Skimming Multiple Perspective Video Using Tempo-Spatial Importance Measures

Multiple perspective video taken by more than several tens of cameras are observed and stored for applications such as video surveillance and outside broadcasting. It is one of the most important demands to grasp what happens in the area from the multiple perspective video in a short time, but there are some problems to be solved. For example, we cannot look at a lot of video simultaneously and it is difficult to understand the entire state of a phenomenon that occurs in a broad area and is sparsely taken by a plural cameras. In this paper, we propose a new skimming method using tempo-spatial importance measures. At first, the video importance is calculated using the importance of elements captured in the video scene. The elements are objects such as buildings and cars, and events such as a temperature rising over and a batter hitting. They have their own importance based on space, time and semantics. Then some video are selected based on their importance and displayed with a map and three-dimensional graphics of the objects for skimming the multiple perspective video effectively. We discuss the tempo-spatial importance of multiple perspective video, the video importance calculation and the display methods based on the importance. We also describe our prototype and some experimental results.