Hello Everyone,

New user here, have been using iPi for a couple weeks now with good success using two kinects, and now moving on using multiple PS eye cameras so we can get further recording distance. Following the iPi guide for recording with multiple Eye cameras and a couple of recording tests, had good success with getting good calibration data using the maglight and all that stuff.

The main issue we are having now is, once we set the ground plane points, the base gird uniformly tilts on x rotation and throws all the camera's off due to odd rotation. The workflow I used was

- record with 5 PS eye cameras in darkening mode, good balance between light and readingmaglight
- walk around with light and touch all corners, cris-cross to each corner and then swirl down in middle
- open mocap studio, go to calibration mode, go to each camera and roughly setup the layout of cameras
-made sure each camera was at correct height in meters, then let it calibrate.
-before marking ground points, the data looks ok...but is above the ground plane a bit

Not sure where I'm making the mistake, on a couple of try's i get the "Auto Detect Camera Initial position" , it worked ok but after the detecting of the light points when it was moving the camera's, it would throw them off completely and some times rotate them out of whack.

Sorry for long post but really want to get the PS eye system up and running since the kinect is limited by having to be so close to it. Thanks for any help, specs are listed below, not sure if that'll help.


3nd Gen Core i7 Quad Core
16 gigs Ram
GeForce GT 640
Trail version installed (waiting to buy pro if we can get this working)


P.S. During recording the FPS stayed at 60 for all the cameras.

Hi, chrisNFSN

Can you share with one of your calibration videos to look in more detail?
FYI, you can not export the animation from 5 cams with the free version - 4 maximum. But still you can do the tracking itself for up to 8 cams.

hi..thanks for the reply. Yeah regarding camera number, we're still in the testing phase, I know once it looks ok the mocap studio, bones wise, it'll be ok when i do export it out.

Here is the link to the video https://docs.google.com/file/d/0B7os5dBy_KLqdG9rZTh2WGhJTmc/edit
Somethings i was going to try next, was 4 cams in a semi-circle to try and avoid me getting in the way of the light while moving it around. Also, after digging the forums a bit I came across a DOC file someone wrote, for his calibration he put the light on the floor and moved it while it was touching the ground for a while and set some points based off of that instead of touching the corners. Again....thank you for your help.

-chris

Hi, chrisNFSN

I've worked with your video, and got calibration results that look similar to real positions, but not good enough for processing action videos. Attached project file - you just need to apply proper scaling knowing camera heights.
I have some tips on your calibration technique below. Following them should improve calibration results.


When marking the floor by touching it you do not need to follow any specific geometric pattern. The idea is that the software just has several (3-5) reliable (viewed from at least 2-3 cameras) relatively distant non-collinear points on the ground, so the correct ground plane can be computed. Corners are just good candidates for such points. In your case, for example, corners near cameras 1 and 3 are not very good for the purpose because each of them is only viewed by 1 camera (4 or 5). For the same reason, these corners should also be avoided while shooting the action. I recommend to add 2-3 ground points at the central part of the area to make ground plane more accurate.

There is a huge amount of "red" points in this video. This is mainly because in many situations your were completely obscuring marker from many cameras with your body. You should consider at all times that marker should be visible to the most possible number of cameras. For your room configuration, the ideal position may be like you are holding mag-lighter in stretched arm, and you move in such way that your body is always between the lighter and the nearest long wall. So in most situations the marker will be good visible to at least 4 cameras.

Also I have noted your marker positions mostly were concentrated at the middle-height level. To make calibration more robust, you should try to cover full capture volume when moving marker. Walk reasonably slow through your area and constantly change marker's height in the range of body height.

Here's a calibration demo I did quite some time ago:

http://www.youtube.com/watch?v=MlG1_65s2_c&feature=plcp

And here are some tips:

As vmaslov points out, the actual geometric shape isn't important to Studio but I like drawing a cube because it's a distinctly recognizable shape--if Studio generates a shape other than a cube then I know the calibration has failed.

To calculate the ground plane accurately, you need a minimum of three points on the ground for Studio understand where to place it, and placing more points on the ground should average out to a more accurate plane. In the posted example, I have five points defined for the ground--the four corner points and a center point at the end of the spiral pattern inside the cube. If you use only a few points like I do, make sure they are visible to all cameras. But keep in mind that creating too many points on the ground may extend calibration time and probably won't result in an appreciably more accurate ground plane.

The spiral pattern I draw inside the cube is another recognizable shape and it covers a broad area which I believe helps improve the accuracy the camera positions. Again, creating shapes that are recognizable is really more for my own benefit but I also believe it's good practice to have a pattern you can repeat for each new session so you can get predictable results.

When drawing the calibration shape, don't move your Maglite too quickly because motion blur can make the points harder to track.

If you have trouble visualizing the shape you wish to draw in space, try placing bits of tape on the ground to mark the boundaries and corners of the shape. At the very least, it's a good idea to mark the boundaries of your capture space--this should be the region that is completely visible to all of the cameras. When marking this boundary, it's good to have a person standing in the space because the actual boundary may be smaller when you account for the height of the performer.

I hope this info is helpful.

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

Watch a one minute excerpt on Vimeo now!

Hi Green,

Thanks for the reply. I got to do some more testing this weekend. I reduced my 6 cameras to three in order to hone in on calibration techniques and general workflow. For my first batch of tests went much much better. Got plenty of green dots, and the ground plane points went well and got everything lined up the grid floor properly, but the main issue I seem to be having now is a scaling issue of the camera height in relation to the video playing and ground of the grid, I made sure the heights of the cameras were as close as possible and the scale of the setting of Camera 1 to 2. I used the ground plane adjustment but seemed like I couldnt get it far enough down. Once it seemed like I got a good calibration I try a simple action test, it went ok, the legs were completely not calculated which may have been cause I was wearing white shoes with small stripes on there.

I attached a link to the files if anyone wanted to take a look, they are the files that start is "3_PS" . As soon as I can I'm going to do some more tests maybe 4 cams this time. Thanks again for all the help guys.

https://docs.google.com/folder/d/0B7os5 ... FsLU0/edit


-chris