Date of Award

Spring 2005

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering - (M.S.)

Department

Electrical and Computer Engineering

First Advisor

Stanley S. Reisman

Second Advisor

Yun Q. Shi

Third Advisor

Richard A. Foulds

Abstract

One well known problem in stereo vision is the trade-off between precision and accuracy.

In a conventional two camera model, as the baseline separating the two cameras becomes smaller, the images that both cameras produce become more identical. This results in a more accurate set of correspondences, however at the same time causing less precise depth measurement due to the small angle used for triangulation. Similarly, in a larger baseline model, the depth measurement is more precise while the correspondences are not as accurate as of the shorter baseline model. This thesis proposes a method that promises to solve the precision-accuracy trade-off problem. It employs an array of cameras, where each adjacent pair of cameras has a small baseline while the entire baseline of the array is relatively large. In such a system, the accuracy of a short baseline model is enjoyed, and at the same time, the produced results have the precision matched to a longer baseline model.

The proposed method in this thesis finds accurate corresponding points in the first camera pair, and then propagates this set of points along the array, searching for the same correspondences in each adjacent camera. The final result is a reliable set of correspondences between the cameras positioned at both ends of the array. The image coordinates of these points are then used to find the disparity map, followed by a triangulation algorithm for a precise depth calculation.

Share

COinS