• 设为首页
  • 点击收藏
  • 手机版
    手机扫一扫访问
    迪恩网络手机版
  • 关注官方公众号
    微信扫一扫关注
    公众号

royshil/SfM-Toy-Library: A toy library for Structure from Motion using OpenCV

原作者: [db:作者] 来自: 网络 收藏 邀请

开源软件名称(OpenSource Name):

royshil/SfM-Toy-Library

开源软件地址(OpenSource Url):

https://github.com/royshil/SfM-Toy-Library

开源编程语言(OpenSource Language):

C++ 96.9%

开源软件介绍(OpenSource Introduction):

Toy Structure From Motion Library using OpenCV

This is a reference implementation of a Structure-from-Motion pipeline in OpenCV, following the work of Snavely et al. [2] and Hartley and Zisserman [1].

Note: This is not a complete and robust SfM pipeline implementation. The purpose of this code is to serve as a tutorial and reference for OpenCV users and a soft intro to SfM in OpenCV. If you are looking for a more complete solution with many more options and parameters to tweak, check out the following:

SfM-Toy-Library is now using OpenCV 3, which introduced many new convenience functions to Structure from Motion (see my blog post for details), making the implementation much cleaner and simpler.

Ceres solver was chosen to do bundle adjustment, for its simple API, straightforward modeling of the problem and long-term support.

Doxygen-style documentation comments appear throughout.

Compile

To compile use CMake: http://www.cmake.org

Prerequisite

How to make

On OSX Using XCode

Get Boost and Ceres using homebrew: brew install boost ceres-solver (you will need to tap homebrew/science for Ceres)

mkdir build
cd build
cmake -G "Xcode" ..
open SfMToyExample.xcodeproj

On Linux (or OSX) via a Makefile

Obtain Boost (with e.g. apt-get install libboost-all-dev) and Ceres (probably need to clone and compile), or on OSX view homebrew as mentioned before.

mkdir build
cd build
cmake -G "Unix Makefiles" ..
make 

On Windows

Use Cmake's GUI to create a MSVC solution, and build it.

Usage

Execute

USAGE ./build/SfMToyUI [options] <input-directory>
  -h [ --help ]                   produce help message
  -d [ --console-debug ] arg (=2) Debug output to console log level (0 = Trace,
                                  4 = Error).
  -v [ --visual-debug ] arg (=3)  Visual debug output to screen log level (0 = 
                                  All, 4 = None).
  -s [ --downscale ] arg (=1)     Downscale factor for input images
  -p [ --input-directory ] arg    Directory to find input images

Datasets

Here's a place with some standard datasets for SfM: http://cvlabwww.epfl.ch/data/multiview/denseMVS.html

Also, you can use the "Crazy Horse" (A national memorial site in South Dakota) dataset, that I pictured myself, included in the repo.

Other

Some relevant blog posts from over the years:

References

  1. Multiple View Geometry in Computer Vision, Hartley, R. I. and Zisserman, A., 2004, Cambridge University Press [http://www.robots.ox.ac.uk/~vgg/hzbook/]
  2. Modeling the World from Internet Photo Collections, N. Snavely, S. M. Seitz, R. Szeliski, IJCV 2007 [http://phototour.cs.washington.edu/ModelingTheWorld_ijcv07.pdf]
  3. Triangulation, R.I. Hartley, P. Strum, 1997, Computer vision and image understanding [http://perception.inrialpes.fr/Publications/1997/HS97/HartleySturm-cviu97.pdf]
  4. Recovering baseline and orientation from essential matrix, B.K.P. Horn, 1990, J. Optical Society of America [http://people.csail.mit.edu/bkph/articles/Essential_Old.pdf]
  5. On benchmarking camera calibration and multi-view stereo for high resolution imagery. Strecha, Christoph, et al. IEEE Computer Vision and Pattern Recognition (CVPR) 2008. [http://infoscience.epfl.ch/record/126393/files/strecha_cvpr_2008.pdf]



鲜花

握手

雷人

路过

鸡蛋
该文章已有0人参与评论

请发表评论

全部评论

专题导读
热门推荐
阅读排行榜

扫描微信二维码

查看手机版网站

随时了解更新最新资讯

139-2527-9053

在线客服(服务时间 9:00~18:00)

在线QQ客服
地址:深圳市南山区西丽大学城创智工业园
电邮:jeky_zhao#qq.com
移动电话:139-2527-9053

Powered by 互联科技 X3.4© 2001-2213 极客世界.|Sitemap