The Study On Counting Passengers Based On Machine Vision

Recently the industry of intelligent video surveillance was activated by the rapid development of the computer vision technology,which contains huge amounts of profit,so more and more enterprises and research institutions come into this filed. Under this background,counting peoples who comes through a ROI area or line based on image processing,pattern recognition techniques were gradually developed. Due to the number of people in the shopping malls,airports,bus stations,subway stations and so on plays an important role in intelligent management and decision, so more and more researchers are interesting in this filed. In this paper, we have researched the topic of counting passengers getting in a bus and proposed a novel approach for it.In our surveillance system the camera overlooks the scene,taking advantage of the special angle of view,people head detection is more efficient than people body detection, because head is more like a solid object than body, and occlusion probability is even smaller among heads than bodies. In this paper, we use Histograms of Oriented Gradients for head feature descriptor and establish the linear classification model for head based on linear support vector machine, through this model combined with multi-scale detection window slipping search approach, we can map the original image into a new space, calling the result image as score-map image. Processing score-map image simply we can achieve the head information in the original image.We proposed an approach for multi-head tracking-by-detection in a particle filtering framework through which can obtain head trajectory. Next, by analysis of head trajectory simply, we determine the behavior of the passengers getting in the bus, then,count the number of passengers automatically.Experimental results show that our algorithm has a high statistical accuracy.