FreeTrack Forum

Surely in theory is as you said, but in practice for me there is a huge difference between 3p-clip and 4p-cap, despite having lost hours and hours to improve the clip.
However, I am happy to know that there's a new version on the way, I will wait for!
thanks for now Kestrel

Sorry to resurrect a really old thread, but I have recently built a 3 LED clip model and I am suffering HUGE problems with instability / sensitivity too.

To be honest, I'm finding the whole thing a vast let down, but I'm wondering if it could simply be the cap/clip model that is the problem?

I'm trying to get tracking working (properly) with IL-2 Sturmovik 1946, but I'm finding it almost impossible to keep my head stable enough to prevent jittery random movements.

I'm using a PS3 eye with IR filter removed. Tracking perfectly using USB powered SFH485P IR Leds. I've tried 320 x 240 and 640 x 480 at all frame rates from 30-120 fps.

I have used Freetrack 2.2 and the very latest alpha of opentrack (2.3 rc6) with all sorts of filtering and sensitivity adjustments, but it just isn't 'right'.

The model is almost exactly the same dimensions as the reference model.

Has anyone apart from Luca actually (practically) TESTED the difference between a 3 LED Clip vs a 3/4 LED cap to verify if there really are noticeable differences in head stability / accuracy?

Thanks

Hi,

check your setup with these default settings. (Of course you need to measure yours to fill up the sections.):



Than go further with the "Profil settings" I mention here:
http://forum.free-track.net/index.php?showtopic=3444&message=19106

Hi Steph

I'm not sure what your point is (excuse the pun)?

The dimensions of my clip are within a couple of mm exactly as per the reference model and the other dimensions have also been entered too with respect to the head model.

The points are being tracked absolutely fine. That is not the problem.

The problem is that there is 'jitter' irrespective of how still your head is. I even mounted the clip on a tripod and the effect is still there.

I'm beginning to wonder if there is a fundamental flaw in the software in calculating the points? Either that or the camera is excessively 'noisy' leading to incorrect points being output? if that was the case though, it doesn't make sense because visually, the ONLY thing on the screen are the 3 tracking dots!

If you compare any number of Trackir4/5 videos on YouTube with my results, then it's like a day and night difference.

The Trackir system is rock steady, stable and accurate, unlike freetrack/Opentrack where the view is drifting/jumping/unstable.

With Freetrack it's almost impossible to move, look and 'maintain' that position.

Further edit:

I just mounted the clip on a tripod again and set the camera on my desk (to eliminate monitor movement). The desk by the way is huge heavy solid pine effect. I reset the view so that all the readings are 0,0,0,0,0 and the Z parameter (especially) just goes up and up by itself! I can occasionally even see the crosses moving on the 'cam' screen as a new position is being interpolated.

I mean WTF? How? The light source is constant, everything is rock solid. What is going on here?

To be honest I didn't actually realise just how bad it was until now, because in opentrack the figures reported are quite large, but in FreeTrack the X, Y and Z translation figures are in 1/100s of a mm which translate to noticeable movements on screen of course.

Is this a hardware or software or clip fault? As a head tracking system, it's completely unusable anyway.

Hi,

this is definitely a problem of your settings. Probably the cam settings.
Freetrack works as good as TrackIR (I have both running on my PC, preferring Freetrack for the advanced settings and TrackIR for it's simplicity).

https://www.youtube.com/watch?v=RSAp_Zpg5Og

Please check if you got three dots with white crosses in the cam review that do not flicker when moving. Like this:



When they are bigger then reduce the point size:



If you got jitter (JPS counter red) your cam are probably tracking more (or less) then three dots.



What kind of daylight filter you are using on the PS3 cam?

You need to be aware also, that you do not have one to one movements, because you are multiplying them ingame. That is quite disconcerting in a first time and it needs a couple of hours to really accustom to it.

Setting some large deathzones to the curves in the beginning can be helpful. Like the "profil settings" I mentioned in my first post above

Hi Steph

Thanks for trying to help.

Unfortunately, I think you're missing what the problem is here. I have no trouble tracking the points and there are NO missing frames or jitters with the camera/filter.

The problem is with maintaining a steady view and I think I've found the problem. It's the software!

If you (or anyone else) can verify the following I'd be really grateful.

First, you need to mount the clip securely to eliminate movement. Set all the graphs to 'linear'.

Set up a typical profile for yaw, pitch etc. say 1.6 or so and smoothing to mid-way. Now centre the view.

OK, here's what I've found:

First, look at the 'Image points' display. Because the clip is securely mounted you should notice that the X/Y coordinates of each LED are rock steady, with the exception of brief transitions between e.g. nn.4 and nn.5, but always between those two figures i.e. not a gradual drift up/down. I think this proves that the hardware i.e. the clip and camera are functioning fine.

Now, on the 'advanced' tab of the profile, set 'Average' fully left to 1%, now hit the Centre button again.

What happens here is that the yaw, Pitch, Roll, X, Y and Z figures change DRAMATICALLY from 0 to all sorts of figures plus and minus of 0!!

If you go to the advanced tab again and set the 'Average' value to 100% then hit the Centre button, the only difference is that the random change in values of the output coordinates is slowed down, not STOPPED!

It seems that the 'transitioning' input coordinates are sending the calculated output values out of whack, because its algorithm is not working right.

You would think that, this effect would tend to cancel itself out when the transition goes back to its original value, but it doesn't. Instead, the output values get further and further shifted away even though the input coordinates haven't changed a bit!

Conclusion:

The software is BUGGED.

At least it is bugged, as far as calculating output values for the clip anyway! Whether it's the same for a cap, I don't know, but someone here could surely test it?

Could you please verify the above? The results speak for themselves. The software as is is unusable.

Further edit:

I get EXACTLY the same effect using the 3 point cap model and 1 point model!

I do verify these settings, but I do not really understand where you want to go...
You set the profile settings to extreme values and the output isn't stable.
I agree.

Every LED dot on the cam preview window is made of of several pixel.
The software is choosing one of them on every dot to calculate three virtual points. These points are marked by a white cross. With big dots and high sensitivity the crosses jumped around within the dots. So the way to avoid this is to limit the tracked dot size and adapting the sensitivity.

That's why I ask you for your filter solution. Please tell me if the crosses stay within the dots and if you haven't more or less than three crosses continuously.

Please set up the profile settings and curves in "every" point at the basic setting I mentioned in my first post.

http://forum.free-track.net/index.php?showtopic=4039&message=19522

With these setting you will be able to get a stable tracking like in the Microsoft Flight video. I do myself use Freetrack in IL-2 1946 and got the same stable tracking.

If not, we have to go further in investigation.

I can't tell you whether or not the observed behaviour is a bug or whatever,
but I can tell you for sure that a perfect stable tracking is possible with Freetrack.

I've already said that I don't have a problem with tracking the dots and I never have more or less than 3.

The crosses always stay on the dots.

The clip is mounted on a tripod i.e. there is no movement.

If the view drifts with no change in input parameters, how can you call that 'stable'?!

I've got the same result like the Freetrack Microsoft Flight video and I only need to center once after starting the tracking, that is what (for me) I call "stable".
Reproducing your experiment I can observe little movements around the centre, but no drift or constantly increasing values. I do eliminate these movements with deathzones in the curves and increased smoothing.
If you don't get to a similar result with the basic settings we need to check again the model position or cam position.
Could you link some screenshoots of the different settings?