A Unified and Complete Framework of Invariance for Six Points

摘要

Projective geometric invariants play an important role in computer vision. To set up the invariant relationships between spatial points and their images from a single view, at least six pairs of spatial and image points are required. In this paper, we establish a unified and complete framework of the invariant relationships for six points. The framework covers the general case already developed in the literature and two novel cases. The two novel cases describe that six spatial points and the camera optical center lie on a quadric cone or a twisted cubic, called quadric cone case or twisted cubic case. For the general case and quadric cone-case, camera parameters can be determined uniquely. For the twisted cubic case, camera parameters cannot be determined completely; this configuration of camera optical center and spatial points is called a critical configuration. The established unified framework may help to effectively identify the type of geometric information appearing in certain vision tasks, in particular critical geometric information. An obvious advantage using this framework to obtain geometric information is that no any explicit estimation on camera projective matrix and optical center is needed.