Research On People Counting Based On Interest Points

The statistics of people counting in video is an important content of crowd behavior analysis,which plays an i,mpo,rtant role in counting people in many fields,getting demographic data,allocating resources rationally,optimizing public management,and ensuring public safety.This article is aimed at counting people in video.In this paper,the main content and innovation embodies in:(1)the extraction of the image characteristics and feature vector was improved rather than using traditionally pixel texture or texture features to describe the crowd,but use feature points to describe instead.First,Use Fast feature point to extract the characteristics of pedestrians,and then use Surf feature points for f'eature matching in order to extract the feature points of pedestrian movement,finally construct feature vectors associated with the number of pedestrians using the crowd walking direction and speed,and use these feature vectors to reflect the relation between the pedestrian movement.(2)In the stage of people counting,after comparing linear fitting,support vector machine(SVM)method and so on,this paper chooses the support vector regression method.First,input the extracted characteristic vector datas into trainers,this paper mainly uses the Libsvm training,so as to get the data model and then using function model can automatically count each frame corresponding to the number of people in the video.In this paper,implenment the traditional people counting method based on pixel feature,and compare this method with the method of this article in the experiment,which shows that our method has a good effect.In this paper,it also develops the implementing system by using Visual C++language,and has studied it by doing the experiment,the results show that the method has a good effect on the people counting in the video and has high precision statistics,which can satisfy the real-time requirements basically.