Perspective and Matchmoving in PFTrack

[MarkHawk]

Hello Everyone,

I've been working on a project for graduation and I've run into a snag. I've got a low angle shot looking up that is dollying around a building that is on fire. My thought process for this was to track the shot in PFtrack and export all the data into Maya and start building the 3d geometry. From there export the particles interacting with the geometry and bring it over to Adobe After Effects to add some 2d plates of actual fire. Simple...

Everything seems to track great. Everything sticks and none of test objects do anything crazy. On screen it looks great. So I try to add a model cube that way I can export the cube into Maya as a reference. When I add the cube it's either to far in the background or to close to the camera and only stays perfect for one frame. Again all the trackers stay locked on just fine but the cube does not match up at all. I purchased the cmiVFX tutorials and have gone through it a few times but I can not get the cubes to match up. Anyone know if setting the origin would affect the cube placement?



So I decided to skip the reference cube. As long as I match a cube from Maya with my trackers and have my grid perfect in PFTrack it should be fine right? This is how it looks when I match up that cube in Maya. The perspective is off... is mine off and if it is can someone give me so good advice on how to match perspective better?



Doesn't work... Here is the video footage I am trying to track if this is any help. If I've been to vague on details please let me know and I'll be sure to get what ever you need to help. Thanks

6mbs Quicktime H264: http://marksvfx.com/videos/Shot%202C.mov

[stevenbray1]

From your screenshots it looks like none of the points that PFTrack s generating are in the right place in 3D. If the track is accurate they should be floating in space, in the shape of the building, in the 3D view.

The one sin your screen shots look like PFTrack has inverted the shape of the building.

Just because the software thinks it's done a good job doesn't mean it's accurate.

Have you compared the camera that PFTrack calculates to the actual camera? Focal length etc.?

EDIT:
Maybe not inverted so much as distorted? It's hard to see from here. Comparing the focal length is the first thing I'd look at anyway.

Then make sure you put your cube next to the correct point? In the first screenshot it's not actually alligned to the tracking point from that part of the building?

Good luck.

[Craigh]

Play around with adjusting your scene orientation, then go into the persp viewport to double check that it's lining up.

[PhilS]

just looking at your setup...you seemed to have included the reflection of the trees on the rhs and lhs. This will cause problems. Track a user feature on the window on the rhs and rear lhs and connect masks to these areas and connect the masks to the user features, ie track the masks along the windows where there are reflections. Make sure the reflections are not included in the equation Haver a look at your point cloud, it should consist of the majority of points on two planes

[MarkHawk]

From your screenshots it looks like none of the points that PFTrack s generating are in the right place in 3D. If the track is accurate they should be floating in space, in the shape of the building, in the 3D view.

The one sin your screen shots look like PFTrack has inverted the shape of the building.

Just because the software thinks it's done a good job doesn't mean it's accurate.

Have you compared the camera that PFTrack calculates to the actual camera? Focal length etc.?

EDIT:
Maybe not inverted so much as distorted? It's hard to see from here. Comparing the focal length is the first thing I'd look at anyway.

Then make sure you put your cube next to the correct point? In the first screenshot it's not actually alligned to the tracking point from that part of the building?

Good luck.

I know the screen shot does not do it justice.

It's not inverted but your correct as they do not line up at all. The zig zag instead of being in a straight line. I haven't checked the focal lengths yet either.

The bottom screen shot from Maya has the cube matched up with the correct point. Perspective is still off which makes me believe the focal Length is incorrect.

[MarkHawk]

just looking at your setup...you seemed to have included the reflection of the trees on the rhs and lhs. This will cause problems. Track a user feature on the window on the rhs and rear lhs and connect masks to these areas and connect the masks to the user features, ie track the masks along the windows where there are reflections. Make sure the reflections are not included in the equation Haver a look at your point cloud, it should consist of the majority of points on two planes

I originally masked them out but the camera resolve left me with about 6-7 tracking points and no camera movement. It acted as if I had the camera on a tripod and was just rotating left to right. I had the camera set to free motion as well.

I then tried tracking without the mask and going in and manually deleting the points on trees and reflections in the windows. When I resolved the camera I would be left with again 6-7 tracking points.

What I haven't tried was deleting the reflections only. There was little to no wind this day so the trees maybe worth keeping but I didn't want to take the chance at the time. I'll try that as well and checking the focal.

[PhilS]

i did an auto track with about 250 points, then deleted all the ones associated with the tree on the front rhs and rear lhs. then I selected the group of trackers on the lhs of the building and tracking/constraint/ added them to a planar constraint. I did the same for the tracks that are on the rhs. Then I did the camera solve.

I forgot to mention, you should undistort the image first as there is a slight amount of distortion

[MarkHawk]

i did an auto track with about 250 points, then deleted all the ones associated with the tree on the front rhs and rear lhs. then I selected the group of trackers on the lhs of the building and tracking/constraint/ added them to a planar constraint. I did the same for the tracks that are on the rhs. Then I did the camera solve.

I forgot to mention, you should undistort the image first as there is a slight amount of distortion

First, any recommendations for undisorting the image? I captured in Adobe Premiere and de-interlaced it there and exported Targas. I notices slight distortion myself but not knowing how to tackle it I left it be.

Also what does rhs lhs stand for? Sorry just have no idea what they could mean.

[MarkHawk]

Comparing the focal length is the first thing I'd look at anyway.

Focal Length in PFTrack 22.1
Focal Length for the camera was f=6.1-61mm

[noouch]

Focal Length for the camera was f=6.1-61mm

Be sure that you're using the 35mm focal length equivalent. Figure out the focal length multiplier for the camera that was used, or at least try pftrack's focal length estimation tool (it seems to be perfect for your shot).