Assisting iPi's tracking

Greenlaw · **Posted:** Tue Sep 01, 2009 2:00 pm

Hi,

I finally managed to get a block of time at home where I could focus on using iPi, and actually got some mocap out of the system. There's some weirdness in the tracking (as expected; it's my first real test,) but overall it's not bad either. It's good enough that I see a lot of promise in this system.

Anyway, I'm a little unclear on how to correct iPi's occasional mistracking. In one test, for example, the right elbow drifts off and points in the wrong direction, and in another case I had a similar thing happen with the hips, and in yet another the right knee rotates too much away from the body. After I correct the position in a given frame, what am I supposed to do next to get iPi to correct the frames before and after this frame?

Also, the rig's hips seem very narrow in relation to my actor (me) which may be the cause for the hips and knee to go out of whack. Is there a way to correct rig proportions for this?

Thanks in advance for any helpful info.

Greenlaw

Greenlaw · **Posted:** Tue Sep 01, 2009 2:44 pm

I just figured out why the elbow got mistracked. Somehow I switched off the T-Shirt over Long Sleeve option, so iPi got a confused about what I was wearing.

I'll have to check this later, but could it be that iPi is sticking with the default Long Sleeve option even after an actor saved with T-Shirt over Long Sleeve enabled is brought into the scene? If so, maybe iPi could recognize this option in an actor file in future release? (It may already be doing this; I'm just thinking out loud right now.)

Still a little unclear about how to correct rig poses when iPi does mistrack thought. (Thank goodness it's a little slow today at work.) :)

Greenlaw

Greenlaw · **Posted:** Tue Sep 01, 2009 2:51 pm

I'm working with a clip where the character places his hands on his hips. iPi seems to handle this okay until the hands are on the hips. At this point, the right hand goes into the center of the chest and stays there until the actor moves his hand away from the body again.

The hand is visible from the front camera but less visible from the side camera because of the high angle. (It's not visible at all from the third camera.)

Just wondering how I can correct iPi for the arm/hand position for this range of frames.

Thanks for any help.

Greenlaw

McWannabe · **Joined:** Sun Aug 02, 2009 10:57 am **Posts:** 107

Michael -

1. could you post the kinds of positions or movements where iPi is currently having problems so that we can try to avoid them?

2. Would a 4 camera setup improve any of these problem areas? I've got 4 QuickCam Pro 9000's - and should be able to start testing iPi this week.

3 What about strips of fluorescent masking tape in certain key areas of the body? If there are some simple things a user could do to make it easier for your software to determine where the body parts are, I'd be quite willing. Even OptiTrack uses markers and a suit.

4. possibly a very loose time frame for when the current problems might be addressed?

Greenlaw · **Posted:** Tue Sep 01, 2009 4:17 pm

Hi McWannabe,

How your tests are going?

As you've read, I'm finally getting some results, though it's a little mixed at the moment. After reading some old posts, I'm thinking that maybe I can improve my camera placement and calibration some. The room I'm using is just big enough, but maybe not quite big enough for the camera positions I established (12, 3, and 6 o'clock.) It's a long and narrow room, so I wonder if the arrangement should be a bit less than 180 degrees so I can move the cameras back farther from the actor.

What's your opinion?

Also, I'm thinking I need brighter pants for improved contrast with the background. My wife suggested bright colored tights like Danskins used in aerobics and yoga. I was thinking tight running pants, but after seeing what they cost, the Danskins are starting to sound good. (Whatever works I guess.) :)

Greenlaw

McWannabe · **Joined:** Sun Aug 02, 2009 10:57 am **Posts:** 107

Greenlaw - just unpacked my DELL - so I haven't started with iPi or JimmyRIG yet (I sure hope JimmyRIG Pro can read these bvh files!!!)

I figure anything that increases contrast should help - and maybe the way to go is going to be with some clothing accessories - which is why I threw the idea up.

The OptiFlex folks have probably been down this path before, and went with markers, etc. (but they cost a zillion $$$ more than iPi)
http://www.naturalpoint.com/optitrack/s ... mocap.html

So - I'm thinking I might look at their site for some tech notes on the best setup. I'd hate for the iPi developers to get bogged down on issues that could be corrected with some simple fluorescent masking tape markers that are cheap and easy to apply. But, I probably don't know what I'm talking about.

This info from Michael might help you re calibration:

------------------------
the sole purpose of Camera Calibration Tool is to measure camera FOV. This is a critical parameter for 2-camera configuration but not critical if you have 3 or more cameras.

