I would really like to see a way to manually calibrate the PS3 eye cameras. Is this a feature of the software that I have somehow missed? I did a shoot outside on an overcast day but the mini maglight was not bright enough. If I could manually adjust the tracking position of the light in each camera I could make this footage usable. I imagine that other people have had similar issues. All I need is to be able to move that little tracking box around in each view.

:8^[

iPi Mocap Studio has no way of manually correcting marker's position. It just would be too much of the hand work to set it in each frame for each camera.
Could you upload your calibration video to some file sharing service? We can probably use it to improve the algorythm for automatic marker detection.

As a customer, I would have to disagree. Here's my situation which I could see other people falling into as well.

1) I have a laptop so I can do portable recording. This in not the same computer that I use for my main license so I won't be able to check the calibration on site.
2) I shot a bunch of footage with four PS3 eye cams and my fingers crossed on the calibration. I used a mini-maglite as recommended and shot on an overcast day. I shot outside as it has been difficult for me to find an appropriate indoor space.
3) I would be happy to manually calibrate if that was my only option to make this footage usable. It sounds like a lot of work but it shouldn't take more than an hour or two in the worst case. Again, as a user, this is a much better option than scrapping the entire shoot. If there was a way to import/export the video I could even do this in another program like AfterFX.

Here are a few suggestions:

* Allow the user to adjust the tracking marker at specific frames like you can do when you are tracking the body motion. This would allow the user to fill in any gaps in the auto tracking.
* Consider adding a manual calibration mode. Do you really need all of those data points? What if the user could just go to specific frames and mark the same spot in each camera? Surely you would only need several points to make this work. Photogrammetry software has been doing this for years.


I'm uploading the calibration footage. I'll PM you the link when it's done.

mike
:8^/

Got the message. Downloading the video...

Hey mperedo!

If you don't mind the leg work you can save your calibration video by editing the video in any editing software. You could potentially mess with contrast or do what 1K0 did for his Danse Kabyle video (http://vimeo.com/22036677). He seems to have added a white block in a video editor that he either had to hand manipulate, or it was tracked by pixels.

I hope this helps

Oli

Hi, mperedo

We've looked through your video. The bottom line - it is a bad calibration video. The reasons are:

• You've placed the cameras too far away from capture area.
• The marker is bad distinguishable even with the eyes.
• You moved only in a small part of capture area. And you should move the marker through the whole capture volume you are planning to use.
• You have not touched the ground even once. It is required to mark several ground points (3-4 at least) for software to be able to detect the ground plane. I suppose you skimmed some important parts of our guide for making calibration video.

At least you can playback your video and estimate how good it is.

Manual adjustments in calibration process will always have less precise results then automatic computations. And this usually leads to low quality of tracking.

The more points - the better accuracy. Just it. Finding the center of several pixels bright area in image is not that accurate in essense to rely on measurements from several frames.

What you can do now if you really cannot reshoot is to manually set camera positions. Adjust actor's model in Studio to match your body height/proportions and appearance. Select some frame where you are standing in a simple pose. Adjust model's pose to match your pose on video. Adjust each camera position/direction to match model with you on video image plane. Iteratively continue adjusting model's pose and cameras positions until your are satisfied with accuracy of matching.
This method (you can indeed call it "manual calibration") will give about the same accuracy as you would manually adjust marker positions. For convenience, you can use a T-pose frame from an action video for such calibration, because it does not need any markers besides actor's body.

IMO, it's probably easier to just reshoot your session than to try to fix bad data. When shooting mocap for the first time, keep it simple so you can clearly see what works and what doesn't. Back when I was starting to use iPi DMC (1.0 beta) there wasn't much in the way of documentation so I experimented with many different setups before I committing the system to actual production. Even today, I may shoot multiple versions of actions just to be sure I have something that I can use. (Not unusual if you consider live action shoots are always done in multiple takes.)

Since we're talking specifically about calibration videos, I suggest shooting two or three different calibration patterns to see which one work best for you. When using PS3 Eye cameras, this has been the pattern that works best for me:

Calibration Video

I always use automatic camera positions and, if the calibration data has been properly shot, it should work perfectly every time! Trust me, this is far easier and more accurate than manually positioning the cameras.

The key things to watch for when shooting calibration video:

- clear and constant visibility of the Maglite to all cameras. Avoid occluding the light with your body as much as physically possible.

- the Maglite should be the brightest thing in the shot.

- avoid rooms with reflective surfaces (shiny floors, large stainless steel appliances in BG, exposed windows, etc.)

- if you're shooting outdoors, shoot on an overcast day, very early in the morning or within a shadowed area. Try using a bright white or chrome ball on a stick as your tracking marker--this may be more visible than the Maglite when outdoors.

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!

Hey guys. Thanks for all the great replies.

You're correct in that I hadn't done all my homework in preparation for this shoot. It was my first attempt with the four camera setup and I really just wanted to see what the results were like compared to the Kinect. I did have the wiki guide on my laptop in PDF form that I was intending to refer to for the calibration. However, I had just done a fresh install of Windows 7 so I could get both Kinects working (no success under Vista) and I had neglected to install a PDF reader! I was also short on time as my actor only had a small window of time that day.

That said, I was still quite frustrated to find that the maglite was not bright enough for the cloudy day and the software did not appear to be tracking the brightest object on screen correctly. I subsequently did a proper shoot in a more controlled environment and I now have some good footage and calibration to work with.



I'm really interested in this. I can see how to export video from the ipi software but how can I assemble it back again into an ipi format file? Can anyone recommend a method for doing this?



There doesn't seem to be a guideline for how far apart the cameras can be in the guide. I used similar camera spacing in the subsequent shoot that worked. Is there a recommended maximum distance between the first and last camera?



Setting the calibration from the actors body is certainly worth a try. I did try aligning the grid to match reference points in the ground but that didn't work well. However, I still think that being able to mark the same point in each camera should be possible. I'm content to put this on the backburner for now.



I still seem to be having issues getting the Maglight to track correctly. In the second controlled environment I shot a few calibration passes with the maglight and another bright source. I still found that the tracking software was getting confused and not tracking the light correctly. Fortunately, I tried a version with the lights out that did ultimately work. The wiki guide says not to do this but Ian Chisholm's guide mentions this as an option so I tried it and it saved my shoot. I did try the darkening option on the PS3 cams but it didn't help.

mike
:8^1

Hey Mike,

I'm just finishing up a video talking about calibration with PS Eyes and from the data I have so far I find taking these steps helps.

- Circular motion from top to floor in middle of calibration space (makes you dizzy)
- Then "wipe the windows" of your capture space. By this I mean that if you have a square "sweet spot" then slowly wipe from top to bottom on each of the 4 sides.
- Then wipe the outsides of the windows, stand outside of your square and wipe each side down.
- Do some last waving inside of your sweet area
- Do 8 ground points making sure camera can see all of these, 4 on the square points then one in between each point. I do this because if any of the ground points are occluded you stil have many more to chose from.

This (so far, still got one more calibration test to do) has given me good reliable results each time. I should note I use 4 Ps Eyes like Greenlaw but in a 360 square set up, not a corner set up. (By the way, awesome video Greenlaw!)

I hope this helps

Oli

Camera locations depend on many factors. They should be positioned to provide best view of the capture volume from different points. Usually, a distance of 1-2 meters from boundaries of the capture area is suitable. However, total distance between opposite cameras should not generally exceed 10 meters, because otherwise actor's image in the most distant camera will be too small for reliable tracking.

The wiki says not to turn lights off solely for the safety of a person doing the calibration. Dimming the lights surely helps with marker detection in case your model of Maglite is not a big one.