Hi,
 
i'm working on a bunch of shots where a head is to be tracked, using MM pro 4. I prepared the head at shooting time with tracking marks and took reference pictures from different angles than the main camera - i even have focal length and film gate info.
 
Tracking works well so far. The solved points fit onto the footage well - as long as i look through the solved cam.  
 
The big problem is that on most shots the face point cloud is inverted along the z-axis (e.g. the nose tip points away from the cam but it should point towards it). Also in some cases the points are spread way along the z-axis. Even using my nice high-quality reference pics i can't seem to make it work. I also tracked both the room and the head as rigid motion - the room went fine, for the head i got the same wrong result. I played with the coordinate system too - with no effect.
 
Whatever i try - always the same result: appearently nice tracking result, but if you look from the side you see very wrong point positions.
 
I heared of Problems with tracking cams that rotate around an object while they are "constrained" to it. My face tracking seems to be very similar to that issue.
 
Is there any way to flip the points into their right positions? Some workaround? (otherwise i have to 3d-rotoscope all the heads by hand )
 
any ideas would be very appreciated
Marcus

Hi Badoo,
 
I have done a head track as well and ran into the same issues.
 
Since your points are spread across a large area and don't accurately represent the tracked head a likely issue is that MM doesn't have enough info to understand the shape, this DOESN'T mean you don't have enough points.  To fix this I had to record my actress turn her head left to right before begining the scene.  This gave me extra data to track but also gave me very solid track.  An object rotating in an object track is the same as having parallax in a moving camera shot, it is very important.
 
When my points were coming out inverted I found the best way to solve it was to rethink my track points.  In some cases removing points made my solve perfect, do this by unchecking the "used for 3D solve" check box, that way if it doesn't help by shutting it off you can turn it back on and try adjusting other tracks.
 
Also, just a quick note, more points doesn't necesarily mean a better solve.  MM asks for a minimum of 7 points, in some cases going over that actually made my solve less acurate even though they were solid tracks.  Also, if 7 points is very difficult for you to get it doesn't have to be 7 across the entire solve.  This is a dangerous cheat but as long as you have 7 at one point that is better than nothing, your solve may suffer, or it might not.
 
I hope this helps.

Hi PudBawl,
 
thanks for sharing your experiences.
 
Since i provide two reference images from different angles to my main camera, MM should have enough parallax info. I know that I use enough points. I use 8 to 12 points for each tracking with pretty good results (survey > 0.4) exept the the inversion of my points.
 
Little feature request to the realviz team:
It would be helpful to be able to give MM something like a point relation info were you can say "point A is closer to the cam than point B". This could maybe easily solve such problems with inverted points.
 
cheers