...

I don't know if this topic has been discussed before, I used the search function to try and find it, but didn't return any results:

I decided to run some tracking with the ground plane collision box unchecked. (using the 6-PS Eye set up), I am not sure of the effect using Kinects, maybe someone else could test it for those.

Although I get good tracking with ground collision on, sometimes the hips and lower spine would react weirdly during most squatting and extreme forward bend actions.

(Note: I like to always use the very flex spine setting when I can, sometimes I just can't to get the best results, so I switch just in these areas to just flex spine to help it, or I can correct things in Post-process editing, but would rather have it work in iPi using very flex throughout if I can, but I don't know it's possible).

I feel I have a good calibration for the cams off floor height generally, and I have tested just lowering the floor in scene after, but didn't feel that was the best option after seeing the results, I never had to adjust it more than 1 cm up/down to keep to video reference, or actual performers height kept within that scale.

So, I refit the Actor with the ground collision box ON, then I turned it OFF and hit refit again and the Actor dropped just slightly, but I left the Actor height the same for tracking.

After tracking, I noticed the hips, lower spine, (that before caused poor belly push) and feet contact appeared to track better, I actually got a better bend in the lower spine, which that would before lock in the hitched position on extreme forward bends and the hips would be forced to rotate forward to compensate, as well as the heels try to lift off the floor a bit more, even the toes sometimes, depending on the action, or reach of the leg.

I am going to re-track some videos where I know I had this happen more severely to see if it helps, I just did 2 so far, but I feel it helped in iPi Studio 3, as well as when exporting to external editor, it seemed to be more stable, I always have exported with AMT on, as it stabilizes the feet better. (I have tried exporting with it OFF also, ON I feel I get better results).

Just adjusting the floor lower in-scene didn't show the same results, it still seemed the hips were locked to the refit of a static hip location, even if the Actor was made to refit accurately, with ground contact OFF, it was able to follow the performers hips more accurately without an awkward push effect.

Ground contact was also not needed for the toes to bend properly, and since I now get good Move controller data and always use them for wrist tracking, I don't need the floor contact for hands either.

You would still need a good camera calibration first, with the correct cam 1 off-floor height set in order for your Actor scale to be very close to actual.

I am going to test this more, I just thought I would mention it if other users are seeing these same affects after tracking and it may help them too.

I will post some videos of the results before and after soon, if I am satisfied this does help, especially after export and testing in other 3D editors.

...

...

After testing this method more thoroughly, in my experience with my set up, I am finding that using my actual calibration height of cam 1, ground collision enabled works better, over disabling the ground collision, or using the floor height iPi seems to want to set it to by checking it with disabled ground collision, (iPi shows it to be 2cm lower), and also re-running the calibration video and setting cam 1 to that shown iPi floor height with adjusted Actor scaling, which was taller than the actual performer height to appear correctly in scene to the video after refit.

Using my actual cam 1 height during calibration, over just using the floor height adjustment in scene I also feel works better, the 2cm difference iPi showed in scene to me personally makes a difference in the outcome of the tracking. (the closer the actual performers height and bone sizing/position in the Actor mesh, setting the calibration correctly to the video reference is more vital IMO, but I have also used several calibration cam heights and if everything is adjusted to work together at a different than actual performer scaling has worked pretty good also, so there is a bit of leeway the program allows when correct adjustments are set).

Although the ground collision isn't completely needed for toe bend reaction, it does follow more accurately to the slight bending in the video when the floor collision is enabled and the floor is set correctly, slight bending in the video won't be picked up when the floor is too low, or ground collision is disabled, I just like more naturalness in the toes when it's more accurate to the video. (some programs don't even allow toe tracking, so it's a plus to have it in iPi).

Other users may feel 1-2cm is neither here nor there to their acceptable tracking quality, (of course anything can be worked out in post processing), I am only mentioning my results and how I feel it performed, but I do feel if the Actor to video to calibration to floor height isn't pretty good, the results will not be as good as they could be, far more forward tracking issues occurring and most likely have more manual cleaning in a 3D editor, which I have tried to limit by allowing iPi to perform better with less tracking errors during the tracking process, even though catching some tracking errors while tracking and fixing at occurrence with some post-process clean up of some areas is always expected, but really far less than other programs, especially for the amount of extreme action there is in my recordings.

I also found by switching the performers shirt to be a more form fitting style in recent recordings, where the torso area doesn't flair out from the torso, even when more loose T-shirt was used and tucked in, and I also added a 6th camera squarely head-on to the performer, from using only 5, with 2 set crossing the head-on view of the performer, (as shown in the iPi Docs for multi PS Eyes set up), for the different set ups videos tested, this 6th camera positioning has seemed to help, but I haven't tested it with the more loose fitting T-shirt yet to compare and probably won't since I have the form fitting wear now.

While I was doing this, I also tested the auto-cleaning only using iPis' ACP (JR-TF-AMT) compared to using none of iPis' ACP and using only Web Animates' ACP and settings, and also using both processes together, all done with FBX format exports from iPi.

I am videoing the results and will post a link when completed in a side-by-side-by-side tri-view to show the results I received on single forward tracked videos, single run post-process cleaning without any manual post-process clean up corrections.

...

...

Well, it seems now iPi has updated the floor and bone adjustments to use the hips as the static point for sizing again, I am testing the effects this has on tracking and will get back to this topic later.

...