The Application Of Ontology In Surveillance Video Event Detection

Video detection of semantic events is an important part and is difficulty in video semanticanalysis，but the researchs on this subject in the past several years mostly use pattern recognitionto extract low level features such as color and material to detect low level semantics, but littleattention has been paid to the detection of simple or composite events through inference.Nowadays, video event detection has been applied in more and more regions, and plays a moreimportant role in daily life, and video event detection has become a hot research issue in relatedindustries. To detect high-level and complex events, this paper firstly constructs a new framework,inference using the combination of ontology and Petri Net to detect composite events composedby simple events; After annotation with algorithms of video analyze ontology, we propose to buildan video event analyze ontology in the higher level, map the video analyze ontology to eventanalyze ontology to describe the video of the event; then use combined ontology and extendedPetri Net technology to inference graphical and asynchronous over surveillance video events, anduse Semantic Web Rule Language（SWRL） rules to describe the detection of surveillance events;at last, we verified that our method is more effective than pattern recognition method throughinference using concrete high level events.To waring the underlized threats and dangerous things, this paper proposes to alert before theincidents or analyze during the events instead of the common method of analyze incidents afterthings has happened, and change traditional passive method of recognize threats in surveillance toinitiative recognize the threats; decrease the error alert and absence of the necessary alerts byhuman, and liberate the operators from heavy manual surveillance works. This paper propose toanalyze the relations of surveillance video events on the base of surveillance knowledge base, thenmatch with the surveillance knowledge base so as to detect the target knowledge, also meanstarget events. Then alert initially, provide more efficiency and is more automatic.