Projection|Dance Choreography|Kinect

this is my newest project …merging dance and projection

http://www.youtube.com/watch?v=ISKV1BeB3pM

project page:
http://princemio.net/portfolio/flow-1-kinect-projector-dance/

i utilized

ofxOpenNI – created by gameoverhack – openNi wrapper to read captrued data from the kinect camera in realtime.
ofxCV – created by Kyle McDonald – fast openCV wrapper.
ofxFluidSolver – created by Memo Atken. After Years, it is still one of my favourite calculation models to illustrate continous flow of a dancer and graphics.
ofxUI – created by rezaali – having worked a lot in processing, i completly fell in love with this GUI library as it speeds up my tweaking processes. Its easy to use and fast to bind to variables.

many thx for this nice libraries!!!

for calibration i used the methods by kimchiAndChips - it was a huge step for me forward in learning about projector - Camera Calibration
http://www.kimchiandchips.com/blog/?p=725

hit me up if you have questions or need help or advice…thank you very much

mio

Great job!

I especially liked the hectic triangle part. Its nice to see how such simple shapes in just black & white really messes with the perspective and the way we perceive the dancers.

Really nice work, congrats. My favorite is the part with the triangles, and the grass-like things growing towards the motion.

I assume you used optical flow, but from the video it shows that you didn’t have much light. Did you use the rgb camera for the optical flow or another infrared camera? Or did some tricks with the kinect depth image?

hey guys thank you very much for the kind words. I used optical flow for the calculation. right now im more interested in this flow then in the actual shape of the body.

So i used farneback optical flow to create a grid of vectors. ofxCv by kyle has a nice wrapper and examples for doing so.
Those vectors are then used to maipulate the individual graphics. The mask is created by a kinect depth image + background subtraction. I additionally scaled down the image to 320 * 240 since processing on images can be intense and especially farneback is sometimes slow.

In order to map those flow vectors to the dancers body…i used the calibration method by kimchi and chips

again thank you very much

mio

Thanks for the explanation. I assumed the kinect depth image does not have enough information to calculate proper optical flow. Probably the background subtraction helps. Thanks! I’m looking forward to your future projects combining dance and projection.

Really nice work. Congratulations

Very nice project