Line Detection in Range Images

Detection of linear structure is a very important problem in image processing and computer vision. The task of finding lines in 2D images has long being studied, but the work in 3D space does not have any promising work yet. This paper investigates the issue of line detection for range images. It proposes an approach to find a wire-frame composed of lines that can represent precisely and comprehensively line features in a range image. Our approach tries to combine two targets together: to locate the lines precisely at where 3D data discontinuity occurs and to keep the lines consistent with high level constraints for the structural lines of 3D objects. A global fitness measure is proposed to evaluate a wire-frame composed of lines and global optimization techniques are used to maximize the value of the fitness measure.