Relative Affine Depth: Structure from Motion by An Uncalibrated Camera

. In this paper, a new projective model for 3D information representation, termed relative affine depth (RAD), is derived for the solution to structure recovery of an object at arbitrary positions with respect to uncalibrated cameras. The RAD of a spatial point is proven inversely proportional (subject to a projective scale) to the Euclidean depth of a spatial point in a camera-centered coordinate frame, which is sequentially obtained in a straitforward way. The point matches are accomplished using a feature point tracking algorithm developed by us. Experiments with real objects show good results for shape recovery. 1 Introduction In recent years, projective and affine reconstruction have significantly influenced the researches on structure from motion (SFM) [1]. In the cases where metric measurements for object shapes are not essential, people attempt to use projective or affine framework to formulate the camera projection and image transformation. Under these frameworks, difficultie..