Research On Video Analysis For People Flow Automatic Counting System

People flow counting system is one of the most active research topics beingvaluable for both theoretical and practical research in computer vision especially hasan important practical significance in the traffic safety, traffic statistics and the stationmanagement. However, because of people usually wear various and movingirregularly so that the traditional method are not able to count exactly and our researchon people flow counting system becomes a challenging work. There is not a valuableuniversal theory works. We propose an effective people flow counting algorithm usingmultiple evidences based on body geography and image features.First, we develop a new color background subtraction method based ontraditional method to get more color information for the human targets. Then thechanged method of shadow removal and a developed mean filter are introduced tomake a preprocessing for people imagesThen, people flow counting algorithm for single frame image. We compareddifferent skin color models and Hough transform model to detect heads of people,then, we propose a new method for head detection based on body geography(BFR)and a new algorithm using horizontal projection analysis to divide multiple targetsinto individuals for the case that people overlap laterally in the monitoring area. Weproposed a special image segmentation algorithm called T model segmentationalgorithm to count the number of people in the current frame image for the case thatpeople overlap in a row in the monitoring area. The movements of centroids are usedto get the direction of human walking. Last, we make the counting algorithmcompletely for single frame image.Finally, we introduce a method using multiple evidences analysis based on videosequences for people flow counting system in this paper. At the same time, we test the whole counting system using video sequences obtained in different scenes. All theresults show that our method is effective to solve the problem that people wear hats orwalking together. When the number of left-right overlapping people is less than6, thenumber of before-after overlapping people is less than3and the whole number is lessthan10, the accurate rate of people flow counting system is in93%above. Themethod counts20to21fames in one second and meets the needs of the project.