Hi folks:

I managed to swap the order of the cameras in my video, and when I load it into DMC, it seems to be in the right orientation and the defalt camera positions only need a little fudging in the scene tab to be close to the coordinates that I measured at the shoot location. It seems to track the calibration light ok, but there are lots of false marker id's on other bright spots in the scene (like my pasty white hands and bright spots on the floor).

I have four cameras at about 2.9 m height in the 4 corners of the room, and two cameras on tripods at 1.3m high. After I run the calibration, one of the cameras goes to 5+ meters, and one goes below the floor. I'm guessing that they use the Bouguet-Svboda calibration technique, and it is unstable when there's lots of garbage data.

(see http://www.vision.caltech.edu/bouguetj/calib_doc/ and http://cmp.felk.cvut.cz/~svoboda/SelfCal/ )

So, how do I get a good calibration? My mocap room is 7 m by 8 m with a 3 m ceiling. It has fluorescent lighting, and I will see if I can shut half of the lights off for the calibration. I'm also getting a white LED light that should be brighter and better than the maglight and blue led lights I've already tried. Also, since the capture area is so large, I walked around the outside of the area while moving the calibration light in vertical up and down motions. The tutorial says use vertical corkscrew patterns, is that

I'm assuming there's no way in DMS to clean up raw data - is that true? If that's the case, I'm going to try to recalibrate using Svboda's matlab programs. The newer versions of matlab have a wonderful graphic data cleanup tool where you show the 2D point from each camera, draw a selection box around bad data, and then it tags that data so you can remove those points with a simple matlab command.

Anyway, please let me know if I'm doing something else wrong. I rent the space, so I'm not at my leisure to just start trying things, I need a good plan to deal with the issues I've discovered.

Thanks, Michael

It should be pretty straightforward. Normally I just run the calibration without manually adjusting the cameras at all, and I always get the 'good' green marks. The only thing I adjust is the height of one of the cameras. I'll upload a typical four camera calibration video tonight so you can see what it should look like, along with images of the final result. I'll post the link tonight.

BTW, did you remember to switch to calibration mode? This gains down the lighting so that only the brightest element (the maglite in 'candle' mode) is visible. If there are elements that are as bright or brighter than the maglite, then you will have poor calibration results. Things to avoid are reflective surfaces (like window, mirrors, shiny floors, bright lights in view of the camera.) Also, make sure you've defined on the timeline the frames you want to use as your clean pass. If you haven't done this, calibration and tracking will probably fail. In theory, the clean pass at the head of your video is used as a difference matte and it will eliminate any of these things, but naturally it doesn't eliminate moving elements like the maglite reflected on a refrigerator door in the background or on a pair of glasses. (The silliest error I've run into is a computer screen in the background with a basic screen saver running. That stupid Windows logo that randomly pops around on the screen totally confused the tracker.)

Of course, don't forget to switch calibration mode off after you've captured your calibration footage. This will hose your motion data.

Hope this helps.

G.

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

Watch a one minute excerpt on Vimeo now!

Thanks - I'll try to run it without moving the cameras at all (except maybe camera1). Do yo turn on the autodetect initial positions?

Michael

Greenlaw beat me to it this time, I was going to say to make sure you're using "calibration mode". That was one of my early mistakes.

I'd love to see a bit of your calibration, i'm sure we'd be able to give you some really good pointers.

Yes. I let the software do this for me and with my typical setup it always works fine; no manual adjustments needed.

G.

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

Watch a one minute excerpt on Vimeo now!

OK, I just wrote a program in matlab to track the LED in the video. When I set the brightness threshold to points>80%, I get almost no mistracks. In other words, the video is good, the DMC software is just not working well to track it. I did have the calibration box checked, and it did dim the exposure slightly, but the room was pretty bright, and everything was still very visible. That was much different from shooting in my living room where the light levels were much lower, and the exposures almost dimmed to black except for the hotspots from the lights. I think my problem is partly that I know too much about how mocap image processing works, and maybe I'm making assumptions and mistakes that someone who doesn't know enough to be dangerous wouldn't make. I'll see if I can dim the lights in the room during the calibration. Maybe for now, I'll try a calibration in my living room with the dimmer lights and see if DMC can track the LED better.

I also left the computer at home working on a calibration with none of the cameras adjusted in the scene tab to see if that helps.

Michael

Yea, the bad points are showing up as green too depending on how it sets the cameras automagically. I think there are just too many light hotspots on the floor (and my pasty white skin) that are being tracked as valid points. When I check the brightness of the hotspots they come in at about 60%, but the LED is always > 80%. This should be enough of a difference that a thresholding slider could differentiate them. Most infrared mocap systems have a way to adjust the threshold for tracking.

More as I experiment, Michael

It doesn't look like I can post images, and I've got a really cool one of the LED tracked in all the cameras over 1137 frames.

Oh well,

Michael

I use imageshack to post images to this forum.

BTW, I will have that calibration video uploaded soon.

G.

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

Watch a one minute excerpt on Vimeo now!

I posted information about our Calibration video in the following thread:

Little Green Dog's Calibration Video

Feel free ask questions in that thread.

G.

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

Watch a one minute excerpt on Vimeo now!