network,it can improve the use ratio of the existing transport resources. Supply safer and higherquality services for each participant of the traffic and, thereby gain tremendous social andeconomic benefits.The traffic video detection is an important part of ITS, ITS system widely using thecomputer vision and image processing technology. Video image processing is used on thetraffic detection is more advantageous, and it can take more information, intuitionistic, andthe hardware's cost is not too high, and very popular. With the computer software andhardware development as well as the video processing technology and the development ofcomputational intelligence, video traffic detection gradually overcome the small amount ofstorage, real-time low, cost, and lack of information and is developing towards systematic,high-speed and information-sharing, and more and more be used in practical applications,security and reliability, and are also increased gradually as a very important part of the ITS.This paper presents a color histogram-based traffic detection system. The developmentenvironment we use is Vistual Studio 2008, combined with Microsoft's DirectShow VideoProcessing Library and Intel's OpenCV image processing library when developing. We usethe Directshow to play the video and to get the frames from the video one by one, while inthe image processing the OpenCV library is used to achieve the common algorithms.After we get the frames from the video using DirectShow's VMR9 class, we first grayand smooth the frames as pre-processing operation, and then use the combination ofbackground difference and interframe difference to find the interest area where the vehicleslocated. And finally obtain the interest area in binary values. After we get the binary images,we then conduct a morphological filter on them, to eliminate the noise and fill the airspace.We use the dynamic background updating strategy while doing background difference operation, to ensure the computing efficiency, and to adapt the changing of light anddisturbance of surroundings, and to improve the adaptive capacity of the system.Analysised and discussed the object detection technology based on continuous frame,compared and weighed the various algorithms has been achieved, we use the line- detectionmethod on detecting the vehicles and the method based on the adaptive rectangular box.After the discovery of vehicles, we establish the vehicle model, and then initialize the colorhistogram of the vehicle, it has 16×8×8 of the HSV color vectors which can be well usedfor recording the color characteristic when the vehicle has just entered the visual field.At the beginning of the system we do the camera calibration by substitute the pair ofpoints measured before, so we can get the mapping from the point in the real world to thepoint in the image. Then we can get the speed of the vehicles according to the mapping.The faster the speed, the smaller the angle of the forecasted place, and the more the framesin all the forecasted directions, and the longer the forecasted distance. We calculate theassessing function of all the forecasted places, and found which of the places has the bestresult, and we make it the new position of the vehicle in the new frame.After we get the vehicles'speed, we do the vehicle tracking based on movingprediction. The prediction area is got according to the speed vector of the vehicle. The twoparts of the information mix in a applicable ratio, and only when the color of the vehicle isclose to the forecasted place and the interest area of the forecasted place is enough. Theforecasted place would be chosen as the next location of the vehicle.We use the color histogram we established, and with the help of the binary imagegained in the step of vehicle detection, we do a vehicle tracking operation based on theforecast of perspective position. In this step, we also used a adaptive rectangular boxmethod just like which is in the step of vehicle detection to make the result of the vehicletracking more accurate. In addition, we also handl the situation of vehicles'adhesion. Byanalyzing the occurrence of several possible vehicles'adhesion, present the vehicle divisionmethod based on scan line and the vehicle chain to deal with the adhesion.We conduct experiments under multiple situations, we get the speed of vehicles, aswell as traffic flow, traffic conditions,traffic violation and other macro-traffic information.Experimental results show that using the above method of traffic detection system, whileensuring real-time circumstances, can effectively detect the vehicles, and the accuracy ismore than 90% rate, and can effectively extract real-time speed of the vehicles, and thetraffic flow of the whole day and other informations. And it can also detect some kind oftraffic violations.Due to some of the features and the limitations of the video itself, there are someproblems which are hard to be solved, and become the focus of the further study later inthis direction. As the environmental situation is more complex, the weather conditions,lighting conditions are constantly in change in the background, and because the wind is alsoa cause of the vibration of the camera, in such circumstances, methods which is valid in the the experiment may be very different with the actual results, so it is necessary to improvethe accuracy of the algorithm and improve the popularity of the algorithm in the futureresearch.