We propose to develop automated video indexing and retrieval mechanisms based on comprehensive representations of video data, and efficient processing for automated in-dexing. Content-based video data retrieval has enormous potential in defense, industry and the government. However, a number of stubborn problems have impeded progress in developing fully automated video database systems. These problems include difficulty of specification of video queries and huge computational requirements. Our effort will focus on developing viable solutions for these two problems and utilizing the solutions in building an automated video database indexing and management system. We plan to employ wavelet-based multiresolution pattern matching techniques for recognition of video objects. The precision of object modeling and recognition can be tuned to the level of available computational resources and time constraints thus allowing trade-offs between levels of computation and accuracy of pattern matching. We will develop pragmatic knowledge-based models of video data to capture complex video semantics. These models will provide efficient Indexing and retrieval mechanisms for a variety of scenarios. Two models, based on object motion trajectories and spatio-temporal logic will be developed for this purpose. The system will provide a graphic user-interface for the users to formulated complex video queries. Anticipated Benefits/
Potential Commercial Applications:The proposed effort will develop tools for automated indexing and retrieval of video data. This technology is anticipated to have wide-spread applications including those in surveillance, intelligence gathering, digital libraries, tele-medicine, news and entertainment, and collaborative manufacturing. Video Processing Spatio-Temporal Modeling Automatic Object Recognition Trajectory Based Modeling Content-Based Indexing Fuzzy Queries Scenarios User Interface
