Modelling and Identification Algorithms for Emerging Applications in Data and Signal Processing;Lodz University of Technology Press : 2014

3D Reconstruction from Images

Francho Melendez

CeTA

Point-clouds combined into a single point cloud. Each colour represents a different sequence. We incrementally align new sequences. The last sequence presents some misalignment (highlighted by a red square) with the first sequence due to distortion of the input data. Note that first and last sequences are not aligned directly but by the intermediate point-clouds

Abstract

This paper is an introduction to the problem of 3D reconstruction. We discuss some of the problems and existing techniques for inexpensive creation of 3D models using photographs, with focus on large outdoor scenes. We introduce the specific challenges that this presents, and review current approaches to solve them. We also overview our 3D reconstruction system and present two improvements to our multi-view stereo pipeline, allowing the combination of several partial reconstructions and using uncertainty on the reconstructed data to filter noisy points. Finally, we show some results produced using our semi-automatic pipeline.

Download

Paper

Bibtex

@inbook{Melendez:2015: author = {Melendez, Francho}, title = {3D Reconstruction from Images}, booktitle = {Modelling and Identification Algorithms for Emerging Applications in Data and Signal Processing;Lodz University of Technology Press}, year = {2015}, isbn = {9788372836571}, pages = {153-170}, }