Mocap Test with Small Child (1 meter tall)

Greenlaw · **Posted:** Mon Apr 02, 2012 11:05 am

Yesterday, right before dinner time, my five year old daughter Sienna Kai and I decided to do a quick motion capture test in our living room.

Mocap Test with Small Child (1 meter tall)

We used iPi DMC 1.0 with twin Kinects positioned at 90 degrees. For raw mocap, the results are really quite good considering the chaotic nature of her 'Happy Cat' dance.

It's my opinion that the 90 degrees configuration produces more stable results than the 180 degrees configuration. The 180 test we did last weekend was more jittery and had a lot tracking errors, and last night's 90 degree test was more steady with far fewer errors. Also, with the full 90 degree positioning, occlusion is clearly not much of a problem.

Both Sienna Kai and I were very pleased with this week's test.

Set phasers on 'cute' and enjoy. :)

G.

Michael Nikonov · **Joined:** Tue Mar 31, 2009 9:13 pm **Posts:** 805

Thanks for sharing. That's a very interesting test case! And nice dance, by the way!

Greenlaw · **Posted:** Sat Apr 07, 2012 3:40 pm

Thanks Michael! I'll zip the files for this 90 degrees test and the files from the 180 degrees test from two weeks ago, and send it to you guys for study.

There's a error in this video that's interesting. Twice in the the motion, Sienna's left knee points inward when it shouldn't, and I think it's because she leans too close to the sofa. During 'Happy Box' I ran into a similar problem when I sat too close to the smaller love seat on the other side of the room. I think what happens is that the kinect 'merges' the performer with background objects if he gets too close, and this confuses the tracker. I haven't tried to fix this yet but I think it will be another interesting challenge for the software.

There is also an error at the very end that I just somehow missed. It only took me a couple of minutes to fix this but I'm not exactly sure why the tracker gets confused here. My guess is that the data lacks enough detail for accuracy because the left arm is visible to only one Kinect from this angle. Plus, her arm is really skinny. :)

G.

Michael Nikonov · **Joined:** Tue Mar 31, 2009 9:13 pm **Posts:** 805

Greenlaw wrote:

I think what's happens is that the kinect 'merges' the performer with background objects if he gets too close, and this confuses the tracker.

Yes. Recommended safe distance from actor to background is a minimum of 50 cm.

Mocap Test with Small Child (1 meter tall)

Who is online