Counting Of Different Densities Crowd Based On Video

With the development of economy and fast growth of urban population,dense crowd frequently trigger stampede events.So it's of great significance for public safety to obtain crowd statistical information.The current video-based crowd counting algorithms have good results,but most of them are based on single crowd density,and rarely distinguish the movement direction of large-scale crowd,which can't meet the demand for crowd statistical information who entering and leaving the public places such as parks and plazas.In this regard,this paper proposes different direction counting of different density crowd according to actual needs.Due to the large differences in characteristics of high-density and low-density crowd,in order to achieve the best counting effect of crowd who entering and leaving,selecting the different crowd counting algorithms in the different density crowd.After graying and smoothing the original image,the foreground image is acquired by the combination of average background modeling and background difference.Then the morphological processing is used on foreground image to denoise.The crowd density is preliminarily classified by the number of edge pixels,the ratio of edge pixels and foreground pixels,which lays the foundation for the classification of crowd counting algorithm.For the low-density sparse crowd,this paper adopts a crowd counting algorithm based on individual detection.In order to improve the accuracy of target tracking,an improved tracking method based on region matching is adopted for pedestrian targets in the virtual gate area.Finally,judge the direction of pedestrian movement by the bidirectional counting line and count pedestrians who entering and leaving.For the high-density occluded crowd,this paper adopts a crowd counting algorithm based on multi-feature fusion.Firstly,select the foreground pixel area,gray level co-occurrence matrix eigenvalues and SURF feature point as eigenvalues.Then sent the eigenvalues after perspective correction to support vector regression for training,and predict the crowd number by regression model.in order to count the different direction crowd number of high-density crowd,a pyramid optical flow method based on SURF feature points is proposed to judge the direction of crowd movement.Finally,fit the number of crowd in different directions according to the relationship between the feature points number in different directions and the predicted crowd number.Experiments show that the proposed different density crowd counting algorithm achieves the preliminary classification of low-density and high-density populations,and can automatically select the corresponding crowd counting algorithm based on the classification results.It has good counting results in low-density crowd,and can effectively distinguish the direction of crowd movement and achieve bidirectional counting in high-density crowd.It can meet the demand of bidirectional counting of different density crowd in practical application.