Research On Traffic Flow Detection Algorithm Based-on Fixed Camera

Intelligent Transportation System is considered the main stream direction of urbantraffic development. The foundation of Intelligent Transportation System is to acquirethe traffic flow and vehicle speed of traffic. In this paper, we mainly work on theapproach of traffic flow detection based on fixed camera. And pay more attention tomoving vehicle detection, shadow elimination and vehicle tracking and counting. Thispaper consists of five main sections.First, video image preprocessing. Described and compared the median filter andmean filter that is generally used in the image processing. Introduced the principle ofRGB color space and graying. Analysis the method of adaptive threshold determiningOTSU, and introduces the morphological processing operation such as erosion anddilation, open operation and close operation.Second, moving vehicle detection. Summarize and analyze the commonalgorithms of detection. Proposed an improved method that combined frame differencealgorithm with GMM, this method updating the learning rate in temporal and spatial.Aiming to the ghost phenomenon when GMM initialization and the problem of largeamount of computation, the algorithm divide the video frame into different regions ofspace by frame difference algorithms within T frame, then value the different learningrate in different regions. Update the GMM background model in the interval of timedomain. Experiments shows that this algorithm can remove the ghost in a relativelyshort time and through a spacing frame update improves the real-time property.Third, shadow detection and elimination. Describes the principal of generatingshadow and the shadow features. Introduced the principle of shadow detection basedon color space and on texture respectively. Compared the four algorithms of shadowdetection by experiment. And summary that shadow detection algorithm based on HSVcolor space is the best. Introduced the shadow removal and reconstruction work in detail.Fourth, vehicle tracking and counting. Research on the tracking algorithm basedon the Kalman filtering. Because of the centroid positon and the tracking window sizeof a same car is similar in two adjacent frame. We matching and tracking the car usethis characteristics combined with Kalman filtering. Fifth, the realization of system platform. Design a traffic flow detection andcounting software system, and test the sample video using this system to statistic thetraffic information. The result shows that the system can meet the basic needs of vehicleflow detection of real road scenes.