next up previous contents
Next: Standard Usage Up: Correlation Stereo Tool Previous: Correlation Stereo Tool   Contents

Introduction

The Correlation Stereo Tool allows the correlation based matching of stereo image pairs. It is a direct substitute for the PMF Stereo Tool, based on more extensive development and testing. The algorithm was specifically developed with hardware implementation in mind (Tina memos 1994-002 and 1995-001).

The similarity measure used by the correlator is user selectable as is the correlation algorithm. The available algorithms are all based on finding similarity between small regions or blocks in the left and right images. The image blocks used for correlation can be the original images, or they can be derived from warped blocks of the original images. There are two methods of image warping techniques which can be used: stretch correlation and shear correlation. In addition to the correlation functions, there are various image preprocessing techniques which can be employed prior to correlation, typical image preprocessing for edge based stereo matching might include horizontal gaussian smoothing, and enhancement of non-horizontal edges. The range of preprocessing available covers a very large subset of the techniques investigated in the published literature, allowing a wide range of alternative approaches to be evaluated for specific data sets.

One key feature of this approach is that the results of stereo processing are available to subsequent analysis. This allows the algorithm either to be iterated in order to refine the set of matches found, or to be used in a temporal context, where the results of previous frames are used to constrain the next stereo match. This gives real benefits in the reliability of match results over single frame solutions (Tina memo 2003-009). More information and references to published literature is available in the stereo demonstration directory.


next up previous contents
Next: Standard Usage Up: Correlation Stereo Tool Previous: Correlation Stereo Tool   Contents
root 2017-09-24