Correlation based Abnormal Motion Detection in Crowded Environment
Gyu-Seong Kim, Dae-Yong Park, Hyeran Byun · Jeongbo gwahaghoe nonmunji. so'peuteuweeo mich eung'yong · 2011
Intelligent surveillance system is proposed for automatically detecting abnormal events by analyzing surveillance video frames in crowded environment. Especially, we focus on detecting abnormally moving motions. To detect abnormal events in crowded environment, low-level features of moving objects are extracted from video frames in optical-flow form. These features are used to get motion correlation model which is used in abnormal event detection phase. There are two steps in abnormal event detection phase. The first detection step is based on local model which is made with local information such as velocities, directions and positions and the second step is based on gobal model which is made with global information. The proposed algorithm shows practicality and sufficient accuracy through many experiments which use many public benchmark video sequences and compare with LDA based methods.