i m actually building for one of my performance a 2m long stick, with in its middle a MPU 9250 (gyro+accelerometer+compass). The idea is to activate hitest zone when pointing in a defined direction. Its relativ to my body ( up right, down left, middle front ect…).
I m receiving good data in a Yaw Roll Pitch format from the arduino library used But… i m wondering of to create hitest zone in 3d, intersecting only one end of my cylinder simulation.
On the patch here you will find a representation of the stick and its orange tip, wich should be recognized when touching one of the two 3D targets.
yes its great ! now fighting with the mpu9250 just to find out how to avoid inversion of data if Z axe is inversed ( roll). I dont know if there is any tool enabling compensation on this classical state of devices ?
thank you again :)
i had a look inside wekinator, at videos, and the idea of “machine learning” and did some tests. Helas, Trained model is just a kind of queue, plus mapping and noise cleaning.
by machine learning i was hoping a somehow “magical” program learning from the moves i m doing, enabling complexees identifications: hey this a 8 figure, hey this is a break, hey you are up to down ect…
anyway continuing reasearch for it, i think there is an old developpement of learning gesture somewhere…
You should be able to do that with Wekinator. Recognizing an 8 figure, going from top to bottom : sounds like a Dynamic Time Warping thing.
If not willing to use Wekinator though you could use motzi’s machine learning pack. He also did two vveekend vvorkshops on the topic : part1, part2, at some point there’s an explanation on how to recognize a gesture, which sounds like what you’re after.
Hello sebescudie i have done a lot of test this morning. Ok, i need to check back so, maybe the simplest interface of wekinator didnt bring me to the right way to handle it. as i saw also an old post of you quiet enthousiastic with wekinator :) Thank you for leading motzi’s machine learning pack. thanks a lot !!!
Yes it helps, i didnt noticed on start this possibility to be in time warping.
but i dont have any result… cube detection is perfect, so i can output some logical from it ( smashing from up to down and so on), but having the figure of the two handed spinning wheel would be really a big plus. Some doubt as my data are inverted when spining the stick…
Here is a link to understand the spin i m searching to detect with the IMU, its the first one : https://www.youtube.com/watch?v=ZsuVokk6Zrk
The inversion of z and y maybe the problem when rotating …