Research Of People Counting By Video Sequence Based On DSP

In the thesis, firstly we learned and researched the image processing method which is related to the people counting by video, summed up the good and bad points of various of counting methods by analyzing the object detection, segmentation and track method for present targets. After learned large number domestic and international documents about the people counting by video, the thesis introduced the present method of people counting by video.This thesis mainly researched the method of people counting by video which is used the bi-direction passage, and the video images are made by the camera which is mounted vertically at the top of the passage. After people detector, Segmentation and tracker for the pedestrian in the image, tracked every pedestrian, then improved the partitioning method for the counting area, I proposed the counting method in this thesis, and achieved the accurate counting for pedestrian.In detecting part, we abstract pedestrian target by temporal difference method, improve the detection results by morphological operation. In the part, a secondary segmentation method is proposed, the method divides the multi-box into sub-box which only include single pedestrian; In part, the thesis proposed the conception of Overlap Coefficient of Sub-box, achieved the stable track for sub-box. Through the pedestrian segmentation and people tracking methods, in the case of vertical shooting, the thesis solved the merge/split problem which influence the accurate pedestrian counting very much, also it is the problem for the researcher who used the vertical shooting method must to solve. On the base of stable tracking, the thesis proposed a tight pedestrian counting method, the method can handle the situation very well that the pedestrian tramping and walk back in the detecting area. In the end, removed the method in this thesis to the network video camera based on TMS320DM642, and made it to be a application module of the network video camera.