I had an idea where each sensor emits a unique ultrasonic chirp at a known time, then the other sensors have little microphones so they can listen for the chirps and know how far they are from each other, this will provide another bunch of measurements (distance from speed of sound/timedelta). So the computer can figure out where each sensor is in space a bit easier (it must lie on the surface of a sphere with radius of measured time).
Thanks for this...it is helpful even if IMUs are being looked into for other reasons (drones ). Plus I like your sense of humour. Great job and thanks for sharing!
I'm new to all this. I'm using the on board processor, which "removes" the gravity. I have tried it by getting the magnitude of all and by component. When looking online it seemed that everyone does it by component then they numerically integrate it to get the "current velocity." The issue I got was that the numbers would never settle back close to 0 when the sensor is at rest. How would you go about this?
dj deb a ha. Yes this is the problem with most of these sensors. They will never really be "at rest" and will fluctuate around the zero point. Not only that but the cheaper ones have a tendency to drift. Have you seen my video on IMU drift? Check that out. Also my other one on how an accelerometer works.
Hi ! is it possible to get intrinsic angles of rotation from a given quaternion ? I mean, if i rotate an imu about his X axis, so i get a changes just in one angle and not in all angles ? Thanks for answer
I had an idea of using 3 IMUs placed at 90 degrees to each other, using the accelerometers and combined gyro data to check the gyros against drift. The accelerometers would confirm no movement, and the output of the individual gyros would somewhat self-correct because they all won't drift the same direction, at the same time, because of the orientation. So if X1 (X on gyro 1) started to drift, which corresponds to Y2 and Z3 (example), and they respond with not the same data, the sensor or input can be reset/adjusted. Not sure if it'd even remotely work, just an idear I had and haven't gotten around to building/testing yet. Does your experience indicate if this even remotely viable?
+gymkhanadog You're spot on! That's how a low of military grade sensors work. Even the high end ones suffer from drift. So they set several out in multiple axis to correct.
If you are interested in a short read, comparison of LPT & IMU, commercial products during the squat exercise: assess2perform.com/blogs/assess2perform-blog/from-the-training-floor-comparison-of-2-vbt-velocity-based-training-technologies
+Izzie Yes yes. That's one of the methods I am looking into, but not with s Kinect. There's some cheaper ways, but i need to test to see if it can keep up with the frame rate.
MickMake too much physical maneuvering. I would try a logic approach. maybe 3d image storing, video recording, matching image, calculation of body position. just thought🐒
Try out RepCam itunes.apple.com/us/app/repcam/id1118115900?mt=8 in freestyle mode. I think it should meet the basic requirements you listed (although not by way of skeleton tracking) and you don't need to wear anything.
i am showing my son (7 atm) every vid i can get him to sit for as a learning platform. thank you for your feed
oh and your humor is great
Ha ha! Fantastic. Never too early to start learning.
I had an idea where each sensor emits a unique ultrasonic chirp at a known time, then the other sensors have little microphones so they can listen for the chirps and know how far they are from each other, this will provide another bunch of measurements (distance from speed of sound/timedelta). So the computer can figure out where each sensor is in space a bit easier (it must lie on the surface of a sphere with radius of measured time).
Any update on IMU potential for weight training tracking?
Thanks for this...it is helpful even if IMUs are being looked into for other reasons (drones ). Plus I like your sense of humour. Great job and thanks for sharing!
Thanks mate. IMUs have come a long way since that video, but still there's more room for improvement.
Awesome man ! any updates so far ? :D
Wonderful video! Any recent updates on your attempts since then?
you cam try oblu (imu based positioning). It is pretty accurate.
Thanks mate. I'll look into it.
Great video! Did you ever post the follow-up?
Not yet! So many things to get through and so little time! :-)
Our Bar Sensei barbell mounted sensor came out in Sept 15, 2016, almost to the day this video, something to consider next time...?
Hi,
Would be interested to see this in action!
