Video Anomaly Identiifcation By Motion Background Subtraction

The rapid development of computer technology and information technology in the fields ofartificial intelligence, digital image processing and pattern recognition makes people can get thedigital signal of object images in the scenes, and then use the computer to achieve the automaticmonitoring of the video information through information processing.In this paper, a motion background subtraction based method was proposed to detectanomalous events in the traffic videos. The approach makes good use of the spatiotemporalcharacteristic, objects identifying, classifying or tracking is not needed at first in this method, and itcalls for less computation and memory consumption. However, a improved anomalous objectlocalization by forward-backward MHI(motion history image) algorithm was used to enhance thecorrectness of the unusual events detection system.There is no clear definition of anomaly events. The method proposed just works at the pixellevel to build spatiotemporal co-occurence models of normal data and then attempts to detectdeviations from the normal models in observed data. The unusual events detection system is dividedinto three main steps: moving targets detection, abnormal events detection and abnormal objectlocalization. First, in the moving targets detection stage, we use the background subtraction to findout the moving objects. Second, at the pixel level, each pixel builds a spatiotemporal co-occurencemodel of normal data and then attempts to detect deviation from the normal model in observed data.Third, abnormal objects localization needs to be done by the improved forward-backward MHI.In the thesis, within the best parameter values,multiple video images are intercepted in eachstep to show the experimental effects. Confirmed by experiments, our approach has a highefficiency.