I have a problem with the motion tracker in R19.
Static camera, Greenscreen Scene with X-markers….
I can’t track clean manual trackers, which I created on the x-markers on the green, throughout the whole sequence. The X-markers on the greenscreen are to blury because of DOF.
Because of that I can’t get trackers to solve the scene. :(
Is there a way to force manual trackers to a static position throughout a sequence without inserting keys frame by frame?
Edit: The project, as it turns out later, is a project for only object tracking with a lock off camera, starting at post #9 /edit
Hi Frank,
Would you mind to define “Static Camera”. To me it sounds like a lock off camera, with no motion. If so, you need the Camera Calibrator.
I guess the DOF blur is only on the green-screen, it shouldn’t be at all on the actor/object, with that I have at least hope that some support/survey C-Stands have survived.
However, hard to tell without any visual example. I could set up an upload link for a one second clip (or even a still), and keep it private. Let me know.
Edit; an upload link is in your Cineversity-Private-Message box/edit
Thanks, Frank, for the files, as well as for the care you have put into it. Very much appreciated.
As expected from your introduction, this is a lock-off shot: no change of parallax. With no change, no motion tracker will give you anything, but perhaps something that can analysis perspective lines. Hence my suggestion to solve the position with the Camera Calibrator. To be able to do that, typically one needs more data. In the supplied images I found four X marks, which obviously do not describe a rectangular square. If you can measure the four points again, each distance to every point, which means six values in total, always from the middle point of the original marking. This would allow for a drawing of the “original” situation. With that “blue print” of the floor, it might be possible to create enough perspective lines to match it. Since we have only informations of the x-z plane and nothing in the y direction, some manual work might be needed.
My suggestion for similar set ups:
Measure the lens nodal point hight from the x-z plane.
With the camera in the final position place some cubes (card-boxes) in the scene aligned to an x,y,z grid), remove these boxes when the filming starts of course. With this information it is often (exception here as well) possible to use the Camera calibrator.
Alternative: after the shot, keep rolling and move the camera slowly around while framing your previous area, thhi will generate hopefully all needed information, and give the wanted position of the other wise locked off position. With slowly I mean to avoid motion blur (less precision) or based on the camera eventual rolling shutter, which can’t often not really been fixed! (I haven’t shot with the BMD, but something rings in my mind that it had a global shutter?)
Please note:
Tracking Markers are there to show a change over time in the picture, hence the name motion tracking. Based on typically at least 7-8 of those trackers, if they are apart enough from each others, the algorithm can re-engineer the cameras nodal point (note that this is not the sensor distance, it is the point in the lens where all parallel rays would meet - considering the front lens element is the only consideration. (There are lots of other descriptions on the web, but please just ignore those, yes, that sounds arrogant, but I have tested most of them.) This nodal point is the point of no parallax change (if the lens/camera is rotated around this axis) and the C4D rotation point of the given camera. Given your small sized scene any millimeter counts!
A lens grid is typically needed as well, but since we are already more on the guessing side, that would do much anyway.
If you don’t have the markers anymore, you have to eyeball the camera. yes, not nice and not fun.
Let me know if you have these six measurements, I leave the upload option open for now!
P.S.: Here is a manual attempt. With the assumption that three points, the three points that had a measurement, are in an rectangular relation to each other.
This is a manual “calibration”, with a certain error in it. But perhaps it will work, since one will not have any other orientation or control option to check it (as audience). The image sequence had no meta data either. At least I had your camera data. I assume the camera has a read out of the UHD as submitted, and not a crop from a possible 4.6K.
I ignore for a moment the lens, as in no lens profile provided and if it is 50mm or +/- a few mm. Lenses do not change, the field of view changes, hence I’m grateful for the sensor size information. Since most 50s are without lots of distortion, I feel comfortable to ignore it for now.
It is a simplified version of Tim Dobbert’s manual matching suggestion (Matchmoving Edition 1 or 2: both around page 13). What he assumes that the reader knows is a little bit of perspective knowledge, that any parallel line (horizontally) on eye-level will create a straight line with other lines of the same elevation. We do not have any P.Y values here, except the unknown height of the camera. The hands in the scene allows for a tiny bit of a clue.
His method builds up in steps, which I have tried to apply here. Some parts are just unknown, so this is my best shot for now.
thank you so much for your effort. I feel like a beginner because of all the design flaws I made with the stage. I am an idiot.
I took all measurements. In addition, I crated a clean plate with the object (3D Print of the 3D Model) and to the measuremtents of the object as well. Maybe its possible to get the y axis that way??
In my last c4d file I added a camera called “Required Cam Position” which is the perfect cam position to match the sequence. I got this manually.
Is it possible to calibrate the stage based on this camera? I already made most of the animations manually but can’t get good results for some sections where the hand hold parts of the object in the air. Thats why I need the object tracker to
get the accurate movement PSR but I cant use the object tracker without calibrating the stage :(
And if a camera calibration would work, how can I use this calibration with the object tracker??
