Small Scale Crowd Behavior Classification By Angle Cosine Distance Weighted Network

The crowd behavior analysis is important and difficult research in the field of intelligent surveillance,scholars pay wide attention in this field at home and abroad.In past research,the most research of crowd behavior analysis,on one hand,is that detect crowd behavior or crowd abnormal,this generally is focus on that extract some whole features on the overall to detect crowd behavior,This kind of situation mostly suitable for massive crowd behavior,On the other hand is that understand a single human action and the interaction between the two parson,this research is usually focus on the characteristics of human body,location,track,etc to understand some behavior,this situation is most suitable for most small size crowd behavior.But both the crowd behavior analysis ignores some information,massive crowd behavior analysis process micro information less,minimum-size crowd behavior process macro information is not.Crowd behavior research of small scale,sports scatter in the daily life is less.In order to solve this situation,we put forward to researching small crowd behavior by combining macro and micro information.Complex network is composed of nodes and edges as a graph,which nodes and edges reflect the microscopic information and the network characteristic parameters reflect the macro information of complex network,therefore complex network is a well tool to reflect the information on the macro with micro information.This paper puts forward a way to use angle cosine distance weighted complex network to identify small crowd behavior.The principle of weighted complex networks for crowd as follows: First,getting the motion trajectory of video moving object in the scene information,judging two goal relative direction on the basis of mixed product judgment criteria model,build the direction coefficient influenced relations between the two objectives;Secondly,measuring a single target specific offset according to the vector model,it quantitatively evaluate cosine angular distance to express human individuals and connection degree between individuals combined with direction coefficient,and build weight of complex network based on cosine angle distance;Finally,the characteristic parameters of gather,meet,together,separation,dispersion crowd weighted complex networks including the shortestpath length and the weight distribution are extracted and compose of feature vectors from the macroscopic perspective to recognize and classify crowd behavior.The experimental results showed that the five kinds of typical crowd behaviors recognition rate is over 80%,the method can effectively express and identify the small crowd behavior.