Kinect poses

Hi all

I am about to make an installation where I would like to use specific poses to select the state of the installation.

I have been looking around and can not really find a method. is there a method for recognizing poses via kinect?


Hi Sunep,
i’m trying to do the same, and at this point i’ve readapted this contribution ((contribution:p-point-cloud-recognizer
Unfortunately, for now i can only use 2d coordinates, so i decompose the 3dvector of joints and recompose it using only x and y coordinates (so it’s a kind of 2dpose recognition).

if you think it could be interesting i can try to clean the patch and make a simple version only with this thing, and i can post it.
and i have the sensation that converting this plugin from 2d to 3d could not be so difficult, but i’m not a programmer, so who knows… :)
if you have other ideas, or if you develop further my first idea, i will be happy if you will post your advances ;)
a simple example patch will come soon

ooh, that sounds interesting. posting it would be very cool.

i’m trying to make it simple, because my patching style is a total caos ahahahah :) stay tuned ;)

it’s gonna be bitchy…
you have to care about depth also…
would be cool to have something that can tell you if point is inside of some shape maybe… otherwise =, and (spectaral) then your epsilon is radius of hit sphere, but even if you take XY without depth you will have troubles with size after… I’d rather try to go from skeleton, and i would prolly project it in 2d first… 2d points and something like image or countour to test hit

I have been thinking about angles in joints. then the pose could be a list of the angles with appropriate epsilons

et voilà, a simplified version (with a quite comprensible root, i hope :)
let me know if it works and if it’s useful, and hope that someone want to try to convert this very useful plugin in 3d space ;) (48.2 kB)

yes, i had the same idea about angles, because the angles has also the advantage to be absolute, and not relative to the dimension or position of the skeleton, anyway in my little experience this patch i’ve attached is quite working for different people too (now i’ve tried only really simple poses, like arms up, arms down and something like “walk like an egiptian” style ;)

a wait for calm is missing

mmm, just a moment ;)

ok, try now (58.2 kB)

mmm, there was some errors related to the waitforcalm (it lost some settings, now it works, at least here) (58.4 kB)

wow, looks like it’s working… I will try taking the p$ route. 2D is good enough for this project

Great, i Will update if i discover a way yo use 3d positions

The recognizer is only for 2D. Don’t think you can rewrite the algorithm to use 3D - based on the math used. But of course it can use 3D->2D projected coordinates.

As for detecting Static Poses with the Kinect using joint angles should be more robust. Unless you are going for dynamic Gestures (Gestures with movement). There is also some Gesture library implementation on mrvux’s Git which you could try out. Think it was handed out at node13.