Detection Of Crowd Gathering Events And Behavior Based On Corner Feature

Recently, more and more terrorist events and violent incidents happened in publicplaces. Therefore, it becomes a vital problem to protect the safety of the people inpublic places. The government had taken a lot of security measures to reduce theincident, including to arrange staff patrol in public places and install CCTV. However,those measures need lots of human and material resources. As the development oftechnology of intelligent video surveillance, people can monitor the real scenario byusing the computer vision system. It saves lots of manpower and materials. It takesmore and more attention to deal with the crowd event by government departments at alllevels. The premonition of the crowd event is crowd gathering, so it becomes ourresearch topic to detect the crowd formation and evacuation.The description of the crowd is first introduced in this thesis. Many factors affectthe segmentation of the crowd, including plenty of individuals in different shapes,occlusion, illumination, etc. Therefore, we deal with the crowd as a whole. We use thecorners as the base of the crowd event detection to represent the crowd. The followingcrowd formation and evacuation detection is based on the corners.Two crowd formation and evacuation detection methods are proposed in this thesis.The first is the convex based crowd gathering detection method. Corner which has thesimilar motion is clustered as a whole by using cluster algorithm and represent eachcluster as convex. After that, we detect the formation and evacuation event according tothe mutual movement tendency of each cluster. The other is a weighted corner areadetection method. The distribution of the crowd is described by the corner area.However, the corner area is easy to be affected by unrelated individuals. Hence, weintroduce the weighted mechanism to reduce the influence of unrelated individuals.According to the experiments and results, the two methods mentioned above havetheir own advantages. The corner area method is easy to implement and has a fastdetection speed. While the convex based method can detect crowd gathering event moreprecisely and has strong anti-interference ability.