On the congruence of noisy images to line segment models.

The problem of model matching is a central issue for computer vision. This paper describes an approach to matching images, in the form of range data or edge points derived from an intensity image, to two-dimensional line segment models. The algorithm assumes that the image and model have the same scale (i.e., are the same size), and estimates, by an iterative process, a translation and rotation that moves the image onto the model. This rigid motion is determined by minimizing a modified mean squared distance. The iterative process includes a novel feature which is vital to its performance. Good convergence has been achieved for arbitrary translations and rotations for a variety of images. The matching is robust against noise, and can deal with occlusions and spurious image points.