The estimation of 3D object motion trajectory

• Main Authors: Jaw-Yeh Chen, 陳肇業
• Other Authors: Sheng-Fuu Lin
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/47066164761883791110

博士 === 國立交通大學 === 電機與控制工程系 === 90 === The main approaches in this dissertation are the discussions of correspondence, pose estimation, and modified optical flow problems for 3D object motion estimation. Binocular image sequence data is used to solve the correspondence problem. When stereo imaging technique is used to produce binocular image sequence data, some properties naturally occur; then the correspondence problem can be solved by a very simple procedure. Pose estimation problem usually needs to find the least-squares solution of the error function; then the singular value decomposition is applied. The procedure always uses all feature points; hence, the computation burden is heavy. A norm vector-based procedure is proposed to obtain the pose and hence the motion trajectory. We use the change of norm vector to determine the plane, in which the rotation axis lies; then another change is used to determine the rotation axis and the rotation angle is obtained. Then the translation can be determined. Since only four points are used, the computation burden is light. Many optical flow techniques are not suitable when rotation is presented. Hence, a modified optical flow technique is proposed in this dissertation. Traditional methods usually obtain the flow in Cartesian coordinate; The flow is obtained in polar (spherical) coordinate in this dissertation. When the flow is calculated, then the rotation and translation component can be calculated very easily. At last, we extend the above discussions to 3D space optical flow.