Knowledge Discovery in Video Sequences

Abstract

Any knowledge discovery process that discerns novel information identifies newlinepotentially usable and logical data patterns. It encompasses multiple stages for completing newlinea specific task by applying a discovery method. Knowledge extraction is concerned with newlineusing effective and flexible algorithms for efficiently storing, retrieving and analyzing large newlinedatasets for interpretation and visualization of results. Video is one of the most viable form newlineof information representation that records events, activities in real time and captures lot of newlineinformation at an instance of time. Due to increase in crime rate, security concerns and newlinemonitoring purposes, surveillance cameras are installed in cities, highways, homes, newlineshopping malls etc. These cameras routinely record hours of videos which can be analyzed newlineto gain previously unknown information. Videos are also used as a primary tool for newlineinvestigation whenever some anomalous incidents are reported. Analyzing these video newlinefeeds is a cumbersome and mundane task as it has to be continuously monitored by newlinepersonnel specially deployed for this repetitive tasks. Use of computer vision techniques newlinefor analysis of such videos has proven to be effective. newlineAny video analysis application that aims to discover hidden patterns in a video or newlineperhaps interpret the video, would fundamentally need to cope with the large data for newlineprocessing. Videos can be analyzed effectively if the amount of data that undergoes newlineprocessing is reduced. This is accomplished by using two methods: reference frame-based newlineand clustering-based key frame extraction. These two implemented methods are able to newlineextract key frames effectively, achieving an average compression ratio of 0.89 and 0.94 newlinerespectively. newlineInterpretation of objects, events, activities in video is primarily based on the newlineaccurate segmentation of moving objects in videos. Moving object detection in dynamic newlineenvironment is a challenging task and getting an accurate solution that performs detections newlinewith sudden illumination changes, non-static background