|
...
In my experience 6 PS Eyes are not enough to accurately have enough constantly seen data to track dual performer recordings, most likely why you state you had so many issues and a lot of clean up.
As with any optical tracking system, (maybe iPi devs can validate if this is accurate with this system), but I believe there has to be triangulation of at least 3 cameras to always have a body part in view and this is virtually impossible with only 6 cameras and dual performers, especially if positioned at a commonly set semi-level view plane.
One main thing that should always be kept in consideration is the center floor mark and all cameras should point toward it at a chest high level, and look similar in all the recorders view ports, set as horizontally level as possible, and I believe this will attribute to better tracking results, although I do this by default, it is just something to be aware of. (Willy-Nilly camera views in the view ports could not help matters anyway).
I have not tried this personally with iPi yet, (although I do know a direct overhead camera will function correctly, as shown in the rock climbing tracking forum post and videos), and commonly in other optical systems there is at least 1, (usually more cameras), positioned high above the capture volume looking primarily down, if more used they are set at slight angles, but always pointing at the same central floor mark.
This positioning should always give the triangulation needed and better results for close proximity tracking, although any hidden part, say a tucked in arm, or arms under the legs holding someone may still cause tracking issues.
I don't have the ceiling height to place a directly overhead camera, and I use only 6 PS Eyes anyway, but I do have 2 positioned at 8 ft. and closer in to the center floor point, which has helped in hip tracking and foot angle placement on the floor for the most part, and also for full body floor contact.
I would expect a minimum of 8-9 cameras to better accurately track dual performers motions, especially if in close proximity for longer periods like holding to the body, hugs and the like and using a particular camera positioning scheme for better triangulation.
I don't even worry with dual actor tracking with iPi much with 6 cameras, but I have done some bar seating and dance motions that weren't in very close proximity using my camera positioning and it worked out ok without too many major errors, attributed to the 2 high side cameras able to look over the head of the performers closer to it and specific positioning of the performers in the scene helped.
Older Vicon systems I have had experience with worked the same way when using a smaller number of cameras, and also how I learned using overhead cameras helps, I just haven't been able to attempt it yet with iPi in an area that would give the proper head clearance for that camera, which is a min. of 12 ft., but optimally would be 16 ft. to allow for raised arm extension, without blocking out the cameras view too much, as the closer a body part gets to a camera, the larger that object appears in the lens view, so with dances this spacing becomes even more critical when directly under the camera(s).
This is why I personally think it is better to have more than 1 overhead camera and place them a bit lower (12 ft.) and a bit further apart (3 ft.) pointing at a slight angle to the center floor mark, and I would prefer those cameras if only 2, to be used as side view cameras, but if only using one, I would suspect directly overhead would work best at a higher height, but still using the 2-8 ft. high side cameras also. (most users would want to use the minimum amount of cameras needed as each one added will add to the processing time in tracking and I haven't really seen any posts of anyone that uses 16 cameras during tracking yet anyway, 10 should be plenty for 2-3 actor recording, using this style camera positioning).
One day I am fully going to test this set up with iPi in an adequate location, (maybe someone can do it before I can though), but I do believe iPi is like any other other optical system, it just tracks color instead of luminescent IR balls, which does give more viewable area for the tracking at pretty good accuracy once all filters are applied, but you can only expect so much at 60 fps with a low resolution camera directly out of the program without some clean up, but that is required with even high end optical systems, even if real-time tracking with less cameras, with iPi at a considerably lower cost for set up, as iPi was only developed as a Pro-sumer software, and does a pretty good job overall, for me anyway using only 6 PS Eyes.
I hope any of this is helpful to anyone attempting similar recording, and maybe iPi devs can shed a little more light into the mix that may help, or explain how triangulation is a factor with their system.
...
|