The disparity pyramid: An irregular pyramid approach for stereoscopic image analysis.

Abstract

The disparity pyramid is a proposed solution for both problems of disparity estimation and object detection using a stereoscopic pair of images. This pyramid can be classified within the class of irregular pyramids. The hierarchy of the pyramid is achieved by successive calculations of a set of levels. These calculations are done over a disparity range to get the difference in intensities between the two images. A cell can survive at the next level if it has the minimum difference within its neighbourhood. This minimum difference plays the role of an indicator of how near this cell, in terms of intensity, to another cell in the other image. This means that the disparity pyramid takes into account the information of intensity of the images under consideration; in other words, it is a data-dependent structure. A cell is a root, if it cannot be linked to any of its neighbours because it is far enough from all cells in its neighbourhood. Unlike some other pyramids, and according to the input values and/or the distance between disparity vectors and/or number of objects contained in the scene, the top level of this pyramid may consist of more than one cell. Thus, in terms of geometry, its shape looks like an incomplete pyramid with a flat top level and irregular sides.