Method for matching general stereo planar curves

摘要

A method is described to solve the stereo matching problem of general closed planar curves under possible imperfect segmentation and occlusions, provided that camera parameters are known. Our method decomposes the parameters of an object plane corresponding to stereo curves (i.e. slant, tilt and scale factor), and uses a histogram technique to individually estimate these parameters. The parameter estimation is based on the disparity information of the stereo curves. Point correspondence plays an important role in the method. We solve this problem in a dynamic programming style. The final matching is assessed by applying a distance transformation. The method has been applied successfully to several practical examples.