MRF Model for Detecting Abnormal Activates in Crowded Environments
K. Poomala, J. Jayageetha · Zenodo (CERN European Organization for Nuclear Research) · 2017
This paper focus on detecting unusual activities in video. The analysis of motions and behaviours in crowded scenes constitutes a challenging task for traditional computer vision methods. To overcome this disadvantage there are different methods are used to detect the abnormalities in the video. This proposed method shows that a space-time MRF (Markov Random Field) model for detecting abnormal activities like bicycle passing through a crowd. This method not only localizes abnormal activities in crowded scenes, it can also capture the irregular interactions between local activities in a global sense. Histograms of Oriented Gradients (HOG) are used for capture the image from the particular video. The extraction of appearance characteristics in Region Of Interest (ROI) tracked over time using HOG descriptor. The robustness of this method in practical application can be understood by applying it on long surveillance videos.