I have a couple of ideas:

Im not sure exactly how the tracking works or if it already does something like this, but would'nt it improve if you could as well as using different colours on different body parts to differentiate between them, also use other colours or bright spots to identify joints more precisely? A few daubs of reflective paint and a decent floodlight would make a hell of a difference unless im mistaken.

If it generated kind of key bone positions weighted by how much data supported the extrapolation of that bones location like a certainty factor, and biased tracking toward an interpolation between these bones positions more so if it is less certain where to place the bone. Bones that are manually corrected could also gain bias toward constraint to the relative degree of rotation between bones or movement between frames.
maybe something like this could add to the robustness, or is tracking completely image based?

A way to pin certain bones may be good.
Another thing that plagues me is certain strange sudden rotation of bones, couldnt they be constrained as a usual rig is? and if the constraint is approached it could reevaluate. Is this becasue the location of an extremity is conflicting with the rotation of its parent bones or something?

I know that most of the problems i have are probably mainly because of my cheap webcams/not optimally alligned as theres no room, and the environment is a little cluttered and even with my new flood light the cameras are bit dark/grainy. It would still be good if it can perform as best as possible with poor equipment though.

Well theres my two cents... for now.. :)

This is just my observation but iPi DMC may not be looking for the performer's joint positions as much as it calculates the volume of the body and then fits an IK enabled rig into that space. In other words, it tracks the 3D 'skin' of the performer which shapes the skeleton inside it...kind of the reverse of what our real bodies do.

As for differentiating left and right sides, this info may not be too important to the tracker; if it can clearly distinguish the performer's front and back sides, left and right becomes fairly obvious. (The human face and the shape of the feet might be enough to inform the program about orientation.)

I'm sure the software has some understanding of the natural bending limits of the human skeleton and tries to avoid doing anything too crazy with it. If it does do something crazy, that's probably because the video data is unclear (i.e., occlusion, poor lighting, confusing background, inappropriate clothing, etc.,) or the scene is mis-calibrated. I believe there are some rotational limits in place with the rig but if too many limits were put into the rig, it probably limits the type of motion the tracker can capture. (The human body can be surprisingly flexible.)

I know I'm oversimplifying what DMC actually does, but I think that's it in a few words. This is just a lot of guessing of course. To me, it's easier to believe it's all magic. :)

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

Well that's how i had assumed it worked which is why here in the feature requests section i suggested joint tracking. Maybe this could improve accuracy..
Im also aware that with better equipment i would get better results.
"I know that most of the problems i have are probably mainly because of my cheap webcams/not optimally alligned as theres no room, and the environment is a little cluttered and even with my new flood light the cameras are bit dark/grainy. It would still be good if it can perform as best as possible with poor equipment though."
Regardless of my capture quality though if those constraints WERE in place the rig wouldn't bent in quite such.. unhuman ways :/ Perhaps they should be optional so if you have good equipment or want to animate someone bending in inhuman ways you can. However a re-evaluation of tracking if a constraint was approached would seem to be a very usefull feature especially with not so good cameras, could save a lot of manual correcting. And i do believe this is a FEATURE REQUEST section.
Oh.. and it is all MAGIC regardless of whether you think you know how it works or not.. :D

Sorry, I didn't mean to come across as offensive. I was simply saying that what you're describing is a bit different from what I believe the current tracking system is designed to do, and it might require a major rewrite of system. iPi Studio is meant to be a markerless system, and once you start adding tape, balls, etc., to the joints of the body...well...

But I agree that it could be nice to have the option in a future version.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

Not at all, i didnt mean to imply that you were being offensive. :D Just seemed you didnt understand what i was saying/requesting..
Anyway, iPi uses the difference in colours/brightness to tell actor from background and different actor parts from eachother i think?
So what i mean is if you had just in key points, some clearly distinguishable small patch of other colour, using the current system, couldn't that be used to improve accuracy?
The only input data is video so all its using to find the light when your calibrating is the brightness difference in the picture no? couldnt that same principle be used with the only added requirement being a few spots of different colours on your clothes?
Its proving remarkably difficult to actually find the right clothes, ive been shopping a few times and theres nothing just right! if i could enhance my sub-optimal clothing with some reflective spots that would be awesome. :D
Well i got 3 new ps3 cams so its improved anyway..they are sooo much better..
Now i just need a new house.. with bigger rooms.. :/

Congrats! Yes, the PS3 cams made a huge improvement to my footage.

Tell me about it! :)

I spent many months getting rid of a decade of clutter from our garage just to clear a shoot space. I managed to clear most of the garage but actual active performance space wound up being about 6 x 6 x 6 feet. It's actually quite usable but a little tight for actions actions that require more than two or three steps of walking. I know some users have had good luck shooting outdoors under an overcast sky, so someday I may try that for 'larger' performances.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!