Nuke 3d Track problems!


Hey guys.

I’m fairly new to Nuke and tracking. I have taken a compositing class already and I’ve been compositing in After Effects for a couple years (Want to be a compositor once I graduate, currently at Full Sail).

I’m trying to get a better understanding of Nuke because it’s mainly used in film and such. Currently I want to get the hang of 3d tracking, but it’s giving me problems and I need it for a side project. I’ve been able to follow tutorials and class situations online, but when it comes to taking my own footage and doing it myself it doesn’t go so well.

Here is a jpeg from the sequence I’m working with as a test shot

Basically the camera is not on any form of tripod and it’s free moving. We put a couple of trackers on the wall because it’s a solid color and not much to track.

Basically I’m still getting a crap track. It gets some decent points and such, and I go through cleaning some of them up. A person walks by into the entrance, but I have him roto’d out so that he doesn’t mess with the track. I did line lens correction to straighten out the lines as well.

I’m not sure what I can do to make this work. I tracked it, roto’d him out, cleaned up points and lowered the error range. I have a couple similar shots and I’m getting pretty much the same results.

I also am having issues getting the point cloud to look like the scene, even after I define some Axis points on the same lines.

Really could use some help here if someone knows a little bit about this. Thanks!


You can give the cleaning lady some time off. Reflective
surfaces are worse than moving object and i see enough
of it. Put some furniture or something in front off the walls,
mask out the floor and see how it goes.


I tried rotoscoping out the floor and it still didn’t really work. It’s even picking some bizarre places to put tracking points, like in the middle of the white wall. I raised the points to 350 and I also raised their time limit to be around 13 frames of life, so that I would get better tracks. Still to no avail.


Without seeing the footage and being a lame matchmover myself
all i can suggest is to solve it with usertracks only (you will have
to correct some of those manually of course). 10-12 should suffice.


Can you host a jpg sequence or mov anywhere? no one can really help you from a still.

Nuke is usually pretty good at making a guess at the match-move and then the rest of the work is refining it. With that in mind you need to give it the best chance…

Try giving it more of the clip, rather than just the portion of the footage you intent to make into a shot; often the few seconds after hitting record, and after the action can be vital for solving a tricky track; the camera might do something outside of the shot range which may just save it, so if you are not doing so already, try adding some onto the top and tail of the clip (would need to see the footage though!)

If you haven’t done already - start again from scratch with a new approach. Your lens distortion may be doing more harm than good if it’s incorrect for example. try it with no correction, or different correction.

lastly, depending on the move I sometimes find it helpful to lock the cameras allowed motion to one of the pre-set like ‘planar’ if the camera is mostly hand-held and doesn’t really move up and down as much as it moves side-to-side. This will force the path to be planar, and then the trick is, after the first solve to change it back to free-motion and then hit the refine button; this will take the basic planar motion and allow it to move a little more if it needs to.

truthfully though no one can help you without seeing the footage; it’s literally an impossible question that is asked a lot - those tips I have given may just not apply at all to your shot.


Have you put your focal lenght and film back size? And try to apply a denoise to your plate like reduce noise from Neat Video.


Well, without seeing the sequence, it’s hard to say.

First I would say, don’t worry about lens distortion, most times it’s not an issue and your ‘correction’ may do more harm than good.

I can see a dozen nice points there and you don’t need more than that to solve a camera, so don’t assume just because you don’t have hundreds of solid points that you don’t have an accurate solve.

Finally, without knowing either your workflow or your footage, it’s hard to say. My first thought would be to set the initial frame in the centre of the footage and then solve backwards and forwards from the middle. Put 1000 low accuracy points in there and then just pick the 10 best.

Otherwise, solve it in another program like mocha and import the data back into nuke.


I would track distinct points manually with the point tracker and then import it into the “UserTracks” of your camera tracker.
If you have the proper focal length AND Film Back (sensor size) pop that in there. Try a solve with and without lens distortion calculation (None & Unknown).
I’ve gotten much better results manually tracking in Nuke (something I’ve not seen many others do, but helps a lot).