A multiscale stereovision algorithm for a mobile robot

We are interested in matching stereoscopic images. In this context we present a multiscale stereovision algorithm based on primitives of type “contour chain points”. We use a pyramidal image as a data structure. The resolution is reduced with a consolidation. We consider a pyramid at its four uppermost levels. The consolidation is done on a 2×2 neighborhood. The matching process is done at each resolution. The research process is done at a low resolution and becomes finer at the upper ones. The primitives we use are groups of connected edge points called contour chains points. These primitives are rich in information and then easier to match. The contour chain depicts the primitive support and allows verification of the disparity smoothness. In order to have an efficient algorithm able to match the large structures of the scene as well as the details, we have developed a multiscale edge detection algorithm based on the multifractal approach providing high quality contour images and keeping the richness and the details of the scene