Send us a note please via our Twitter
what would be the best way to go about tracking the velocity of a IMU? I'm using the linear acceleration, a trapezoidal summation isn't working.
For my purposes I resolved acceleration down to a single acceleration value by doing this:
Square root(X^2 + Y^2 + Z^2)
Is that what you tried?
Oh also... don't forget that you'll need to remove the gravity component out of the final result.
I'm new to all this. I'm using the on board processor, which "removes" the gravity. I have tried it by getting the magnitude of all and by component. When looking online it seemed that everyone does it by component then they numerically integrate it to get the "current velocity." The issue I got was that the numbers would never settle back close to 0 when the sensor is at rest. How would you go about this?
dj deb a ha. Yes this is the problem with most of these sensors. They will never really be "at rest" and will fluctuate around the zero point. Not only that but the cheaper ones have a tendency to drift. Have you seen my video on IMU drift?
Check that out. Also my other one on how an accelerometer works.
ruclips.net/video/8E7C36wawqg/видео.html
Hi ! is it possible to get intrinsic angles of rotation from a given quaternion ? I mean, if i rotate an imu about his X axis, so i get a changes just in one angle and not in all angles ? Thanks for answer
@MickMake
Just spotted this video here: great tests and explanations!
I was wondering: have you tried using an altimeter/pressure sensor ?
Thanks mate.
Yup. Tried using altimeter. The issue with using an MCU for location is that it’s still a dead reckoning. Every error will multiply out.
I had an idea of using 3 IMUs placed at 90 degrees to each other, using the accelerometers and combined gyro data to check the gyros against drift. The accelerometers would confirm no movement, and the output of the individual gyros would somewhat self-correct because they all won't drift the same direction, at the same time, because of the orientation. So if X1 (X on gyro 1) started to drift, which corresponds to Y2 and Z3 (example), and they respond with not the same data, the sensor or input can be reset/adjusted.
Not sure if it'd even remotely work, just an idear I had and haven't gotten around to building/testing yet. Does your experience indicate if this even remotely viable?
+gymkhanadog
You're spot on! That's how a low of military grade sensors work. Even the high end ones suffer from drift. So they set several out in multiple axis to correct.
If you are interested in a short read, comparison of LPT & IMU, commercial products during the squat exercise: assess2perform.com/blogs/assess2perform-blog/from-the-training-floor-comparison-of-2-vbt-velocity-based-training-technologies
Any update on this?
Have you implemented kalman or complementary filter yet, to minimize drift?
+Johan Sandgren
Yup, that was all in place.
Interesting stuff!
Just looked up what you guys are doing over the pond. Interesting.
Thanks very much!
What about using a Microsoft Kinect?
+Izzie
Yes yes. That's one of the methods I am looking into, but not with s Kinect. There's some cheaper ways, but i need to test to see if it can keep up with the frame rate.
Kinect has serious floor issues. Any exercise requiring you to lie down horizontally seriously fails. It is also very noisy.
What did you do to him (gym owner), last time?
He asked a gym member if they wanted to MickMakeBabies.
+Suraj Grewal
Ha ha! It was a setup gag.
MickMake too much physical maneuvering. I would try a logic approach. maybe 3d image storing, video recording, matching image, calculation of body position. just thought🐒
Bruno Hoffmann
Yup, that's what I've been working on instead. I'm also thinking of using double or triple IMUs to counter drift.
MickMake probably a combination of autodrive recognition software and facial recognition and anatomy software.
what kind of software that you are using?
I wrote my own from scratch.
@@MickMake Thank you. Is it including the 3D model? you wrote by your own ?
What type of IMU module were you using? :) I sub'd
Thanks! I'm using an MCU9150
I think for 3:40, you have to adjusted your position by yourself
Try out RepCam itunes.apple.com/us/app/repcam/id1118115900?mt=8 in freestyle mode. I think it should meet the basic requirements you listed (although not by way of skeleton tracking) and you don't need to wear anything.
Lol, what unlucky font and logo , the company atlas has...it looks like straight out of call of duty