Attach point is drifting inside track box

[Haugen]

I'm stabilizing using stabilize motion, and I'm doing this frame by frame. But the attach point inside the tracker is drifting from frame to frame. How to avoid this?

[Haugen]

I switched from "adapt feature" to "continue tracking". That seemed to work. I'm tracking an object that are a bit far away, so I guess the confidence went down? Am I right or wrong?

[Haugen]

Nope, didn't work. It's still drifting...

[Roei Tzoref]

Haugen, you need to read more about point tracking

Tracking and stabilization motion workflows in After Effects

the feature region needs to be as distinct as possible. this looks to me like a better candidate:

the search region needs to be smaller too. remember, Ae will search the feature region in the search region so don't make it too big.

practice some more and run through more resources in the link I sent you. if you still can't make it work, upload the footage so that we can take a look.

[Haugen]

Honestly, I have been reading about motion tracking, but I'm having trouble relating the text to what I challenge with. Therefor I'm having some trouble understanding everything.

But, both distinct and smaller search regions are good advices. I'll try some more.

But why is the the attach point drifting? Is there an easy answer to it?

[Rick Gerard · Accepted solution]

Motion Tracking in AE relies on tracking a detail area and it needs to be sufficiently large and have significant areas of a high degree of detail. Personally I think are not tracking enough detail. You are tracking all of the pixels in the inner box.

The attach point does not need to be in the tracking area if the detail you are searching is the same distance from the camera as the attach point.

The search area needs to be large enough to keep the tracker, the inner box, inside the outer box in the next frame. Your search area is big enough that the top of the tower could move almost twice it's width to the left and right in a single frame. judging by the distance between keyframes this antenna is moving around a lot so your search area is probably about the right size. I don't see the keyframe on the left ??? but the distance between the next two keyframes shows me that the tracking box is about the right size. I would have made the search area about as big as the red box.

Here's the other problem that I see that may be causing your attach point to drift a bit. Because the vertical lines are very thin the aliasing around them will make them shift a bit on the pixel grid. By making the tracking area larger you are reducing the potential error but the motion blur that may be in the other frames could be throwing things off. It's kind of hard to tell because you didn't tells us what the Magnification Factor for the Layer Panel is but a close up look at what you are tracking reveals a lot.

1. All of the color fringing tells me that there is either a lot of compression in this footage or you have an inexpensive lens
2. The antenna in the frame is so far from the camera that the top of the antenna is just barely wider than one pixel

If you check the colors and the edges inside your search area and compare them frame by frame you will probably see see a lot of changes in each frame. I suspect this is from a consumer camera and that it is MPEG compressed.  The lens is probably not that good either and the detail level in the camera may be set too high. The inconsistency inside the small tracking area is why your attach point jumps around. The more pixels you have to average the more steady the track will be.

Also, and this is a final point I will make, If you have a few frames where the attach point just falls out of position it's perfectly OK to delete those keyframes and make adjustments. There is no rule that you have to have tracking keyframes on every frame. You can easily average movement, especially jittery movement by simply deleting a group of keyframes and adding a couple of corrections manually. The only way to judge if your tracking is successful is to attach the artwork or element you are going to use to the track point and see how it looks with the comp at 100% Magnification factor, the resolution set to full and playing back at full frame rate.

I hope these suggestions help you understand how tracking works. If you are free to give us a link to 30 or 40 frames of the original footage where you are having problems I'll be glad to give the track a try and let you know what I think. Depending on what the shot looks like you might be better off tracking a good portion of the antenna in Mocha so there are a lot more pixels to average.

[Haugen]

Many, many thanks Rick!

I don't have a problem, but I had to manually stabilize every frame. And I couldn't understand why I had to and why the attach point would drift.

The camera and lens is actually a 5D mark III and Canon L lens, 17-40mm. So a rather expensive gear. I have exported RAW-footage from Lightroom to Jpegs. That's the reason for the compression. And I have to work with JPEGs since raw files will take to much time. Haven't tried TIFF.

And the tower is quite far away, so I had to zoom in a bit.

A was on a lake on a cycle boat. That's why the movement. I also could only stabilize position. I might try rotation and use the tower as vertical guide. But I find the position tracking quite good. I'll upload, so you can take a look. I'll provide a dropbox link later.

Here is the first frame:

[Haugen]

I guess, since the movement, it will be a lot of cropping if I want to keep this boat ride really smooth. When I was done stabilizing a had the tower in the middle, but in order to keep the image in the frame, I had to crop a lot.

Anyway. Here is the h.264.

Dropbox - On the lake.mov

[Rick Gerard]

The footage looks like time lapse or some would call it hyperlapse. That kind of footage is insanely difficult to track because of the tremendous differences in the position of the pixels. Next time you want to do that kind of thing shoot video and use warp stabilizer, render a DI (digital intermediate) and then speed up the footage using time remapping. You will get a better result with a lot less work.

There are techniques for stabilizing that shot but I would not have used the radio tower for my target. I would have picked something on the shore that was more solid, then in would have used Stabilize Motion, and probably cut the clip when something went out of frame. Then I would have parented the stabilized footage to a null and animated the position of the null to put a little smoother motion back in the shots. Then, because you were shooting way bigger than HD stills, I would have dropped that comp in a standard HD comp and scaled and adjusted the motion to get as smooth of a shot as I could.

The other option would be to just load your image sequence into a Hyperlapse app and let the app figure out the stabilization and the time lapse. I have an app on my phone that does that amazingly well.

[Haugen]

Yep, it's a hyperlapse, which I'm practicing for a bigger project. Of course, I need to carefully choose my hyperlapses to make it easier on me in post. This one on the lake is really hard because of all the movement.

In experience the automatic stabilization does a good job, sometime a really good one to. But if there is to much movement warp stabilization and other apps have trouble doing a good job. That's why I'm doing manually stabilization.

I'm better off shooting stills as hyperlapses than take a video since I haven't got 4K recording.

How do I drop composition in HD? Not sure what you mean.

But, using something on the shore and more solid are great advices. And I now understand more about tracking, and why the attach point is drifting.