 |
 |
 » » oglop | forgive my poor english,i hope i've made myself clear
yes and i've read about the manual and the book Invisible Art of Camera Tracking and Gnomon's Camera Tracking 101 Tutorial, but I still came up with problems when matchmove a shot.
the footage was shot by a SONY MINIDV TRV 60E , and i dunno about its ccd size ,but i figured out through gnomon's instruction.it's got a 1/3.6" CCD,i found the right film back setting on gnomon's lecture notes,they said i should use the actual sensor size as the film back parameters, yap i know the 2/3 rule and the aspect ratio of the ccd is 4:3 and 1/3.6" means a Diagonal of 7.056mm , and then i multiplied 2/3 to it , I then got the "real" diagonal of the CCD sensor ,and that is 5.000mm and through pythagorean theorem ,'cuz i know the ratio is 4:3 so i got the final film
back parameters and that's 4.000mm ,and 3.000mm,this combination is from gnomon's notes ,i tried it but didn't get a good result,and i tried the values calculated by the 2/3 rule by myself still no luck.
the footage is shot in PAL 25f 720*576
my symptom is that yes i can define a coordinate system and my features r sticky but when time comes to the coordinate part,things got really weird.the coordinate system was sucessfully defined ,but one of the xz axis can not be aligned with the marker i put on the desk, i'm pretty sure the angle is 90 degree .but the oz axis just can't align with it ,after taking in to some inspection of the 3d scene , i found that the associated 3d point r just not perpendicular so the coordinate system cannot be applied to the solution.maybe i didn't make myself understood ,but i'll sent the project file and the image sequences on rapidshare
i searched associated thread on this forum ,some guys mentioned nodal pan shot ,but i think my shot is a handheld shot hence it should have some parallax in it or there is just not enough parallax?
already tried adding a nodal pan constraint but the discrepancy between the actual axis and the solved 3d helper r just more obvious,so is there anything wrong in my implementation of the matchmove?
and i found the following response of u guys in this thread
Quote :
Hello,
In fact, your shot is what we call a "nodal pan". That means the camera rotates almost around its optical center.
This kind of camera move does not give any perspective information (it's not a matchmover issue, it's a physical limitation). So you'll only be able to compute a camera rotation with this data. The 3D tracks will be useless.
In this case, you can add the nodal pan constraint to the camera to force it to only rotate to have a more accurate result.
Therefore, you won't be able to define a coordinate system based on the tracks. You should export this camera move directly in your 3D package and work with it.
What you can do to add perspective information is either:
- add some images from another point of view (one can be enough)
- input the 3d coordinates of some of the point
Hope this helps.
|
what do u mean by "You should export this camera move directly in your 3D package and work with it."
do u mean i should add the translation by myself (we r using maya here) ,since the rotations of the camera have been solved?
looking forward replies...
btw ,my best result was file diehard_quite_close_test.mmf
here is the project file
http://rapidshare.de/files/1620449 [...] 0.rar.html
A o ,sorry here comes the right sequence
http://rapidshare.de/files/16271721/match.rar.html |
niko | hi,
I'm afraid you didn't uploaded the good sequence...
Could you check and upload the correct one?
Thanks,
Niko |
oglop | Quote :
hi,
I'm afraid you didn't uploaded the good sequence...
Could you check and upload the correct one?
Thanks,
Niko
|
the new sequence has been uploaded... |
oglop | the right sequence has been uploaded... |
niko | hi,
Please find below 2 MM projects:
- the first one "diehard_quite_close_niko3.mmf" is one i made from scratch with manual tracks and two relations (share Y) and it seems to me that it solves your problem: the tracks you wanted to use to define your coordinate system are nearly perpendicular.
- the second one "diehard_quite_close_niko4.mmf" is your project "diehard_quite_close_test.mmf" a little bit cleaned: i removed tracks with bad residual and shortest ones, i calibrated the camera and it appears that the result is more consistent with what you wanted to obtain.
Let me know if it helps you,
regards,
Niko |
oglop | It's so kind of u,u r my no. 1 big life saver,
I have no access to the software at the moment
I'll check this in the afternoon
thanks again |
oglop | again thank u 4 ur time for invesgating and analysing my project files
question 1:
do i have to make them align?
i mean the coordinate axis and the line between the two markers , do i have to make them align exactly as in real world in order to get high-end (not slipping) result?
question 2:
does the coordinate scale really matters on 3d solving process? i learned from the manual that u should solve the camera first , then define coordinate system 'cuz these r 2 definitely different process, the 3d point relationship is already there after the 3d solving process.
but the confusing thingy is that when i changed my distance in my coordi 01,the MM system says "some point relations got obsolete" and it just tries to disable them.i really don't get it.as far as i have learned about the solving process,the point relations r just some input for the solver in order to facilitating the solving process,how could them become obsolete when i am defining the coordinate scale? |
 » »
|
 |
 |
|
