Research And Realization Of Developed People Counting And Tracking Algorithm Based On Video

Computer version is a subject that aims at realizing people recognition、tracking、behavioranalysis and so on. This paper mainly researches the head extraction and tracking in two casesbased on the video collected by camera. These two cases have the differences between the cameraangles. The first case has the vertical downward view, while the second one has the view that has acertain angle in both vertical and horizontal directions.By analyzing and identificating kinds of head characters, it can realize head extraction. Forkinds of reasons, such as the angle of the camera and the dress of the people could be different, orthe light would bring shadows, the head detection has some difficulties. The difficulties may be thehead characters are not general, or the heads are connected and hard to make a distinction. Aim atextracting the head, this paper uses the method that sets two kinds of models for head detection andswitchovers them according to the parameters of camera. For the first model, the combination of theextraction of hair region and the gradient hough transform is taken to extracte the head, while thecombination of the extraction of hair region and skin region is taken for the other model. What’smore, for the second model, the detection whether the skin region is next to the hair region or notshould be done.This paper adopts the area matching algorithm based on Kalman filter for tracking. TheKalman filter was used to update those targets which disappear temporarily. For the combination ofthe background difference algorithm with the frame difference algorithm that used for the extractionof moving foreground, this tracking algorithm can remove the interference of the static target. Also,this algorithm sets the cover area between the targets in adjacent frames as the key property, andsets the histogram of color and the distance between the central point of two targets as thesupporting properties. Then it is possible to handle efficiently the emergence of new targets、thefusion of target、the split of target and the exit of target, and make it possible to solve the overstatingproblem in the target splitting part. In the end, this paper achieves a system which can count thenumber of people of exits and entrances.