Research On Pedestrian Counting Method Based On Flow Mosaicking

Pedestrian counting based on videos is a favorite issue in computer vision. Only using a camera to capture a picture of specific area can achieve pedestrian counting of the monitored area. With the development of society, more and more cameras put into use in the community. A stable and reliable pedestrian counting method can extend the functionality and the range of applications in camera. Therefore, many researchers have proposed various methods of pedestrian counting. However, due to the scene complexity and diversity of people’s movement in practical application, the difficulty of pedestrian counting is greatly increased. These methods can achieve satisfactory results only in a particular scene.This paper focus on the studies of pedestrian counting base on Flow Mosaicking. Using an improved optical flow algorithm to estimate the velocity and direction when pedestrian crossing the detection line. We propose a velocity correction algorithm. It can more accurately reflect the overall pedestrian velocity and walking information after correction. The average velocity is used as thickness to crop the slice. According to ViBe background modeling method and improved active attentional sampling, we decrease the affect of illumination and shadow to slice cropping. Finally, with the method of linear regression analysis, we estimate the number of pedestrian in mosaic by the area and edge of mosaic. The results of experiment on various videos show that the proposed method in this paper can achieve relatively accurate estimation in real‐time.