Well; I have run through a number of trials and tests using OpenCV and a webcam looking at my own implementation of the sixthsense style of interaction. While the ways of implementing it are there, I sat down with a co-worker one day and realised, why not buy a Kinect? It has less of the over the shoulder portability of a sixthsense system but more of a presentation/presence potential and MS is about to release 1.5 of the SDK.
The ideas I've come up with are interesting, especially since Kinect for Windows is just a bunch of smarts in a box with a camera, IR sensor and IR laser emitter.
Milestone 1 became:
"While the skeleton points are provided easily, I need to locate the object at the point"
After a nice side track off onto depth camera smoothing (a limited implementation of this project which I kind of deduced myself but it's a nice link regardless) and change only depth frame display, I ended up with a quick and dirty depth and skeleton point based bounding box.
Walking points up, left and right using a quick guess at the range of depths to include (to allow for precision problems, angles etc) it gave me a nice framed area for further depth and image mapping.
A happy coincidence was that, when tracking hands, the open palm has a larger area automatically than a closed or half closed hand due to those parts falling outside the allowed +- Z axis. (Here I wrote "Handy!" then slapped myself for the bad pun).
So, Milestone 1 is completed. It's not great but I can make it better and it's required for two of the ideas I'm going to attempt to implement.
Next up, Milestone 2, depth map storing and comparison along with some image recognition.
I'll probably throw my "fun with Lync 2010" things on here as well since that's been driving me up the wall recently.
No comments:
Post a Comment