(Another question, if this will all work out: How can I bake the object tracking in the end?)
Since you mentioned tutorials on youtube and such, you’re right. These were not helpful. All the more I am grateful for your help.
(Since the object in the plate is a prototype, I would be grateful if you treat the data discreetly.)
Thanks again for all your effort.
Best,
Frank
(i will upload the files via dropbox)
Ah, and about the camera, lens and crop….
It was shot in 4,6k on a BM Ursa Mini Pro with a 50mm Canon 1,4mm lense.
Than scaled down to 4k uhd.
a new approach:
Since the object tracker can not be solved without a solved motion tracker, how would this approach be?
From the Motion tracker, I generate the Background Plate, then assign the trackers to the Object Tracker, attach a Null to the Object Tracker, bake the Object Tracker in keyframes (assign it to the null) and then attach my object to the null. Then I would have the movements reasonably on my object. Then manually animate all rough movements.
What do you think? Would that be a rough workaround?
Could you tell me please, how I can bake the Object Tracker to keframes (Null)?
Thanks again, Frank for the additional information. Very helpful. (Especially that the sensor was used completely, as in downsampled and not cropped)
You wrote how you feel. Well with all the information shared, that should be a something from the past, I do not share things to make people feel bad, the opposite is the case. I had to learn all of that at one point as well, and back then the tracker were not that good
I assume the prototype is the 3D print out, which will not be published here at all. What I did was to use just the four tracker markers, which have no secret at all, right!?
So, I needed that top-view measurement from the four trackers, so I could create a texture that matches the “set-floor” or the table. In the moment I was able to match this, I was able to exchange that with a texture that gives the tracker more date. Since we have no motion in the camera, I produced a few seconds of motion. This would be the initial motion for your sequence, to calibrate the Motion Tracker. Once that data is given, you should be able to do the object tracking (I have no object movement here, so this is for you to explore.
Based on a small test, the tracking features might be added every half or full second again, to get enough features, then 2D track and repeat.
Eventually you need to add on the end some motion, or extent the initial a little bit. I hope all of that makes sense.
With my little rant about Nodal point information, I didn’t want to slash all the YouTubers (that would be wrong!), just a fair amount who do not do a good research and warm-up what they have just learned in a video. I have, by now, watched over 6,000 tutorials and this is my base to say, there is quite some nonsense out there, and I’m unhappy that many people have the wrong information. But going by books, I’m not in agreement with all there is either. Yes, I’m picky, but we live in a time where evaluation of the sources is perhaps the most important part to create a useful base to develop one’s skillset from. Again, not all is bad, but …
Please note that this wouldn’t be a suggested workflow, if the camera could do the take in one step, that would be much better of course: Motion Static Motion. With mote trackers as well.
Anyway, if I had only this given material to work with, perhaps that would be my way, besides manually tracking the object of course.
Erica Hornung, has written a book about “The Art and Technique of Matchmoving” (Solutions For The VFX Artist) Focal Press.. Her main part in the book is manual “matching” the motion.
As mentioned, and based on the progress of getting more information, let me summarize, a camera position from a locked of (static) camera is not possible with the Motion Tracker. The Camera Calibrator would be the key. Which I have discussed a while ago for a panorama shot of a “drone”, to establish the position and use the Motion Tracker for the Pan. I mention this, as we have several options to combine. In anyway, either perspective lines or changing parallax is needed. Those wer my suggestions for the initial question.
Later you mentioned Object Tracking. This is possible even while ignoring the camera all together, but some data needs to be known. Since I haven’t seen any object-motion-footage, here is my one minute clip. Please note that it can be based on automatic or manual tracking features. IF the tracking features are very low in numbers, a support geometry can help. Anyway, here is a quick run-down for a still camera but object motion based workflow with auto trackers and no geometry. The resulting keyframes are in the Object Tracker itself. Those can be transferred via Xpresso and Cappucino. Alternative: Xpresso and Timeline> Functions> Bake Please ask me if you have a specific question.
With shift selection, one can select more and more features, in the capture below, I have added them after each single selection, you pick which method you like more.
There is of course one key question, if we don’t solve the camera point of view/point of interest, field of view, what is the solve-algorithm in the Object Tracking doing with the data. The simple answer is, the camera that is a child to the Motion Tracking object will be used as reference. Since one can’t change the position or rotation of this camera, nor the field of view, how do one get a specific still camera adjusted, and with that the solve data? The camera position and rotation can be set by adjusting the Motion Tracking object, the field of view/Focal length, etc can be adjusted in the 3D solve section. All of that should be done before the Object Tracking object solves the data.
Please keep always in mind, even the C4D tracker is easy and fast, the field of tracking is a profession in its own rights, and many artists focus exclusively in studios just on this work. To have these nearly “one push button” solutions lead sometimes to the idea that there is not a lot to know about, but I guess that has to do with something some of my friends like to say: “If you can’t see it in a movie, then it was done well. It is an invisible art after all.”
sorry for the late reply and thanks for the book recommendation. I will get this in a timely manner.
