The Fluid Describe Of Crowd Movement And Abnormal Behavior Detection

With the improvement of living standard and human safety consciousnessenhancement, the human put forward more and more high demand for social security. Asa new intelligent video surveillance system which is to use computer technology to realizethe automatic analysis and detection video alarm, thus getting more and more people’sattention. Based on the video and the crowd abnormal behavior detection become moreactive in this field, in recent years it gets many scholars’ attention and research, and alsoput forward a variety of solutions, which completed the security task efficiently and bringsa great economic benefits for the society; and the meantime realizing the alarm timely, itsolves the lag phenomenon of traditional monitoring system. This article from thetheoretical and practical application makes new exploration for the crowd scare detection;the proposed algorithm has been applied in practice. The main contribution of this papershow as follows:(1) The motion estimation for the different videos. In this article the Horn-Schunckoptical flow algorithm is adopting for the succession frame, and then the velocity and thedirection can be got.(2) When we got the optical flow results, in this paper puts forward a kind of methodwhich combining fluid mechanics features such as vorticity, divergence, gradient tensorinvariant which are used in the crowd, this method considering the similarity betweencrowd movement and the fluid motion, so we can put the crowd movement framework tothe fluid movement. Through a lot of experiments results we can get that this method canbe applied in different indoor and outdoor scenes effectively.(3) Using the concept of information entropy for each characteristics of fluidmechanics, normalized and quantified these characteristics to different levels, throughstatistical different pixel number of each level the entropy value can be calculated, in theend using the entropy results to judge the crowd abnormal events happened. To differentscenes’experiments results show that this method can judge the crowd abnormal eventseffectively. (4) For different numbers of crowd, this paper proposed combining line integralconvolution (LIC)、OTSU segmentation algorithm、the least square linear fitting methodfor different numbers of crowd abnormal behavior detection. Through the arrangementthis research can be applied for more areas, the performance can be expanded greatly.