Matchmove camera properties. Help?

#1

Howdy,
I’ve been slowly experimenting and self learning the art of matchmoving for around a year now, and am by no means amazing, but am able to solve the shots I work on for personal projects to some degree.

It’s just becoming increasingly apparent now though, that I need more info about the camera and lens.
Whether in Nuke or PFTrack, the packages generally do a fairly good job of calculating the lens distortion, so that’s not such an issue.

The Issue:
My main concern is regarding focal lengths and the dreaded ‘film back size’!! My only camera to hand at this moment in time is a Sony Handycam (HDR-CX106e). I’ve derived from the spec sheet on the Sony site that it has a 1/5" Exmor sensor, and focal lengths from 3.2mm-32mm.

Focal Lengths:
Am I right in assuming that when no zoom is applied, the focal length would be 32mm, and when at its optical limit, it would be 3.2mm? Or is it the other way round? Or, worse still, have I completely misunderstood this?! :curious:
I’m hoping that this will ‘solve’ my current issue with the CG elements just not fitting with the 2D scene, as per the attached screenshot. The building itself tracked really well, but when I add the textured cube on top to extend the building vertically, it’s clearly skewed and not in keeping with the scene. I even tried manually skewing the cube and deforming it etc, but nothing really solves the issue.

Film Back:
Also, I have no idea how on earth I can calculate the film back size. I’m hoping it may be possible to calculate this from the sensor size (1/5"), but don’t know where to start.

Apologies for the epic post, I’ve got everything crossed for someone being able to shed some light on this one…
Huge thanks in advance! Feel like I’m hitting my head against a 2D wall here!
Chaz

#2

Can you post some video of what you are trying to do? It is hard to tell from the still what is going on.

A quick note about film back and focal length: These two variables are used to calculate the field of view of the lens. Even if you know the film back and focal length exactly it is generally a bad idea to lock it down as a known value in your match-move software. Getting really close to the true FOV can help you with a tricky shot but isn’t a magic cure to matchmoving problems.

3.2mm would be the wider end of the lens, then you zoom in to 32mm. I expect that lens distortion may be a bigger issue than you are giving it and I have found that getting lens distortion information is way more useful than focal length or film back. Take some video of some grids and see what the distortion really looks like at various zoom levels. Most of these handycam-like cameras have a tendency to produce quite a bit of distortion especially wide.

I had a tricky matchmoving project using a similar camera a while back and instead of fussing over the proper dimensions I shot a bunch of shots that were easy to matchmove and then figured out what film back and focal length values were calculated by PFtrack and then used those values on the nasty shots. The values were way off because the film back was set to 35MM size so an equivalent focal length was picked that was way outside of the real cameras specs. Like I said above all that is important is the FOV, so as long as the film back and focal length were off the same way it all worked out.

I only addressed a few of your questions, but hopefully this has been some help.

#3

Thanks for the info Aaron,
I’ve uploaded a short clip of the shot in question to MediaFire as per the link below:
http://www.mediafire.com/?ys6bf1m5bdqnbvm

It’s only 85 frames on account of the roto being incomplete at this stage, and very rough.

I’m going to try what you said with a lens grid. I’ve seen a few different types floating about on the internet, does it make a difference which one I use? Would one like this be ok? http://eyefish.tv/files/images/Lens-Grid-Chart.jpg

Also, perhaps this is an obvious question, but, do you have to be strict about framing when shooting the lens grid? ie. Does the grid have to fill the frame and be centered exactly? Or should it just be as close as possible?

Thanks again, this is a constant learning curve!

#4

For a lens grid you need horizontal and vertical lines, it needs to fill the frame, but you don’t have to be exact about it being parallel with the film plane or anything. It should give you an idea of what the distortion looks like for the lens.

What I am seeing from your test is that your solution isn’t very good and there is a lot of slipping. If you could roughly measure your FOV you could check and see if PFtrack is coming up with a reasonable guess.

Another possibility is that the stuff you are putting in top of the building isn’t in the right place, even if your track is perfect it won’t help if your geometry should be somewhere else. The way I usually do these kinds of things I model something existing in the scene (in your case I would start by modeling the top floor. Make sure the stuff you model sticks to the scene geometry. From there you can export your model to your 3D program or whatever to do the digital set extension. The key is having the reference to what is currently there and making sure that it lines up. It can also be helpful to use some of the stuff you modeled from the scene to use as masks.

#5

If you are using PFTrack or PFMatchIt you can input the film back size in inches. I did some math of the top of my head to calculate the focal length. 1/5" is about 5mm, a normal 35mm frame (22 x 16) is about 4.4 times wider.

3.2mm x 4.4 = 14mm
32mm x 4.4 = 140mm

Seems reasonable to me.

And dont forget to shoot distortion grids, 14mm is very wide (barrel distortion) and 140mm is very narrow (pincushion distortion).

#6

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.