And thanks for your effort with the texture and the scene setup and the video file.
I’m really thankful for your efforts.
I have added 4 cubes to your scenes as further references and rendered the sequence. I have now used this sequence for the motion tracker. Thanks to its texture, the Motion Tracker finds a lot of tracking points. Nevertheless, I still have problems with the calibration. I tried to get it done with a planar constraint, didnt work, then with vector constraints. The latter worked one time, but the y-axis wasnt right. Although I’ve read a lot, and so on, I’m probably still doing something wrong. :(
I didnt try your “brain-storming” idea yet (Reverse, moving world, static camera) due to timepressure on my running project. I really need to improve my matchmoving skills and I will definetly do this asap.
I`m a bit desperate beacuse I try to get the scene calibrated for weeks now and I really have to solve this problem now.
I’d be grateful if you can help me with a workaround for now:
Since I already have the right camera position and already did a lot of animation manually, how about that:
I already use manual tracking markers on my object-tracker. It´s working fine, its just not calibrated to the stage (for that I`d need to calibrate the motion tracker with is not working for me at this time).
If I could bake the Object-Tracker Movements to keyframes of a null, then put the null into another null, I should be able to reposition this keyframe animation manually to my 3d-print-Reference of my film-sequence, right? The in and out to these object-tracker-keyframes I´d animate again manually.
Is there a way to bake the Object-tracker to keyframes?
So sorry that you have so much trouble. If the scene gives not enough information, the camera path needs to be expanded. However , as you didn’t mentioned the Object tracker in the initial post at all, I was focused on the Camera tracking, and how to help you to improve it. In the image-sequence was a bluish “blob”, later on I was told that is a disguised prototype, and this was the main thing you wanted to match in terms of tracking. So, that information from the start would have helped a lot to shorten your time. (I deleted the reverse idea very quickly, as it needs a lot of math to counter the movements… . It would take a week to make a tutorial series about alone, going from the math basics to the problem, so, no way).
With that new and more specific information I tried to find any tutorial to show how this would be done. I hope I didn’t miss anything, as Cineversity has quite a few tutorials about tracking, and they are all great -no doubt, but object tracking from a still camera source was so far not explored. AFAIK.
Hence why I wrote post #9 (please explore the content of this post). Since you have the camera position, I described above what to do to get the Still camera work in concert with your settings. It is possible to create the needed Object Tracker information without a moving camera. However, I have not seen more than a second of the footage and there the prototype was just a disguised blue blob. So, how far you get with the tips of post #9, I really don’t know. Get as many tracker features on this object.
I always try to improve my skills and and like my projects built in an orderly manner thats why I was starting with the camera tracking in the initial post to start from scratch the clean way. My fault, I should have mentioned my time pressure. Although I`m not a fan of workarounds, the baking has to be enough for now.
But I really apreciate all your tips and tricks and food for thought. I will get into this deeper asap.
And thank you very much for the baking video etc in post #9. Sorry for the repetitive questions about it.
I must have skipped it due to my workload. So sorry….
About tutorial resources for object tracking from a still camera: Yes, I was searching a long time until I posted here… couldnt find anything.
I will check #9 out right away.
If I still have questions after that, I would be grateful if I could get back to you again?
Thank you very much for everything. You were really helpful, once again.
I mentioned the sequence of question/information along the thread only to make clear how it went, not to blame or make wrong. No sorry needed. It only encourage me more to ask more about the real target, as I get often not the real target, what I get is the problem “half the way” and fixing from that point introduces often more problems than to really start from scratch.
You ask as long as you feel not clear about it. OK? I have all the patience in the world, as long as I see collaboration
My interest here is only to support your learning efforts, and my credo it: there are no mistakes, only gained experience.
Never hesitate to ask, as others might be happy to find a question already answered. So, feel good about anything! Thanks.
Keep in mind that I can’t really see what the final result will be, so I simulated the 3D-print-surface with noise, and did it many times, to share data based on my experience with a (hopefully) close approximation of your set up.
If I had to rotate this prototype, I might would have put little needles in it, these with the colorful spherical heads, to expand the size and overall visibility. (The spheres on Per Holmes SciFi gun dummy, in the tutorials, will support my idea here.)
Edit: XPresso and baked was the question, it is now gone, I leave it here. /edit
After it is baked, just delete the XPresso. Alternative, there is an Attribute Manager> Enable checkbox (while the XPresso tag is active)
You also can bake a copy. This copy is not part of the XPresso.
If you use a Null, never use a short cut to group things, while there is animation anywhere involved. I know, this short cut is mentioned a lot, and I would love if it would just vanish, based on the automatism (muscle memory) it creates and the HUGE danger it has. A Parent-Null to any animation should be on position zero and no scale nor rotation either, in this way, the animation stays intact.
One more option to move an animation is, while you go into Animation Mode (you find that in the left bar inside of the Model/Object mode), then you can just move, rotate or scale if needed. The rotation will be based on the time (with this the position) the object has. Then go back to Object or Model mode.
