120 likes | 256 Vues
This document outlines the process of creating calibrated cylindrical panoramas, focusing on methods for stitching images together using techniques like RANSAC for outlier rejection and SIFT feature matching. It discusses the mapping of two views related by rotation, cylindrical projection, and converting 3D points onto a cylinder. Additionally, it addresses common issues in image stitching such as exposure differences, vignetting, and misalignments, while introducing multi-band blending as a solution for preserving detail and hiding seams.
E N D
Cylindrical panoramas Jana Kosecka, CS 223b
Rotation Only - Calibrated Case • Calibrated Two views related by rotation only • Mapping to a reference view – rotation can be estimated • Mapping to a cylindrical surface Jana Kosecka, CS 223b
Matching SIFT Features [Brown 2003] Jana Kosecka, CS 223b
Reject Outliers using RANSAC [Brown 2003] Jana Kosecka, CS 223b
Stitching Images [Brown 2003] Jana Kosecka, CS 223b
Do we have to project on to a plane ? Camera Center Jana Kosecka, CS 223b
Cylindrical Projection 360˚ Panorama [Szeliski & Shum 97] Camera Center Jana Kosecka, CS 223b
Cylindrical projection Y • Convert to cylindrical coordinates X Z • Convert to cylindrical image coordinates unwrapped cylinder cylindrical image • Map 3D point (X,Y,Z) onto cylinder unit cylinder Jana Kosecka, CS 223b
Cylindrical Projection Y X Jana Kosecka, CS 223b
Inverse Cylindrical projection (X,Y,Z) (sinq,h,cosq) Y Z X Jana Kosecka, CS 223b
Removing Seams • Differences in exposure • Vignetting • Small misalignments [Brown 2003] Jana Kosecka, CS 223b
Multi-band Blending • [Burt and Adelson 1983] • Multi-resolution technique using image pyramid • Hides seams but preserves sharp detail [Brown 2003] Jana Kosecka, CS 223b