Mixing tracking imput to get more accurate Head Tracking?

[moukrea]

Hello again, new topic today, I'm wondering about something that could bring a really neat head+positional tracking to the next level... If only it's possible!

So here's the catch, Oculus uses two imputs at the same time if I understood well: The internal sensors + Camera on computer and IR sensors on the headset. That said, I guess they use a kind of algorythm mixing both imput to make the most accurate headtracking possible!

So here's what I would like to do, I get head-tracking from my phone (actually pretty accurate already) or could be from a DIY sensor. I'm getting this as a TrackIR output (could be anything else).
Now, what if I put IR leds all around my headset and use FreeTrack (allowing 360° tracking)?

I mean I would like to get both data into FreePie, and mix them into one much more accurate output! Like one imput correct the other one, and make them more smooth, less jitter... make an average out of the two! Do you think that's possible?

Thanks

[CyberVillain]

It's possible with some sensor fusion, but its not built in to freepie, have a look kalman filter for example

[moukrea]

Mhhh I looked quickly at Kalman filters... Seems kinda hard to do such a thing! I'm wondering... Is that means the sensor data from our phones are filtered this way? Knowing they got Yaw, Pitch and Roll from Gyroscope + Magnetometer and accelerometer?

I noticed OpenTrack can use Kalman filter, is it possible that it makes Sensor fusion by itself?

And one last question, I know all of that is a bit out of topic but well... Can't stop thinking about it... Is the Oculus Rift doing sensors fusion too? Or they just use the camera for positional tracking?

[CyberVillain]

They all do sensor fusion, we too do some sensor fusion, if you use a Android with raw data we use a fusion called Magick to fuse the data into yaw pitch roll

[konstantin_lozev]

Mhhh I looked quickly at Kalman filters... Seems kinda hard to do such a thing! I'm wondering... Is that means the sensor data from our phones are filtered this way? Knowing they got Yaw, Pitch and Roll from Gyroscope + Magnetometer and accelerometer?

I noticed OpenTrack can use Kalman filter, is it possible that it makes Sensor fusion by itself?

And one last question, I know all of that is a bit out of topic but well... Can't stop thinking about it... Is the Oculus Rift doing sensors fusion too? Or they just use the camera for positional tracking?

Good optical + IMU sensor fusion for positional tracking is not an easy business. It took many iterations for some of the brightest minds at Oculus to get the low-latency positional tracking right and they had to go to LED blinking codes on each LED in the Constellation LED array https://www.youtube.com/watch?v=O7Dt9Im34OI and https://www.youtube.com/watch?v=X4G6_zt1qKY. Reports say that even the OSVR HDK has not yet nailed its own optical positional tracking yet to reach same level as the DK2.

[moukrea]

Thanks for this informations, much appreciated

Indeed I'm not looking into getting an as good as Oculus tracking at a extremely low latency, it would be more about the optics correcting an eventual drift coming from the sensors while getting positional tracking values... The blinking leds is a smart idea for fastest recognition of position of each leds, I'll be looking deeper into it, thanks!