This paper aims at a new challenge caused by a specific type of real-life problems that require to not only process tremendous videos, dig out cross-video information, but also guarantee a real-time responsiveness to the user. Moreover, users want to adapt the resources to the actual amount of visual objects, rather than to the number of videos, in order to better match resource consumption to the true business needs. All the state-of-the-art solutions cannot meet these requirements altogether, while this paper developed a series of techniques to address each specific problem, and then proposed a new adaptive solution for large-scale, cross-video, and real-time visual analytics.
展开▼
