UPDATE 05.19.2014

Progresses :

Kinect skeleton is very jumpy, because of the noise of the sensor.

So here is a script which SHOULD smooth the value using 

"Holt’s Double Exponential Smoothing Algorithm" 

which uses two factors, a and b...

a should be 0<a<1 -> a is the smoothing factor

b should be 0<b<1 -> b is the smoothing trend factor

I put 0.25 for both...

 

References :

[[OpenNI.holtsDoubleExponentialSmoothing.Alessandruino.py]]

GroG

10 years 7 months ago

Nice work Ale !

Do you notice a large difference with the arm itself?  I always assumed we would need to do PID or some other smoothing algorithm - but wkinne started his experiments I was pleasantly surprised to see how smooth Kirk moved.  

I know some smoothing will occur just from mechanical load and details of the servo.  Although I am a big fan of solving issues at the source - is "jumpy" an issue?

Regardless, its will be great to add to our toolbox !

Without smoothing i got crazy values( negative values too) which were no acceptable for ik service...i got to remap kinect to "real" distance too... Angle itself is more smooth than the single point.. Print hand.z as test and see how much jumpy the value is..