MavVStream: Extending Database Capabilities for Situation Monitoring Using Extracted Video Contents

Query-based video situation detection (as opposed to manual or customized algorithms) is critical for diverse applications such as traffic monitoring, surveillance1 , and other types of environmental/infrastructure monitoring. Video contents are complex in terms of disparate object types and background information. Therefore, in addition to extracting complex contents using the latest vision technologies (including deep learning-based), their representation as well as querying pose different kinds of challenges. Once we have a representation to accommodate extracted contents, ad-hoc querying on that will need new operators, along with their semantics and algorithms for their efficient computation. Extending database framework (representation and real-time querying) for processing queries on video contents extracted only once is critical and this effort is an initial step in that direction. In this paper, we extend the traditional relation to R++ (vector attributes) and arrables to accommodate video contents and extend CQL (Continuous Query Language) with a few new operators to query situations on the extended representation. Backward compatibility, ease-of-use, new operators (including spatial and temporal), and algorithms for efficient execution are discussed in this paper. Classes of queries are identified based on their complexity to evaluate with respect to video content. A large number of small and large video datasets have been used (some from the literature) to show how our work can be used on available datasets. Correctness of queries with manual ground truth, efficient evaluation as well as robustness of algorithms is demonstrated. Our main contribution is couching a framework for a problem that is becoming very important as part of big data analytics based on a novel idea.