Constraints for Recognizing and Locating Curved 3D Objects from Monocular Image Features

Abstract. This paper presents viewpoint-dependent constraints that relate image features such as t-junctions and inflections to the pose of curved 3D objects. These constraints can be used to recognize and locate object instances in the imperfect line-drawing obtained by edge detection from a single image. For objects modelled by implicit algebraic equations, the constraints equations are polynomial, and methods for solving these systems of constraints are briefly discussed. An example of pose recovery is presented.

• Type: Conference
• Year: 1992
• Where: ECCV
• Authors: David J. Kriegman, B. Vijayakumar, Jean Ponce
• Keywords: Computer Vision, Constraints Equations, Curved 3D Objects, ECCV 1992, Viewpoint-dependent Constraints