| Summary: | 碩士 === 元智大學 === 工業工程研究所 === 89 === To facilitate the development of a human-popularized environment, a low-cost, easy-to-operate motion capture method has been developed in a previous work (Chang, 1999). This method is based on an important property of conics theory: the 3D rotation and translation of a planar circle with known radius could be derived from its 2D elliptic projection.
This work focus on the improvement of several drawbacks suffered by previous system. First, we employ the Java Media Framework (JMF) to integrate the video camera with our motion tracking system. Secondly, (Chang, 1999)’s approach has problems identifying boundary pixels when ellipse are broken or overlapping with each other. In this work, we develop four constraints to correctly extract boundary pixels of ellipses in consecutive images. Thirdly, a forward kinematics approach, together with the quaternion transform technique, is used to automatically specify the pose of the 3D computer graphic model of a virtual human. Finally, we discuss the accuracy evaluation of our motion tracking system under digitalization error.
|
|---|
