Research On People Counting System Based On Digital Image Processing

To know the real-time passenger flow information in public places such as shopping malls and stations, can provide scientific basis for the administrator to distribute service resources and security assurances reasonably and make the best use of resources. The traditional statistics methods for passenger flow volume (e. g. infrared automatic detection counting, mechanical transmission type automatic detection counting, light screen sensor automatic counting, etc.) have problems in practical application, which, as a result, makes the statistical precision of passenger flow volume relatively low. So it is very necessary to find out an intelligent method of higher accuracy to count people at entrance. Based on the conclusion of predecessor’s researches, I put forward a guest flow statistics method in entrance and exit which combined image processing with neural network,And using VS2010and OpenCV2.4.4to prepare procedures related to.The main work of this study include the followings:(1) In order to solve occlusion condition concurrently occurred by two or several people when counting, this paper adopts the method of vertically placing single camera to get video frame. Adopt background differencing method when extracting the foreground image, considering the complexity of the scene, make use of Gaussian Mixture Model to build a background model, we can gain foreground image by video image of current frame deducting the background image when making background subtraction. The experiment results show that this method could avoid the effect of scene changing on extracting images effectively;(2)This article aims to construct the nerve network categorizer with the ratio of length and width, duty ratio, ratio of arc and Fourier descriptor and radius, and identify human target and multi-player parallel through nerve network categorizer;(3)In the section of human body target tracking, this paper uses the tracking method based on Camshift and Kalman filters. This method combines the advantages of Camshift tracking algorithm and Kalman filter tracking algorithm. It can avoid being dumped by the pedestrian target when the followed target was sheltered if using CamShift tracing algorithm only, which could also avoid the inconsistent condition between the tracked pedestrian target and actual position of pedestrian target when only using Kalman filter.