Implemented my own framework for pose comparison & pose/motion retargeting as well as the demo for motion learning platform.
Pictures below could be played on click (it's animated GIFs).
Video (attached below) was prepared for the Perception Neuron Mocap's contest .
* Input: global rotations (quaternions) for each supported joint
* Output: Pose vector that consist of Degrees of Freedom of each joint
* Could be applied not only to human's body but for any ball-and-socket group of joints
* Joints could be configured separately so it really doesn't depend on input device: Kinect v1/v2, Leap Motion, Mocap data
* Basic filtering applied on top of Output DoFs vector
* Comparison metrics (pose & timing) are consist of weighted average between compared DoFs
Motion Learning app demo features:
* The main goal is to provide real-time estimation & metrics
* Per-joint highlights based on DoF distance metrics
* Ghosting mode
* Pose & Timing scores