Hello Everyone,

I cant seem to be able to get calibration to happen. It really seems like its blind luck that it will calibrate. Out of maybe 40 times I have tried to get calibration to happen 2 have been PERFECT, maybe 3 have been GOOD.. and the rest ofcourse FAILED. I really want this to work as I am excited to work mocap into our flow.. but so far its been a fairly tedious experience unfortunately.

Anyhow, I have some pictures below in hopes that someone can shed some light on this process. Details below...

- 2 Xbox 360 Kinect cameras
- One laptop, running dual GTX 560m graphics cards (its a big laptop)
- I suppose thats the important machine info...

- Cameras are 10.5 feet (3.2 meters) away from central point
- Cameras are 10.5 feet (3.2 meters) away from each other (65 - 70 degree angle)
- The Floor is in view
- I generally calibrate the whole anim clip of me waving the board which is generally over 1500 frames
- I am doing this at NIGHT, no natural light around, I only have a small light on... and I also tried to shut all lights off and it makes no difference
- The board I am using is 35 inches x 35 inches... (.889 Meters x .889 meters) At first the board was just cardboard... but the tape I used to put it together was shiny so I thogught that might be causing issues. So then I covered it with a sheet.. so it was nice and smooth cloth in the front .. but that made no difference.

Anyhow, as I hope you can see in the attached images below... the board I am using is just covered with Yellow spots and I am guessing this is what is failing the calibration. YES, there is a chair behind the board.. but I just used it to just prop up the board for these pictures. It looks EXACTLY the same as when I am standing there holding the board myself. And yes I do hold the board away from my body. I have tried different camera angles... more floor.. Less floor... cameras together.. cameras more apart.. makes no difference.

I dont know what else to do, I am out of ideas. Please take a look at the images below and I Thank you in advance for any help

I think you have to hold it like a crane in the middle.

That chair is proabably not recognized.

That's odd...it really shouldn't be that difficult. We typically use two to three Kinect for Windows (v1) and rarely need to calibrate more than once per session.

I'm not sure what's going on in your situation but here are some tips:

I don't know if the size of the board is critical but the board you have seems a bit large. The board I use is a sheet of foam core, probably around 30 x 40. Perhaps smaller--I need to check on that at home--but definitely not larger.

When moving the board, a back and forth action is more important that 'waving' it around. The motion I create is typically back and forth and slight rolling on y and x axis. Keep the motion steady and smooth so that the device can capture the plane accurately.

Try to keep the board away from your body as much as possible. For example, if you bring it close to our legs, the 'changing' shape (as far as what's captured) may confuse the tracker. Try to visualize exactly what you are capturing and how the software will use the data. This is true for calibration but especially true when you start recording actions.

I see you're already avoiding sunlight--that's good. I find strong IR from sunlight can corrupt the data.

1500 frames is a bit excessive. I usually only record about 500 - 600 frames for calibration, or about 15 to 20 seconds.

I like to record calibration before and after a session. If the session is particularly long, I might do another one during the session. It's not necessary to do a calibration for each action--just 2 to 3 for the entire session should be plenty to fall back on in case one fails.

After recording the first calibration, I like to check it to make sure it's going to work out. Since the footage is only 15 to 20 seconds long (at least the part I want track for calibration,) it should go very quickly. This early tracking may reveal other issues that may be present that can interfere with motion capture later, and help you avoid ruining an entire session. If all is good, I'll switch back to Recorder and proceed.

Room lighting with Kinect isn't terribly critical since the tracking is primarily based on depth data. Even so, it's good to get decent lighting so you can see the reference video, which is often helpful when correcting errors. This is especially true with Kinect for XBox (v1) because it doesn't allow adjusting the video quality like you can with Kinect for Windows (v1).

Watch out for shiny reflective surfaces--this can mess up the data. I find that solid black surfaces may confuse data as well. If you look for where the yellow color is appearing, this can help you identify problems in the room. BTW, yellow in the distant background can be ignored--you want to watch for anything close to and interfering with the action.

try to stay away from walls and large furniture--if the performer gets too close to a large surface, that can confuse the shape the software is trying to track.

Hope this helps. If I think of anything else, I'll add the info to this thread.

G.

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

Watch a one minute excerpt on Vimeo now!

Thanks so much for these Tips... before reading your reply I actually went out and bought a bigger board... maybe I will cut it up and stick to the smaller one.

I will say I am still not having much luck, I have been tinkering with it all day and still yet to get something that is GOOD. I do find that when I go into MOCAP STUDIO 3 the cameras are setup as if I am doing a 180 degree camera setup.. but I am NOT!! Is this just due to me moving the board too much? This is clearly one of my big issues.

I am going to try a few passes with just moving the board forward and back with minimal movement and see what that does.

Thanks again for the help.

OK,

So I tried again... only moving the board back and forth.. with some slight roations without waving it around a lot. Didnt seem to help... but One thing I wanted to ask about is I noticed there seemed to be a fair bit of background noise in one of the cameras... is this normal? The other camera doesnt have this much.




