Video Material Management System For Virtual Scene Construction

Rapidly growing quantities of digital video has made video data become a more important role in many applications than ever. Virtual scene constructing also use video materials as a rich resource for its strong sense of reality. In order to manage the video materials like salient object and scene, an effective video retrieval system is required.In this paper, we first propose a hierarchical video data model. The model presents a hierarchical structuring of video material and a hierarchical annotation of video material structure. Based on this hierarchical data model, we analyze the video data structuring process and its key techniques. We present a video scene classification method based on Support Machine Vector (SVM) algorithm. The experimental results show that the accuracy rate of scene classification is 97%.For more effective queries on these structured video data materials, we build an indexing tree which supports top-down access, bottom-up access as well as the access among the salient objects. Besides, a semantic indexing structure for more convenient and efficient semantic retrieval is built. We also present the design of the video database.In order to build the virtual scene, salient objects need to be cut out in advance. It also constructs the bottom layer in our data model we present. In this paper, we improve the video cut out procedure by separating the time-consuming step, alpha matting from the user interactive steps as well as using parallel computing to accelerate several steps. Experimental results show that it accelerates by 67% and results in better user experiences.Finally, we present our video material management system and explained the important modules respectively. Our system supports effective video accessing and video editing, which includes semantic query, scene query, hierarchical browsing, video preview and annotation, etc.