MoCap-Solver: A Neural Solver for Optical Motion Capture Data

Motivation: In optical motion capture, errors occur.

1. Eliminate the error
2. A skeletal agent mesh must be prepared for the actor/actress and used to determine skeleton information from the markers (re-targeting or solving). (What does this mean?)

Motion Capture and Application Process

• In this paper, we only consider motion capture with Marker. Here is a video to show what is motion capture:

https://youtu.be/s9CiGsy7VcM

• As you can see when we do the motion capture, the marker usually will lose track due to the occlusion. I think this is the majority of the errors.
• After the motion capture with markers, we have a sequence of noise markers. We need to associate markers with the skeleton template. So that the model can move smoothly. Here is the video for instance association:

https://youtu.be/3WZSCVeGblU

The Definition of Mocap Solver

The procedure after obtaining the marker and building the link between the template skeleton

MoCap Slover Method:

• Find the intrinsic relationship between the template skeleton, marker configuration and motion relationship.
• The authors used an encoder-decoder architecture to encode the markers to a latent vector and used a decoder to get a sequence of clean markers and skeletons. (There are a lot of traditional methods that can transfer markers to the skeleton, so I think we should only get the true markers)
• Provide a novel normalization strategy, learning a pose-dependent marker reliability function