Structure inference and pose-estimation based on symmetry of object

Symmetry identification of an object is one important part of perceptual organization in the human visual system. This paper presents an algorithm for inferring structure and estimating pose or orientation of an object based on its symmetry. We first analyse the geometric relationships between 3D symmetry planes of an object and 2D symmetry axes of an image. Based on them we can determine the existence of symmetry planes in 3D from image symmetry and recover the structure of the object. Then we calculate the orientation parameters of three mutually orthogonal symmetry axes in 3D, and obtain the pose of the object. The experimental results are very satisfactory.<>