www.ssontech.com

[lightfuzz]

Greetings,

I have been reading and experimenting with Syntheyes for the last couple weeks now and am still having trouble properly matching my Syntheyes solves with my 3DSMAX model.

Maybe someone could point out what I'm doing wrong here.

Here's what I've been doing so far:

I open my 7000 frame helicopter footage (720x480) and Image Preprocess it so I'm using only the Red channel, with a high pass of 40.

I do the AutoTrack.

I clean up the Trackers. Get rid of moving car trackers, far trackers, low quality (high hpix) trackers, and add some supervised trackers. I take a look at the graphs, pinpoint troublespots.

I do a refine solve.

I don't setup a coordinate system because I don't no any accurate distances in the footage. So, instead I use three supervised trackers and lock them to specific X,Y,Z coordinates. My specific x,y,z coords are just educated guesses from my model in MAX.

It's here that I've tried many variations:
I solve with 'Constraints' off, on, in Peg Mode, etc. Different constraint combos like "Any Y", "on XY plane", etc.

In short, the 3 constrained points I use are spread out in time throughout the footage. I can't see all contrained points in one view. One's in the beginning , middle, and end of the footage.

The issue is that while the points do end up at their specific 3D coordinates, the model is not sitting right in space. Either in MAX or Syntheyes.

Do I need more constraints in this situation? I didn't think that setting up the standard 3 point coordinate system would apply to such a long shot.

I've tried scouring the forums and the web tutorials (invaluable, btw). I've done successful tests on small scenes but when I try a huge scene it doesn't come out right. Looks like scaling issues are at work.

So, for a big project like this, what's the best approach? Anybody done a highway visualisation with MAX and Syntheyes?

Also, my calculated FOV's are 12 to 25 degs. Is that normal for a helicopter camrera?


Anyways, I love this software and want to get it right.

Thanks

[damian_d3]

I can't really tell you what kind of problem it is.
It would help to see a picture of the solved shot. side view or so. with the placed geometry. How high is your RMS error. is it a solving issue or doesn't your object match to the real scene. are you allowed to show the footage? or
maybe a preview movie?

Damian

[ShadowMaker SdR]

12 to 25 degrees sounds low for a helishot.
Also as a tip, usually you get better solves in SE with less constraints instead of more. Especially if you don't know the right measurements.
Russ even has a tutorial on this.

[lightfuzz]

Hey, thanks for the replies...

In regards to my solved scene:

I do have a great camera solve, RMS error .4338 hpix.

My FOV is now 25.8 degs.

I did use the 3 point technique for setting up a coordinate system to initially align things. I then disabled them and have been trying to eyeball it by scaling, rotating, and moving.

I agree that too many constraints tends to cause too much trouble. My track and solve look good but my object isn't lining up correctly. I've been trying to manually align things but it is very tedious and it still doesn't line up. I can get one or two points lined up and try to scale from them (with the camera locked and the 'whole' button pressed) but still nothing useful. Should I be scaling only from the camera? Can I lock points in x,y,z and still scale the whole scene safely?

I would love to post some pics but this forum doesn't allow attachments. Know any reliable sites I could use to post images and movies from?

[lightfuzz]

I'm still working on this beast.
This scene represents almost 3 miles.


I'm not using the 3 point coordinate system. Manual alignment was not possible.

But now I'm getting great results by adjusting the world size to 130 in the Solver Panel. I did this because my maximum distance from the origin is -15000 units. I don't know if that's recommended, but that was my interpretation of reading about 'World Size' in the manual.
I also started using seed points on my imported OBJ mesh, which represents the road). Once I figured out how to adjust the clipping plane of my perspective view I could then place seed points for selected trackers. I then did a constrained solve and it pushed and pulled the point cloud into place. It's not quite right yet but getting there. My placement of the seed points on the mesh are just best guesses so I'll be tweaking those tnite to make it better.

Big breakthrough though. Use seed points to force the point cloud to fit the mesh better.

I wish there was a tutorial on this

[lightfuzz]

Here is the link for the above mentioned shot I'm working on:

http://www.vimeo.com/1020656
password: sr179


Check it out. I'm still baffled as to how to correct the highway from snapping into place as another pegged or seeded tracker comes into view.

Would appreciate some new ideas.

And yes, I know the water under the bridge is terrible.

Thanks.

[damian_d3]

Very very difficult shot actually.
You should now start to finetune the jittery parts.
But how is difficult to say.

I bet most jumps are the result of "locked trackers" that go offscreen.
You could go to refine mode and unlock those trackers or reduce their weight.

If you could post the sni file i could tell you more.

But good work so far

[lightfuzz]

Yes, it is a difficult shot. The original source was interlaced 640x480, then it was scaled to 720x480. The video quality is killing any thoughts I had of a smooth cam path. it's not too bad though. Regardless, Syntheyes is coming thru like a champ. I'm actually shocked it is doing so well and even under my naive guidance.


Yes, the jumps are in areas where a supervised tracker becomes enabled.
All of my 2D trackers (Automatic and Supervised) are locked. Shouldn't they be? I will look up more about locking but what do you know about the differences between locked and unlocked trackers?


My project file is 256 MB's. Still want to see it?