Abstract
Information about camera operation such as zooming, focusing, panning, tilting and dollying is useful not only for efficient video coding, but also for detection of moving objects in the scene. We have developed a camera operation parameter measurement system and applied it to inferring the camera-induced image motion directly from pan/tilt/zoom parameters. However, when the camera is undergoing translational movement such as dollying, the camera-induced 2D motion depends both on the camera operation parameters and the depth of each point in the scene. In this paper, we propose a multiframe image matching algorithm for robust depth-map estimation. The algorithm exploits camera operation parameters to establish point correspondence among multiple image frames. The experimental results show significant improvement in the accuracy of depth-map compared with the conventional two-frame matching method.