With 3 or more cameras, you can just use "Auto adjust FOV" processing option in iPi Studio. In other thread, you mentioned you are using Logitech Quick Cam Pro 9000. These cameras should have diagonal field of view of about 74 degrees. You can skip Camera Calibration Tool and just use 74 degrees value for FOV.

Just for the record, head on short is not useful for camera calibration (due to specifics of camera geometry it gives no information). Only angled shots are useful. It is supposed that angle shots should be loaded into Camera Calibration Tool, each followed by a click to "Calibrate" button, and the tool will do FOV averaging.
---------------------------------
Cheers!

Greenlaw · **Posted:** Tue Sep 01, 2009 5:51 pm

McWannabe wrote:

Greenlaw - just unpacked my DELL - so I haven't started with iPi or JimmyRIG yet (I sure hope JimmyRIG Pro can read these bvh files!!!)!

I've been messing with it here at work as it hasn't been terribly busy today (for once!)

I loaded a iPi .bvh file into LightWave and was surprised that it included rotation bones, which can be used to correct or otherwise overide the deformation bones. Neat!

I studied the tutorial scene a bit more closely, and I'm going to try calibrating/capturing again tonight, this time with two lower cameras and one high. I was using two high cameras, and I think this may be obscuring where my hands are in some angles. Also, I think think without gloves, iPi is sometimes confusing my face for one or both my hands, or confusing one hand for the other when I'm making symmetrical moves.

Maybe I'll block some of the furniture and bookcases in the room with a large solid backdrop. This may may help simplify the track for iPi.

Greenlaw

McWannabe · **Joined:** Sun Aug 02, 2009 10:57 am **Posts:** 107

I know you're using 3 camera, but here's some info from Michael for both 3 and 4 cameras.

As to the backdrop, wouldn't iPi need some reference points like furniture or markers to be able to place your motion data in a 3d space? I think the 'how to place tracking markers' tutorial here might be useful if you're going to go with a backdrop?

http://www.hollywoodcamerawork.us/vfx_sampleclips.html

--- suggestions from Michael ----------
Ideal 4-camera setup would be 4 cameras installed in a circle at 10 feet high looking down to actor, they should all be pointed down at 20-30 degree angle, the CAN be facing each other.

It is important that you do not point opposing cameras directly into each other. They should look down so that there is some 20-30 degrees angle between their optical axes. Parallel optical axes would be a degenerate configuration (impossible to calibrate).

In practice, you can use 4 cameras in some kind of semi-circle setup just because you do not have proper background for 360 degrees in your room. This should also work OK.

I would recommend that you start your testing with 3-camera setup (central camera at 10 feet high, side cameras at 3 feet, 60-70 degrees angle between central and side cameras). 3 cameras is just easier to manage for novice users.

Greenlaw · **Posted:** Wed Sep 02, 2009 10:28 am

Thanks for the tips. You've given me something to think about.

I currently have two of the cameras at about 8 feet high. I saw elsewhere that the recommendation is 10 ft, but after reading your tips, it occurred to me that the angle of the camera is actually more important than the height, especially in a small space.

For example, if I positioned the cameras at 10 ft high in my current (tiny) space, the angle would be way too extreme to be useful. Even at the current 8 ft, my downward camera angle is probably around 40 degrees, which is much more than the recommended 20-30 degrees.

Regarding the backdrop: The reason I'm thinking of blocking the room background with a backdrop mainly has to do with our furniture, which is a pale blue-gray, and I just realized that the gray sweat pants I'm wearing may be confusing iPi when I'm in front of the furniture. Of course, it makes just as much sense to just change my pants. (Which I'm looking into because I think the sweat pants I'm using are probably still too loose for proper tracking anyway. Time to check out those Danskins) :p

If I can clear this with my boss, I think my next big test will be in the cafe area outside my office. (We occasionally use the space to shoot green screen and effects elements, so it's plenty big.)

Greenlaw

Greenlaw · **Posted:** Wed Sep 02, 2009 10:47 am

Hi,

Could somebody please list the proper steps for manually assisting iPi Studio's tracking? This feature doesn't seem to be documented, and I presently feel like I may be missing some important steps entirely, and possibly taking a few unnecessary ones.

I'm getting fairly decent tracks maybe 70 percent of the time, but when an arm or leg drifts way off track, I'm not sure how to give iPi the hints properly. I can reposition the rig, but then what? When I click track foward, it seems to overwrite my correction with the erroneous track.

Greenlaw

Assisting iPi's tracking

Who is online