英文摘要 | Image based object pose estimation is an important topic in computer vision, and finds its applications in a variety of fields, such as vision based robot navigation, virtual reality etc. This thesis is focused on object pose estimation by point correspondences from space to image, such as new estimation approaches, and solution multiplicity and stability etc. The main results include: 1.We prove that the popular cost functions of the PnP problem in the literature can be reformulated as some kinds of equivalent potential energy of a dynamic mechanical system composed of a rigid body and springs, then the PnP problem solving is converted into finding the system’s configuration by iteratively minimizing the energy cost. A simulation of this physical progress with resistance and force of springs is carried out, and stable solutions are obtained. 2.One dimensional calibration is an important calibration approach. In the literature, either geometric intuition or some ingenuity is needed to derive the calibration constraints for 1-D calibration. Since the calibration stick’s position before and after the motion is coplanar, and the homography encapsulates all the information of the projection process from the space plane to the image plane, it must be possible, methodologically speaking, to derive the same calibration constraints via the planar homography. In this work, we present a new method to derive the similar calibration constraints via homogaphy, and our method appears general, and systematic. 3.It is shown that if the projections of a circle and its center are both known, the image rectification can be done up to either a rotation or a reflection. However in general, the projection of circle’ center is not the center of the projected ellipse, and additional information is needed to locate its projection. In this work, we investigate the possible constraint on the projection of the circle’s center by adding a coplanar pair of lines with known including angle. It is shown that the resulting constraint is generally a curve of 6-degree. However when the two lines are perpendicular, the constraint degenerates into a curve of 2-degree. In addition, for the perpendicular line pair, its geometric interpretation, and specific curve forms are provided depending on whether the lines intersect ( not) the circle, or tangent to the circle. 4.We systematically investigated the relationship of the solution’s multiplicity and stability for the PnP problem. W... |
修改评论