A RealTime Adaptation of Inverse Kinematics for Motion Capture

Abstract

In this thesis, the process of animating characters from motion capture in real-time is presented. A model of a human skeleton is defined, and the joints of this skeleton are located using positions of markers placed on the body during motion capture. These methods for locating joints are presented and evaluated. During motion capture, markers might be occluded from cameras, resulting in joint positions that can not be determined. This problem is, in this project, solved by estimating the missing joint positions with the technique Inverse Kinematics, in a way that, to our knowledge, has not been used before. Four different inverse kinematics solvers were implemented in order to determine if one algorithm performs better than other. The implemented algorithms are the Jacobian Transpose, Damped Least Squares (DLS), Cyclic Coordinate Descent (CCD), and Forward and Backward Reaching Inverse Kinematics (FABRIK). The implemented system receives marker positions from the motion capture system in real-time and rst performs joint localization followed by estimation of missing positions with IK. The result is a skeleton with a similar pose as the actor frame by frame. The application works well, but need improvements for smooth animation. The fastest and most accurate algorithm was CCD, but the other three are promising for possible future implementations.