Real time pose - tutorial
Hi. I have been trying to adapt the real time pose estimation tutorial for my own data, but i cant get the registration to work right.
I have try change SolvePnP modes and parametrs, difrent mode gives difrent results but non is so close as ITERATIVE shown at the picture here. Change of parametrs make same results or none.
Is there someone who would be able to help me make it more accurate?
All i did is set my own camera matrix, load my own(undistorted) image and own box in .ply format.
Edit: here is tutorial i am working with: http://docs.opencv.org/trunk/doc/tuto...
Some points almost fit, maybe is problem with 3Dreprojection, but i also didnt found something fishy there...
I am going mad, how i am trying to found the problem. Thanks for any hints. Gabriel
Did you make the model registration step? You only get 8 points, which seems very low.
The tutorial is composed from two programs. One for registration and one for detection of pre-registrated object. My problem is in registration. Detection work fine. Results ar inaccurate but is becuse of wrong registration.
The program is based on manualy inserted 8 points (vertex of box) which are paired with corner of box from .ply file. (as shown at my picture) here it waits for user to press a key. After that feature points are detected, and checked if they lay at surface of the box. If they do, they are saved for detection program. If they dont they are tossed away. Detection and next steps of registration program are working alright. Problem is that Box is not placed,as result the 8 to 8 matches, in picture as it should...
in the picture. Red dots are users imputed points(next to them are coresponding x,y,z value based on .ply) Green dots are backprojected 3D points from .ply based on computed pose...