| With the development of mobile Internet,networking and other technology,an increasing number of video applications are widely presented in every area of social life.There are vast amounts of videos produced in surveillance cameras,smart phones,drones,Google Glass and many other sensors.It is a challenge to develop the search techniques which can help individuals to obtain their interested video in fast and effective way.Some techniques have been proposed to meet this challenge of videos'organization and retrieval of unstructured to structured video,such as the text annotation assisted retrieval and the content based video retrieval.However,because of the large amount and the complex content of the videos,these two methods are both limited either in manual level or technique level.Text annotation is simple,but needs a lot of work be added by hand and hence this manual task is laborious and cumbersome for large video collections.Otherwise,it is also very sensitive to the difference of subjective consciousness.Content based video retrieval is limited by the development of image and video processing.There is a "semantic gap" between the video analysis results and the human's understanding.Hence,a metadata which contains precise semantics and can be acquired easily is needed to help individuals obtain what they need in a fast and effective way.Spatial-temporal data is an ideal metadata of video.Most videos(expect some manual videos)are spatial-temporal related.Besides,spatial-temporal conditions also appear frequently in video retrieval usage.It's the necessity for spatial-temporal data to be regarded as the ideal metadata of video.Varity of location sensors(GPS,WiFi location sensor etc.),position sensors(electronic compass,gyro etc.)as well as electronic clock are integrated into video sensors.It's the possibility for spatial-temporal data to be regarded as the ideal metadata of video.These two conditions make the spatial-temporal based video search becoming the research hotpot.The paper are located in the study of spatial-temporal video retrieval method and committed to answer "what is better temporal video search result set,how to build a better spatial-temporal retrieval of video collections".The attribute,evaluation,search algorithm and application are studied.Content and achievements are as follows:1)Panoramic perception.The theories of video sensor coverage and network optimization are associated by the paper to propose a new evaluation mechanism called panoramic perception.Panoramic perception is an index which is comprehensively respect the condition of perceived target,the perceived role and the perceived source.Both panoramic perception(including the concept,attribute and algorithm)and its dual problem best video set are studied by the paper.2)Panoramic perception based video spatial-temporal retrieval.The research results of spatial-temporal model and index are introduced in video retrieval problem.A Historical R-tree(HR-tree)based video spatial-temporal retrieval method are proposed in order to get better retrieval results.3)Panoramic perception based video spatial-temporal retrieval application.A video spatial-temporal retrieval and search engine are developed based on the panoramic perception.The search engine serves as the basic module of the applications.Two application systems are also developed in order to meet the demand of specific areas(one for the field of security,the other for the field of social network). |
