Lab
Lab
| This project is concerned with feature point matching between image pairs. When two views of a scene are taken by close cameras with unknown relative positions, the first step, in many computer vision tasks, is to establish feature point correspondence between the views. These correspondences are essential towards calibration (identifying the camera positions) or reconstruction (building a 3D model of the environment). The goal of this project was to establish, and then use a framework for evaluating matching constraints. Image pairs were selected, on which feature points were identified. Then, all possible matches between the selected feature points were established manually. These ground truth sets could then be used to evaluate the accuracy of various matching schemes. Links to data including ground truth sets, and links to the publications resulting from the project are found on this page. |
Links to image pairs, lists of feature points, and ground truth sets of all matches between the feature points.
Detected feature points in the left view |
Detected feature points in the right view |
Detected feature points in the left view |
Detected feature points in the right view |
Detected feature points in the left view |
Detected feature points in the right view |
Detected feature points in the left view |
Detected feature points in the right view |
Detected feature points in the left view |
Detected feature points in the right view |
Detected feature points in the left view |
Detected feature points in the right view |
Étienne Vincent and Robert Laganière,
Matching Feature Points in Stereo Pairs: A Comparative Study of Some Matching Strategies,
in Machine Graphics & Vision, vol. 10, no. 3, pp. 237-259, 2001.
PDF [3.5 mb]