Hey guys, I've been wanting to do this project idea for a while now. I want to mocap dancing and maybe do facial mocap as well (or animate the face myself) (I have an iphone I could use the iphone facecap for face)
but I was looking into the full body mocap suites, and I seen Perception Neuron, Rokoko, and some new AI stuff Deep Motion and theres ipisoft with kinects or ps eye cameras.
I want to mocap the dancing, and then cleanup the frames, but I need the mocap data to be somewhat stable (I notice foot sliding or hovering above ground in many examples) Is there ways to fix these imperfections easily? Like "anchor foot to floor, auto apply to all frames"
I see there's Motion Builder (I've never used) Is this the only viable motion capture clean up software out there? I mainly use Blender now TBH, I use to be big on 3DS Max and some C4D. so I could hop back into autodesk world, I haven't seen much about mocap with blender TBH.
anybody got any advice? I basically want to mocap dancing animation, hand animation, (and maybe face) and not have the legs sliding/glitching everywhere or have an easy way to clean it up, then apply it to a universal rig for different characters, and desync the animation on each character for different timings and stuff, or make little adjustments, so I can have alot of people dancing at the same time and not have it all be exactly similar (hint: Idolm@ster games)
Any advice?
Replies
All mocap data will have errors. The errors are caused by different things depending on the mocap system.
Visual mocap is done by tracking reflective points on an object - this is the kind where you see actors covered in ping pong balls. Because it's visual you get issues where some of the points are behind the actor or too close together and that causes problems with tracking. Most of the systems that use this have ways to filter this out in software but it's not always successful depending on how it works. The best systems use lots of high resolution cameras from different angles that reduce the number of these kinds of errors. To do it properly you need to setup the position of these cameras as accurately as possible and this is best done in a dedicated studio.
Magnetic or angle based mocap uses a suit that has multiple tiny accelerometers and angle meters all over it - it looks like a bionic suit with lots of wires running over the outside of the suit. This has the advantage that you can mocap anywhere with enough flat space and because the data is being recorded on the suit you can have complex movements that would normally be impossible on a optical system. The problem is that sometimes you get errors from interference by nearby electrical sources. And any small errors from the sensors will get compounded over time to produce large drifts in data. Sensors moving position on the suit can be a bigger problem than optical systems.
With both these kinds of systems the best way to avoid errors is by knowing what kinds of actions produce the least amount of errors. For example if you have a dancing scene where you have to dance with a partner an optical system might work better if you recorded each dancer separately - so they don't occlude each others sensors. This can make recording fighting scenes interesting. Magnetic systems might not work well for scenes with complicated tumbling and jumping as it won't be able to interpret the spinning of the actors properly.
Motionbuilder is the industry standard for cleaning up mocap. It's designed to be used with maya. It has many great tools that will let you do things like lock the feet selectively and transfer animation data between two different rigs. I've also used biped in 3dsmax to do similar things - it's not as good as Motionbuilder but does do the job.
However you can do a lot of these things using standard IK systems if you're good at setting up rigs and technically minded. I would not be surprised if there wasn't some scripts and rigs for Blender that let you do similar things. What you'll need is a good IK rig and animation blending and animation retargeting tools.