Monday, January 16, 2006

Trifocal tensor estimation Matlab code

I came back to the office after coding at home last week. Tested out the trifocal tensor estimation code and it only threw me one error and works all well after I fixed that one up!! It's a very good surprise as those 18x27 matrices and 3x3x3 tensor are quite tedious to calculate. And I've just understand how to read tensor notation 2 weeks ago, so I was very happy that all the code worked just like that.

Again, carefully laying out the theory and equations before jumping into coding proved to be a good practise. The 25-line notebook is a bit too small for the crazy matrix expansion, filling up the whole page... It kind of looks something like phd work when I flip around those pages of matrices. Hahaha!!

After the code runs, I still wasn't convinced that it is correct, so I cross check it with the code I from the Hartley & Zisserman's book from Oxford VGG website, and it's all good! Yahoo!!
I wrote this code because I searched the internet and couldn't find trifocal tensor estimation from noisy >7 corresponding points in three images. I'll publish this code on my website soon, so vision researchers can use it in the future. And I might get famous, hehehe!! :P
See my matlab page for its future appearance.

No comments: