www.ssontech.com

[PaulWinterton]

I'm working on a scene that has a 175 degree pan and a slight dolly move. 6 second move inside a factory. There is motion blur in the middle of the shot.

Although I'm getting a fair number of accurate trackers and I can setup a coordinate system, my ground plane changes SIGNIFICANTLY from the start of the scene to the end of the scene. I also think my camera is "jittery". I haven't exported to 3d software, but it looks like it has more erratic motion than it should.

I tried creating supervised trackers throughout the scene, mostly on the ground, but I got the same result.

Any opinions? If a manual method works, where can I find information on fixing it by hand?

Thanks for reading.

Paul Winterton

[GraphicsKid]

In what way is the ground plane changing?

Also, have you tried helping syntheyes detect the dolly move by telling it what the camera does? In the solver panel, the second drop-down box gives you 7 options: Automatic, Left, Right, Upwards, Downwards, Push In, Pull Back. Try taking it off of "automatic" to help it out a little.

Now when you say a "slight" dolly move... you're not trying to say it's negligible enough to where you've set it to tripod solve now do you?

Motion blur is the most likely culprit here. How bad is the blur? Make sure the trackers are at the FRONT of the blur, not the MIDDLE.

Have you tried looking for bad trackers without the image? It works surprisingly well. Just right click in the camera view and uncheck "show image". Now play the sequence a few times. Any really bad trackers will stick out like a sore thumb. You may also want to look at the graph editor for spikes in tracker velocity. I assume it's a fairly smooth motion as it's on a dolly right?

You say you've tried creating supervised trackers, so that leads me to believe you've tried doing an all-automatic solve... Well when all else fails, I start from scratch and go straight to supervised tracks, because that's the only way to ensure it's not the trackers fault.

[PaulWinterton]

The ground plane is tilting about 30 degrees. Each time I've solved the scene I've tried creating a coordinate system at the start of the footage, and at the end. I can't do it in the middle because of the lack of floor features and the motion blur. Let me clarify that the motion blur is moderate, not extreme. I can see features. If I create a coordinate system at the start, the ending plane is tilted. If I create at the end of the footage the start is tilted. Hmmm, looking at the camera icon I see that it appears to be "rolled". Maybe it's not the coordinate system, but the solved camera position/roll instead.

The dolly move is about 3 feet, with a pan of about 175 degrees. I have tried tripod solve, but only as a last resort. Funny, I have best (bad) results from this, but gave up because lack of distance information causes my objects to slide. Tweaking makes me want to jump off a bridge.

I have kept a version of the SNI file where I created 26 supervised trackers, unsolved. Most of the trackers are on the floor, some on vertical highlights at different distances from the camera, but not toooo far away. I have tweaked each tracker to smooth-out the FOM. Velocity etc. all seem smooth.

Shouldn't SE trust that I followed each tracker from start to finish, and create a solution based on THAT without looking at the footage?

When I solve, I see a recurring message line that reads (something like) " Cannot get crisp solution, using different algorithm"

I will try the different options that you recommend, rather than automatic.

I will look for bad trackers without the image.

Thanks for your insight.

[GraphicsKid]

Not a whole lot else I can say without actually seeing the footage, or at least a few screencaps. One thing did stick out to me though... you said you had a few trackers on highlights??? Please tell me you don't mean reflected highlights. NEVER NEVER EVER put a tracker on something that changes position with the camera, like reflections off a non-planar surface.