Spine still bends oddly

Greenlaw · **Posted:** Wed Feb 27, 2013 1:16 am

Okay, I will send something if the next tests fail. Thanks!

One thing I've never done before is to move the board forward and backwards in addition rotating. I will give that a shot.

G.

Greenlaw · **Posted:** Wed Feb 27, 2013 1:34 am

I seem to have gotten much better results this time after adding the front to back motion you suggested.

Calibration scene looks good but, according to the stats, I got about 500 good frames out of 700--that's seems lower than what I've gotten in the past. Not sure why this is happening.

Question: is it better to hold a rectangular board in tall orientation or wide? In the past I've held it wide but after reviewing the wiki, it seems like tall is being suggested. Or does this matter at all? The reason I ask is because on some errors I saw earlier today, Studio placed green markers in the air above the card, defining a tall rectangle, even though I was holding the card in wide orientation. In this case, it looked like Studio might have been expecting tall orientation. (Note: in this particular case, it is possible that I held the card too close to my body, making Studio think I was part of the card.)

Thanks in advance for any info. :)

G.

Greenlaw · **Posted:** Wed Feb 27, 2013 1:42 am

I just finished the tracking test using this data. Wow, what a difference that added motion made in calibration--the tracked motion results are spot-on this time, no clean up necessary.

I'll continue testing using this method in the morning (US time) and let you know if all is good here. Thanks again for your help. :)

G.

Greenlaw · **Posted:** Wed Feb 27, 2013 1:53 am

Getting back on topic...thanks for bringing back 'Stiff Lower Spine' in this build. When Flexible Lower Spine is chosen in my current test, I noticed that when I am jumping forward, the spine will curve sideways as I come down--but when I re-tracked this region using Stiff Lower Spine, the spine looks much better.

Also, arms tracking is significantly better now. :)

G.

Greenlaw · **Posted:** Wed Feb 27, 2013 11:49 am

Okay, a new day, a new session.

Adding greater variation in the calibration board's movement is definitely helping--a lot! My first calibration of the day got perfect matches for 433 out of 457 frames. Nice! :)

I think I have a much better understanding of how this works now. In the past I only rotated the board from side-to-side, which probably does not give the new Auto Detection feature enough information to distinguish between 90/180 configurations. So, instead of the regular side-to-side rotation, I now move the board in a more random 3D motion--back and forth, slight tipping and banking, plus side-to-side rotation. About 500 frames of this action is plenty of data.

Of course, you should be careful not to move the board too quickly--remember the Kinect records only at 30fps.

If feel like I may actually get some actual shots recorded and tracked today. :)

G.

Greenlaw · **Posted:** Wed Feb 27, 2013 12:07 pm

For some reason, background subtraction has become a little 'hit and miss' since yesterday. Sometimes I get very clean backgrounds, and other times it's very messy. (Image from yesterday is a typical example.)

I'm not sure what I'm doing differently when this happens. The problem doesn't appear to affect tracking but it does clutter the screen with unneeded data. I'll try a few more tests for this to see if I can figure out what's messing it up.

Hmm...could the fact that it is still morning here have something to do with this issue? While the items in the room have not moved at all, I'm sure the lighting in the room has changed significantly during the last few hours as the sun travels overhead. Thanks for any info or opinion.

G.

Greenlaw · **Posted:** Wed Feb 27, 2013 12:16 pm

Yes, that must be it--room lighting that comes from outdoors can affect the quality of subtraction over time. I just ran Background Evaluation and recorded a new motion, and this time the background subtraction looked very clean.

If your room lighting situation is similar to mine, it's probably best to run Background Evaluation often.

BTW, when I'm all done here, I'm going to compile these tips in a handy guide for everybody. :)

G.

Greenlaw · **Posted:** Wed Feb 27, 2013 3:01 pm

The tracked results from this morning's test look wonderful! The Motion Controllers worked great too. New copy and paste options and Refine have come in very useful for fixing any occlusion issues, especially because part of the test involves a circle walk. I wasn't sure if dual Kinect at 90 degrees could handle this motion but the walk looks surprisingly good, even when an arm or leg gets 'hidden' from both devices.

Right now I'm trying the new finger posing system. Not doing anything fancy with it, just 'relaxing' the fingers a bit. So far, it's working well.

I should have the test mocap applied to characters this evening and will post a video shortly afterward. When I get to that point, I'll carry this discussion over to a new thread about how iPi Mocap Studio is being used for Brudders 2. I'll make the announcement here for anybody interested in this. :)

G.

Greenlaw · **Posted:** Thu Feb 28, 2013 11:26 am

The finger animation system is actually very useful! I'll comment further in another thread but it's very easy to use and I can see where, if you modified you controllers to free your fingers, keyframing in iPi Mocap Studio can productive because the software provides immediate reference video. If anything, you get a great 'first pass' for finger animation that you can tweak later in an animation program.

BTW, since I started using random movements for calibration, the system has been very accurate. Thanks again for the tip.

The test motion looks fun on the cats but I want to re-record the test for 'Sister' before posting anything. I was trying to walk like a child but Alisa tells me it looks more like an old man, and she's right. So this morning I studied child/toddler motions and noticed that they have 'sticky' feet and they make very little arm and torso movements. It's almost a 'Frankenstein' walk but with the arms wide to the sides and slightly more knee bend. It's going to look really weird when I try emulating that but it should be funny and hopefully somewhat more believable when applied to the character.

Anyway, I think this thread has served it's purpose so, for now anyway, I'm calling this issue closed.

My next post will be in a new thread--see you there. :)

G.

vmaslov · **Posted:** Thu Feb 28, 2013 11:41 pm

Greenlaw wrote:

Anyway, I think this thread has served it's purpose so, for now anyway, I'm calling this issue closed.

OK, topic is locked. 100 posts qualify it for a new record :)

Spine still bends oddly

Who is online