| In recent years,as the technology of electrical information product is increasing and the cost of hardware is decreasing,which give the chances to the computer vision develop rapidly,meanwhile,the demands of all works are getting high,all of these make the intelligent computer vision becomes a warm research field both on theory and application,intelligent numbers of people statistics system based on video is one of the directions of research.The system of this paper mainly on the numbers of people statistics in shopping mall,make the reasonable distribution of shops though the count up the people of main gallery;count up the attracting rate and busy level of each section;make the effective assessment about the return of the marketing and sales promotion;display the state of passenger and changing trend,the security apartment can take actions in advance to the high people flows section,and they can watch the people numbers and image in real time,comparing with the traditional,it becomes more intelligent.This paper designed the hardware and software,for the hardware,the difference between the traditional one just on the App client,so we do not describe the detail here.The key point and the difficult point is detection method for moving body target,machine learning method for head recognition and the moving target tracking count,so,we separate from the following steps.Firstly,the environment of statistical analysis system is the gate of shopping mall,where is crowded,through the program comparison,we choose the Rossi et al.Proposed vertical camera to video,and detect the human head area.The advantage is when the pedestrian is close or in contact with each other,it still can extract the head information,as far as possible to reduce the leakage caused by covering.Secondly,using Frame difference method to extract the moving objects from video frames,then,the outer rectangle frame of the moving object is used as the detection region of the subsequent human head detection,the research shows that pedestrian head detection is based on HOG feature extraction and linear support vector machine(SVM)as a classifier is the better way to detect the pedestrian.Thirdly,Tracking the number of pedestrians in the video scene.Do the study on the classical HS optical flow method and LK optical flow method,LK optical flow method is more suitable for target tracking because its calculation method is high flexibility and less computation.For the Optical flow calculation of large displacement of spatial motion,decompose the image in Pyramid to improve the accuracy of optical flow vector,when counting,set the interested section in the image,and only count the numbers in this section,meanwhile,it can determine the direction of entry and exit accurately.It can according to the actual needs,set the interested region randomly,so as to improve the practicability of the system.Finally,for the developing of APP,this paper mainly introduce the development of APP client based on iOS,the develop process is:the sever save the JPG image of people number change on server database,then inform the remote control client,control client decode the inform,then through the protocol request the picture on the server in order to get a real picture of the results.In this paper,we use the technology of moving object foreground detection,machine learning based on head recognition and target tracking to achieve a specific application case in the system.At the same time,it can get the inform of people numbers changing and pictures on iOS devices.In order to verify the effectiveness of the proposed algorithm in this paper,the hardware configuration is not high and can make system efficient,real-time and reliability,and made the tests under the condition of a large number of different scenes and different numbers,the results show that the system can detect the pedestrian head accurately and effectively,And it has a good practical effect in tracking and counting,it achieve the desired design objectives. |
