Why does my .7 solve error track result in a weird, thin Tracking Scene?


So! I’ve done some more research into Camera Tracking - and one of the things I’ve discovered is that I tend to set wayyy more markers for footage than most compositing Blenderheads need for good tracks and end results. Like, I’ll set 90+ markers, but often I see great tracks made with 20-40 markers.

I wiped my progress and started again with my crappy iPhone 4 footage and managed to get a solid 0.7159 solve error (using the phone preset w/k1+k2+focal length) with just 21 markers (including my new Favourite Thing, perspective markers). Huzzah!

Except when I Setup Tracking Scene - the presented solution shows the camera and the markers in an almost completely straight line, or has the camera stretched out to a ridiculous distance, covering the markers (which are in an almost completely straight line).

I’ve set the floor, the origin, the x axis, the scale - all to only minor adjustments.

How can I convince Blender that it’s looking at a fairly broad street, not a tiny, thin path?

–Rev

I don’t believe your phone camera has a lens with a focal length of 457mm as you have set in the lens settings

Huh. Weird. You’re right, of course. I’m not sure how it got that value.

Resetting the Blender Camera focal length to 1, removing the Camera Solver, changing the Camera preset to iPhone 4 in the Movie Clip Editor properties (sets Foc to 3.85) and then hitting Solve Camera Motion gives me…


… goddammit. But then I’ve just tried solving the camera motion again directly after I took the image and a) I got a solve error of .6420 and b) the camera problem is solved.

… So, honestly, I have no idea what in fuck is happening here. But the problem seems to have gone away. I dunno.

–Rev

A couple of things, this sitesays the iPhone 4 has a 35mm-equivalent F.L. of 29mm. Divide that by the F.L. multiplier of 7.6 gives you a F.L. of 3.81.

Also, how much parallax does your video have? The choice of your two keyframes can also have a big impact. Post the video on Youtube and give us a link. We can critique it for you.

You probably don’t need 90 markers unless the camera is in a car while driving down the street. Here’s a rock-solid track I did using just 10 markers. There’s plenty of parallax.

Steve S

Hi Steve!

Well, it turns out that the track I got last night wobbles like a jelly sculpture of a hippo in gale force winds towards the end of the footage, so I’m back to square one.

I should point out that the first time I tracked this footage, I got a fairly stable, consistent track with a solve error of 1.3343 - and I only needed 124 markers to do it (keyframed between 80 and 200). As I say, I’m trying to reduce the amount of markers that I use.

Ultimately I’d like to use the tracker weighting trick mentioned on thispage to get lots of good geo reference. I’d almost rebuilt the street using my previous track, so I’m excited to see where I can take this…

The culprit, below:

There’s a fair amount of parallax and some decent features to track (especially with my lovely perspective markers), but the footage is quite wobbly.

I’m going to give the footage another go now, but I appreciate any notes you can give me!

–Rev

EDIT: By the way, just looked at your Youtube account - some really lovely tracks there! It’s great to see this sort of thing done really well in Blender, gives one hope. =)

You’re getting some pretty serious “jello” effect from the phone’s rolling shutter, especially during the high velocity wobbles. I’m sure that’s partly why the reconstruction is so difficult. I don’t think Blender’s tracker includes a way to filter out jello effect yet, but you can remove it in Virtualdub or After Effects.

Dove into the video last night and was getting a fairly stable track (until the very end) with about 77 markers (with a solution of 1.1829).

Have learnt to identify from the Movie Clip Editor that I get red in the MCE timeline when I don’t have overlapping markers (also, the blue line of the solution in the graph editor completely cuts out). Huzzah!

Unfortunately, having decided to start all over again with the jello removed via VirtualDub, I’m getting a 'No audio decompressor could be found to decompress the source audio format. (source format tag: 00ff)" error in Virtual Dub. (I’m running Windows 7 64Bit)

I’ve already installed Quicktime, I’ve installed the ffdshow codec pack and I’ve tried installing the AACACM 32-bit codec (I’m running a 64-bit system, but I’ve been advised to use the x86 version of Virtual Dub as it has more plugins). I’ve run the 32-bit codec.bat file with AACACM, but I’m unsure that the codec has installed.

Has anybody experience with Virtual Dub that can help me?

Some success with Virtual Dub with the kind assistance of folks on thisboard. Original:

End result, post VirtualDub/Deshaker plugin:

Hmm.

Won’t be able to work on the footage for a few days, but I’m curious to see how Blender works with this new smoother-yet-with-odd-hiccups-and-blurs footage. Feedback appreciated…!

–Rev