Using pyramid decomposition and iterative refinement, OF is calculated. Includes a demo and a paper that thoroughly explains the methodology.
Please don't use this code. It is a wrong implementation. The convergence depends on the number of scales, which it shouldn't.
Thanks.Recently,I'm researching this.
Nicely written code. Thanks
the SAD before / SAD after is always at 1.6 ~ 1.9 for my pictures
Has anyone got beeter SAD ratio?
The terms Ix^2 and Iy^2 denote the second derivatives wrt to x and y. However, that matrix is sometimes ill conditioned, so you may want to google 'good features to track' by Shi&Tomosi which discusses that problem.
I am a french student and I tried make your algorythm work with python but I have a problem (I don't know if it is me or the pdf). when I compute the G matrix , this one is always uninvertible (I dont see how it would be possible to invert a Matrix [[Ix²,Ix*Iy],[IX*Iy,Iy²]] because the determinant is always 0) could you help me please?
I added two example images and some more comments, I hope they'll be online soon.
Thanks for sharing this. Could you please also upload the images yos9.tif and yos10.tif referenced in the demo.m file?
Borders are handled better, and refinement is updated.
Example images and a few more comments are added