Interpreting image curve from multiframes

摘要

A method of reconstructing the structure of a rigid space curve from multiframes is presented. The problem is formulated as: Does there exist a unique reconstruction of a general smooth space curve from images if the curve is moving constantly in the space? What is the minimum information needed to allow such a reconstruction? The motion here is taken to be general, but rotates about a fixed axis uniformly. The only featured points are the two endpoints of the curve. We first establish a necessary and sufficient condition on the number of frames for determining the motion with only two feature points observable. Also, whether the underlying motion meets the formulation of the problem can be checked from the observables. Next, we show that the ambiguities of matching a given nonfeatured point on the curve in one frame are limited to the intersections of a straight line and the image curve on the other frame. The ambiguities of matching nonfeature points can then be resolved and the reconstruction of the space curve follows readily. Experiments and examples are provided to illustrate each step of the method. Furthermore, we find that conclusions obtained by Yuille and Poggio [8] (generalized ordering constraint) in the case of parallel projection are either special cases of or can be easily derived from this method.