Semantic browsing of video surveillance databases through Online Generic Indexing

This paper gives a thorough overview of EADS UrbanVIEW indexing and mining platform aimed at providing police forces and security officers with advanced tools to efficiently browse large video surveillance databases for investigation purposes. A scalable indexing architecture that works indifferently with smart or classical camera networks as well as for real-time or a posteriori indexing has been designed and implemented. We introduce the concept of online generic indexing strategy (OGIS) aimed at systematically enriching each video stream with real-time extracted generic metadata allowing to dramatically decrease post-event investigation time. The indexing strategy relies on the systematic detection, tracking and characterization of all observed moving objects. Semantic and non semantic metadata produced by embedded or distributed video analytics modules can be used either to browse the distributed video databases or as inputs to higher level characterization modules (object identification, multi-camera back-tracking, event recognition...). Once a first observation of an object of interest has been found, it can be forward and backward tracked thanks to an interactive multi-stream player taking into account the multi-camera context. Our platform has been assessed on the NGSIM and I-LIDS datasets which consist of real heavy traffic images, showing both high recall and high detection rates in its retrieval capabilities.