A New Approach for Multi-view Reconstruction of Complex shapes

A-DEWS 2015: Design Engineering in the Context of Asia, Asian Design Engineering workshop, 29th - 30th October 2015, The Hong Kong Polytechnic University

Year: 2015
Author: Juliana Tam, M.Y.; Sun, R.L.1; Yu, K.M.
Institution: 1 State Key Lab of Digital Manufacturing Equipment and Technology, School of Mechanical Science and EngineeringHuazhong University of Science and TechnologyWuhan, China, 2The Hong Kong Polytechnic UniversityHong Kong, China
Section: Business
Page(s): 179-184

Abstract

There is a strong interest in effective reconstruction of complex shapes from multi-view capture in aero-engine maintenance application. A significant limitation of existing iterative-based approach for 3D reconstruction comes from the fact that it is a time-consuming; need of good initial pose and substantial overlapping; and inefficient handling of multi-view. That partially explained why most practices of blade inspection are limited to the analysis in partial, in section or at the tip area. In this paper we present an alternative solution by coupling mechanical tracking and added geometric constraints to solve the corresponding problem of featureless data to enhance robustness and efficient multi-view reconstruction. The approach uses a mechanical based transform to initialize an optimization search. The concept is making the data set implicitly connected to virtual axis that provides rigid spatial relationships for robust multi-view alignment. The methodology of adding geometric constraints act as “anchors” to further limit incorrect movement across featureless regions. Then, the refine optimization based registration is performed using non-linear least square algorithm. The Bespoke Measurement system was developed to demonstrate the feasibility of proposed method by integrating the scanner with hardware mechanism and intelligent algorithms for efficient multi-view reconstructions from of complex shapes. For reliable performance, we also present reference based method for system calibration with relatively simple and fast procedure. A k-d tree structure is used to facilitate the data matching. Experimental results demonstrate the validity of proposed method. The proposed approach represents the basis for automation of the accurate reconstruction of complex shapes.

Keywords: range images; automatic registration; mechanical tracking