Operation: Motion Capture - The Event

I don’t know what being an “indie” developer means to you dear reader (I suspect it’s different for all of us that frequent this site ...that’s another blog for another time perhaps) ...for us it means having a day job that isn’t in game development, doing all this stuff in our spare time and not having a budget aside from what you can syphon from the housekeeping and savings.

So I hope you’ll understand what a big deal Sunday the 10th of October was for us when I say that, using the trial version of iPi Softs markerless motion capture system, 4x PlayStation 3 “Eye” cameras, a bunch of tripods some active USB cables and a couple of PCs, we attempted to motion capture 80% of our player character animations in the public hall of a sleepy village in Kent.

I’m guessing that a lot of indies will be interested in such a reasonably priced solution to such a (usually) expensive problem like Motion Capture, so I’ll cut to the short version:

The Short Version
The Jury is still very much out at this point, the system was very “twitchy” (even in what I would consider were ideal conditions) and our early results from the data that we’ve processed so far are being described as “mixed” as at 08:24 on this cold Monday morning.

The not so short version
We had the venue from 9am to 5pm ...allowing 1/2 an hour at either end of the day to setup and scratch the rig that’s still a fair chunk of day.
The hall itself is the biggest at the site (its main purpose is as a theatre for local amateur dramatics) and as such has reasonably controllable lighting conditions (curtains, blinds and spot lamps).
Those of you familiar with our early tests will understand why I highlight those features specifically (see here and here).




We arrived shortly before 9am and, following a period of frenzied activity, the rig was setup with all four cameras ready to go by 9:30am. We had 1 PC capturing and one PC processing the data so we could see what we were getting and still carry on capturing stuff on the main PC.



To ensure that the background was as controllable as possible and it wouldn’t interfere with what the system was “seeing” we brought along a green-screen ...turns out it wasn’t needed, but better safe than sorry.
We also brought along an inflatable mattress to act as a “crash mat” for falls etc. ...turns out I killed that (with a loud pop) as soon as I landed on it during a practice fall - oops :)

Once setup we immediately set about calibrating the rig, this basically involves waving a light about in the performance area while the cameras track it in 3D.
The light was being picked up great by the software and so we set about going from the initial T-Pose and into to a bit of *real* simple movement to see what the rig was “seeing”.

Here’s where the horror starts.

Almost immediately the virtual version of me wrapped his legs round his own waist, put one arm behind his head and did all sorts of stuff that the real me:

a.) Wasn’t doing
b.) Would find way too uncomfortable

Cue about 4 hours of “trying stuff to make it work” and quite a lot of head scratching.




It was frustrating, we’d actually had better results a couple of months earlier testing with just 2 cameras in a confined space (see the "uncleaned" version of the data here).

Eventually all our “trying stuff” brought us to the point where we replicated as closely as possible the conditions of the previous tests ...we removed 1 camera from the system so we were running with just 3 (3 is the minimum), we made the room a lot darker by closing all the curtains, blinds etc. and we tightened up the performance area bringing the cameras much closer to the “action” and it’s at this point that it started working.
By now the performance area was too small to do a run cycle and only just big enough (at a push frankly) to do a walk cycle.
The results we were seeing were quiet encouraging but still contained some glitches but we really wanted to salvage something from the day (besides experience and knowledge obviously) and so we pressed on and started working through the “must have” section of the 113 or so total moves that we have worked out so far.

At this point it was about 13:40.

Two hours later we’d done most of the moves that we thought we could do (plus some adlib stuff for emotes etc.). We had to remove a bunch of stuff from the plan, stuff like ladders, running, sprinting, but we did get a couple of walk cycles, weapon reloads and miscellaneous other bits and bobs (basically anything that didn’t require a lot of room).

As I mentioned above the resultant video footage is being processed at the moment so we should know in the next couple of days what it looks like and where and how much we need to bring in our animation clean up guy.

It’s obviously too early to say at this point whether we’re going to use any of what we captured yesterday (and therefore stump up the extra investment needed in the full version of the software if we do).
I am going to drop iPi Soft a note and link them to our results to see if there was anything we could’ve done better and hopefully our experiences will help their development process in someway.

All that aside... it was loads of fun if very tiring. This is one of the most epic things we’ve done as a team so far and I wanted to be sure that we didn’t miss what happened and had a record of the day ...last I heard our official photographer (my wife, Charlotte) had taken some 300+ pictures some of which are here, the rest of the (heavily edited) highlights can be seen on the Epoch: Incursion Facebook Group {here:}



Thanks to all involved in the day, Team EiKON would especially like to extend their heartfelt thanks to our friends the two David's and to my wife Charlotte. Thanks for stepping into the maelstrom and acting as support crew guys, both physically and morally - it was very very much appreciated by all of us.

If anyone in interested to see videos of the raw data (without any clean up done) then I’ll post a reply to this blog with a link to the video as soon as I have it, just subscribe to the thread so that you'll know when it arrives.