Abstract
Global motion estimation is one of important research issues in a computer vision field and opens a way to advanced video processing such as video codec, robot vision and object detection. Most of the methods to achieve motion parameters rely on statistical image-based matching. Recently there arise the sensor fusion based methods in which camera measurement technology helps estimating motion parameters. In this paper, we propose a new motion estimation method that takes explicit advantage of salient geometric image feature correspondence. The image features are lines existing in the image and are detected by Hough transform. In the proposed method, computation is carried out mainly on transformed hough space. It drastically decreases difficulties in performing line correspondence. The method is validated through simulation experiments.
