Building Indices and Analyzing Semantics in the Image Database

点击翻译按钮获取中文摘要

This research aims to build a content-based image database which can store high level information such as "what object does exist in the frame sequence?" or "what do they do?". In the research, from two different stream of frames," stereo pairs" of several point on both the right-eye and the left-eye images are analyzed, and 2-dimentional movement of some specific objects (such as the cap of a worker, the left eye, red glove etc.) are detected on the frames. Also we could calculate the depth of the points by comparing the positions in the both right and left frames. Also we introduced the "model" object in order to find a specific object (such as a human head) in a scene, and developed a matching algorithm between the model outlines and the external appearances of regions in frames. As an experience, we brought several video camera into a soccer stadium, got movies from different orientations and analyzed the posture and the movement of each player. Then we developed the algorithm to infer the "intention" of the collaborative works in actual soccer games. All the detected information and the original data are stored into databases in a uniform way to show the possibility of spatio-temporal retrieval of the above semantically inquiry.