Motion and Structure from Significant Segments in Man Made Environments

Abstract

An algorithm to determine the camera motion and the structure in a 3D man made environment is proposed. The algorithm is based on straight edges extracted in the image and some plentiful geometrical relations between them (verticality, horizontality, parallelism, perpendicularity). In this paper the computation of motion and structure using these constraints is tackled. The information of the tips of the straight edges are also used to determine motion. Some partial results using two images are now available. Its application to a mobile robot running in man made environments could be considered.

Motion and structure for vision-based navigation

“…6 An anisotropic noise model for the location of the point that takes into account the noise of the projected line has also been proposed. 20 Besides that, to have information of relative depth, the constraints in the direction of significant lines provided by the previous process ( §4) are considered. This information allows us to uncouple translations along an axis with rotations around its perpendicular axis in the image plane.…”

Navigation from Uncalibrated Monocular Vision

“…As neither previous map nor robot heading is assumed, we must obtain structure and motion information from vision (Sagüés and Guerrero, 1995). The computation of the direction of the translation and the time to collision based on the focus of expansion (F OE) is a classical result (Burger and Bhanu, 1992), (Micheli et al, 1993).…”