Multiview Registration And Surface Development Of Range Images In Optical 3D Sensing

This dissertation addresses the following three problems of optical three dimensional sensing: coordinates calibration in phase-measuring profilometry (PMP), multi-view registration of range images and flattening of doubly curved surfaces. The principal contributions of this dissertation are:1. It presents a new method for the system calibration in PMP. With a special calibration gauge, both the phase-height mapping and lateral coordinates calibration could be done simultaneously through only one measurement. By an iterative procedure, this method is able to recover the continuous phase distribution of all steps with same height.2. It introduces a fast iterative pairwise registration method. Which combines four acceleration techniques: fast detection of overlapped regions by reverse calibration, more-careful selection of control points, fast surface closet point computation with closet vertex caching and compatibility test of pairing points based on the difference of modulation. With respect to the registration of multi-view range images, a relaxation strategy is introduced to spread evenly the accumulated registration errors among all views, with the assumption that the pairwise registrations are nearly good.3. It presents a method for constructing the initial flattening guess of triangulated surface. By this method, parametric and polygonal surfaces could be flattened in a general way. Residual energy distribution pattern is proposed to evaluate the global quality of flattened pattern. Also considered is the flattening of surfaces with arbitrary darts and gussets. With the application of flattening of surfaces in shoe CAD using laser range images, it proposes a whole processing procedure starting from the preprocessing of points cloud.