Skip to content

Latest commit

 

History

History
86 lines (52 loc) · 2.72 KB

README.md

File metadata and controls

86 lines (52 loc) · 2.72 KB

panorama with MatLab

GitHub - yihui-he/panorama: Image alignment and stitching with MATLAB

https://raw.githubusercontent.com/yihui-he/panorama/master/results/intersection.jpg

https://raw.githubusercontent.com/yihui-he/panorama/master/results/GrandCanyon2.jpg

https://raw.githubusercontent.com/yihui-he/panorama/master/results/redrock.jpg

features

  • able to handle 360 panorama.
  • Random sequence of images input is welcomed.
  • use color blending and smoothing to make the image more continuous.

how to run

Prerequisite: matlab 2014b or higher images sets are already in ./imgs

  • If you want to see results directly, go to ./results folder
  • If you want to test all images sets with only one click,run RunAllDatasets.m.(10 image sets, about 1 minute)
  • If you want to specify the image folder, run main.m with path to images folder as argument as follow:main('./imgs/redrock'); Note that, this currently support image sets in imgs folder. If you use your own image set, you need to set focus length and other parameters in main.m.

360 panorama

  • mapping image to cylindrical coordinatewarp.m

recognize panorama(random inputs)

I select two random sequence images set:family_house, and west_campus1 They are already shuffled. You can see them in imgs folder. Or you can run shuffle.bash to shuffle them again. As described in Brown’s paper, I use $N_inlier>k*N_pairs+b$ to compute whether a pair of images match or not k,b are const. Set to 5.9 and 0.22 respectively. See recognizing panorama for details

imorder.m

merging and blending

  • Alpha
  • Pyramid
  • Noblend

merge.m

transformation

  • homography transformation.
  • translation transformation.( This is more robust)

computeTrans.m

matching

  • RANSAC
  • exposure matching

RANSAC.m

global adjustment

  • end to end adjustment(comput shift and subtract shift/n to each image)
  • bundle adjustment(difficult way)

create.m

getting features

  • use SIFT features(using VLFeat library, professor allowed)
  • SURF features, (SIFT is better)

getSIFTFeatures.m, getMatches.m

resize

  • I resize image larger than 400 pixel in width

main.m

References

A nice tutorial on panorama I find useful.