道路画像におけるパン・チルト・ズームカメラの状態の推定

概要

論文の詳細を見る
In this paper, we consider the problem of estimating the state of camera from the real-time traffic images obtained with the fixed camera with pan, tilt and zoom functions. This problem comes from the research of image processing of the road image for traffic state estimation. Since this camera can be controlled from the authorities of political surveillance, the camera is subject to move suddenly. The state of the camera means the camera direction and the amount of zooming. This processing needs real-time processing from the nature of the problem. If we apply the simple 2-D matching method, it requires a large processing time. Hence, in this paper, we will use two 1-D vectors for matching using two low-dimensional vectors of the vertical and holizontal average values of the pixels of certain area in the image. The feature spaces are defined from the primal eigenvectors of sample images, and the distance between the current image and the feature space is used for matching. Moreover, when we estimate the zoom magnification, eigenimage is derived from the non-zoomed images after positional matching. Similar methods are applied both for the calibration of camera direction (pan and tilt) and zooming.
甲南大学の論文
2005-07-31