I use 3 Logitech Quick & Webcam Pro 9000.

I have done some tracking but as usual I have some problems in tracking the arms.

View my original video:
(feel free to view and download)
http://www.4shared.com/video/FUifMFN0/video_2010-07-19_18-42-31.html

View the tracking video:
(don't mind the red sleeves it was not there during the tracking, its might just be a bug)
http://www.4shared.com/video/Aw-rNO6e/track.html


1.) I used sunlight from my windows and a small double light florescent lamp. Is the light good enough on this recording? I was wondering what is the best way to use lighting in order to help in tracking?

2.) When I press Analyze Actor Appearance while viewing on either Cam 1, 2 and 3, the actor's clothes color is different in each camera. Is that normal?

3.) Since I am using Logitech Cameras, should I record in MPEG or RAW video codec?

TheOwl, have you absolutely free background (without actor and even without separate body parts) in each camera during first several seconds? Based on posted original video you have not. It is extremely important for the system to calculate correct background.

Please, upload file with scene parameters (*.scene.xml). We'll try to process the video and answer your questions (especially about red sleeves phenomenon).

Note, you can post the content of the scene file right in post as text like this:

Apart from lacking a 'clean plate' as Andrew noted, there are a lot of black shadows in the background and the black sleeves can 'disappear' into the darkness. This can make it more difficult for the tracker to isolate your arms, even with the clean plate. You would be better off with a brightly colored long-sleeved shirt and a dark short-sleeved shirt over it.

Also, you may want to take time to match the colors in the individual cameras too. (You can do this in the Logitech control panel.) If I understand how Analyze Actor works, iPi DMC may be looking for determined color cues to help isolate body parts (like sleeve color for arms for example,) and when the color is different in every camera, that may add confusion.

When you prepare a session, do as much as you can to make it easier for the tracker to see what's going on (shoot the clean plate, remove or hide unnecessary background clutter, avoid reflective surfaces, make sure there's reasonable visual separation between the performer and environment, etc.) Believe me, the extra effort is well worth it when it's time to track the data. Few things are more frustrating than to spend a lot of time shooting a session only to have it fail during calibration or tracking because of something that could have been easily prevented or corrected when you the shot the video.

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!

the red sleeves come out after saving the project and returning to it later.

viewtopic.php?f=5&t=3575

My Scene File:

Thank you for calibration parameters! We process video. The root of the problem is not in background (fortunately the area where you perform movements very slightly intersects with broken areas of background).

There are three tips which should help to obtain better results in the future session:

1. Greenlaw is absolutely right in all observations but the following one is extremely important in this case:

The problem is in sleeves which color is very different in #2 and #3 views.

2. It is not recommended to set one camera exactly in front of the other in case of 3 cameras arrangement. Instead you should choose 120-150 angle between optical axis of #1 and #3. For more info see http://www.slideshare.net/iPiSoft/scene-setup-guide. It helps the system to see actor arms in each view.

3. Minimize motion blur: either use more light, or move slowly (your hands), or buy PS Eye cams and record at 60 fps.

Here is the result of our processing: https://docs.google.com/leaf?id=0B5I10yTpqXLuNDQ5NGRjNGQtNjQyZS00NTI0LWI3MjktMWMxM2IyZGUyZjk0&hl=en&authkey=CPXy1owO

How it was obtained:

• First of all, we set light parameters: position and ambient (Scene tab). It helps to overcome problems with left hand (its different color).
• We have paid additional attention to actor appearance by repeating "analyze actor appearance -> refit pose -> analyze actor appearance -> ..." many times
• We produce series of experiments (in range from 1.72 to 1.80) to choose actor height. It seems, 1.75 is optimal
• Due to low contrast between right hand (sleeve) and background was regions with incorrect recognition of right hand. To correct them we have used "Track Backward" button: go to the end frame of the problem region, manually correct actor position if needed, press "analyze actor appearance" to refine colors for concrete circumstances, then press "Track Backward". If something goes wrong again, repeat (note, there is the random seed in recognition algorithm...).

Note, that there is some problems in the result:

• From frame 640 to 700: you need to use more cameras to capture such movements (in these frames hands are not enough visible at least 2 cameras)
• Frames near 883: wrong position of right hand (too low contrast!)

P.S. We didn't apply "Jitter removal" to the result animation. Do it yourself if you want to try this animation in action.