Also... as you can see below... the calibration ONCE AGAIN thinks I am doing a 180 setup... My cameras are more like 40 degrees apart.. IF THAT... Its infuriating that it keeps reversing things. Is there ANY WAY this has to do with the ordering of the camera windows in the setup? I cant imagine that has anything to do with it does it?



Thanks again for any help

The noise doesn't look that bad to me.

I think you're board looks huge. I don't know if that has anything to do with it but iPi Soft recommends around 1 - 1.3m x 1 - 1.5m or roughly 3 x 4 feet. The sheet of foam core we use is about 30" x 40", which I think is quite a bit smaller than what you appear to be using.

I'm not sure the red color is helping either since the Kinect uses infra red scanning to detect depth. That's just a guess though--I'm not sure if red surfaces actually have an affect on how the device sees its own infra red light.

I would recommend trying what I'm using, a plain white sheet of 30 x 40 foam core, since it works perfectly well here. White foam core is easy to find--almost any office supply or art supply store carries it for about $4 - $6.

It's a bit hard to tell but it looks like you're pretty close to that wall too. I would step a bit closer to the sensors. When I'm calibrating a session, I try to stand as close to the center of the capture space as possible. I like to mark the floor with a piece of gaffers or masking tape so I know where the center is and where my bounds are.

I also recommend read the docs in the wiki more carefully. The link is here: http://wiki.ipisoft.com/User_Guide_for_ ... figuration

The 'good' example is what you should follow for size and how you hold the board. I recommend holding it farther from the body though. Moving the board back and forth about a yard or so is good. When I do this, I hold it out in front of me at arms length and step back and forth, being careful not let my body's volume get confused with the board. I also like to roll and tilt the board but try to keep the the surface facing both sensors when you do this. If you rotate it too far, the sensor will not be able to read the surface of board accurately, resulting in bad 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!

UPDATE!!

So I THINK I got things working... I wanted to update to help anyone out that may be having the same issue. I have to say this is frustrating since I am already close to 2 weeks in to a 30 day trial of this software. They should make it 30 days from first exported action or something.

ANYHOW!!
I went out and bought a small Foamboard..(Elmers is the company that made it)... its 20inches x 30 inches

The last 2 calibrations I did came out perfect... I did them the same way. I moved the board around the space.. but tried to keep it facing both cameras the whole time. I would rotate it in the Z ... (which is like a Propeller motion). I did VERY VERY LITTLE rotations on the Y... which to explain without 3D would be rotations as if it was hanging from a string and spinning.

So thats about it. And that brings me to my final question. This motion wasnt very Dramatic... at least in terms of what I was trying to do. But my quesiton is...

If I just stood there and did NOTHING, other than hold the board still for 700 frames. Would the calibration STILL come out as perfect? Or would it Come back and say failed due to not enough motion?

Thanks in advance for the response.

Glad it's working out for you.

Re: rotation, you do need motion in the board...it helps the tracker resolve a 3D solution. When you rotate it, it helps the system identify the corners and figure out which camera is where relative to the board. Without any rotation, the system has to do a bit more guesswork, which can result in incorrect camera positions.

Rotation on Y is fine, and it may even be preferable...you just need to be careful not to over-rotate it. I think when you rotate it at too sharp an angle, the rays may bounce or just not read to correctly to the Kinect that emitted the rays--naturally, this can become worse if the surface is relective. Also, if you rotate the board too far, one of Kinects may be able to see all four corners of the card or read them accurately.

I'm not sure rotating on Z is helpful...probably. I usually rotate on Y and X a few degrees to help the tracker.

Observe the shapes made by the point cloud. This will show you what Mocap Studio is looking for when it creates its tracking data. If the rectangle in the point cloud is vague and without clear edges, then Mocap Studio will have a harder time reproducing that rectangle in 3D space.

G,

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

Watch a one minute excerpt on Vimeo now!

Hello
The requirements for successful calibration with the board are:

• The board is well recognized by both sensors simultaneously. This depends on its size, material, whether it is apart from other objects and has a shape close to rectangular from the sensor POV.
• There are considerable changes of the board's position during the calibration.

General recommendations to meet these requirements:

• Use board that covers about 1/10 to 1/4 area of the image. Sizes between 30x40 and 40x60 inches usually work good.
• If the board surface on depth image has many yellow noise, then probably it has too reflective material. Consider replacing it with another board or covering with some not-glossy fabric.
• Hold the board away from your body by the middle of a side rather than by a corner. Corners should be clearly visible and distinct from other objects.
• The board is recognized better when its surface faces a sensor. This is very straightforward with 180 degrees setup. With 90- degrees setup, facing to about the point in the middle between the sensors should be preferred. Slight rotations around the vertical (Y) axis during the calibration are allowed (and desirable in 90- degrees setup).
• Move (not rotate) the board back and forth and side to side maintaining its facing direction in the range +/-0.5m. This improves the accuracy of calibration.

You could just try it, not that much time :) Range of the board's motion is one of the criterions to evaluate the success of calibration. I do not remember the exact limits, but definitely standing in the place won't get you a PERFECT result.