[attachment=13291]
ABSTRACT
Optical three-dimensional non-contact profilometry has been widely used for 3-D sensing, machine vision, robot simulation, industrial monitoring, biomedicine etc. Several 3D object profilometry methods including phase stepping profilometry (PSP), Fourier transform profilometry (FTP), modulation measurement profilometry (MMP) and spatial phase detection (SPD), have been exhaustively studied. Measurement of shape through any one of these fringe projection techniques involves- (1) projecting a sinusoidal fringe pattern on to the object surface, (2) recording the image of the fringe pattern that is phase modulated by the object height distribution, (3) calculating the phase modulation by analyzing the image with one of the fringe analysis techniques (such as FFT, phase stepping and spatial phase detection methods- all of them generate wrapped phase distribution) (4) using a suitable phase unwrapping algorithm to get continuous phase distribution which is proportional to the object height variations.
Earlier investigations on the real-time 3-D shape acquisition methods used in diverse fields indicate a growing demand for the development of 3D shape measurement techniques with as minimum number of input images as possible (Ideally one image). It is generally recognized that the discontinuities present in the objects invariably pose several challenges to all of aforementioned profilometric methods and impose a demand for larger number of input images to achieve reliable phase unwrapping. Investigations reported in this thesis seek to enhance the performance of fringe projection based 3-D profilometry, primarily in respect of the number of input images required. This is achieved by proposing a new phase-unwrapping strategy.
Principal contributions emerging from the investigations reported in this thesis are listed as under:
1. A new phase unwrapping approach is proposed and demonstrated. In this new approach, information extracted from an additional image of the object recorded under the illumination of a specifically designed color-coded pattern, is used. The proposed procedure led to realization of accurate and reliable unwrapping with lesser number of the input images than any of the contemporary unwrapping algorithms even in the presence of true phase discontinuities.
Use of the proposed unwrapping approach invariably requires a color-coded pattern for projection. Therefore use of two specific color-coded patterns, which are named PATTERN-I and PATTERN-II, in achieving fast unwrapping is demonstrated. This has resulted in the following two contributions:
(a) Proposed unwrapping algorithm when used with PATTERN-I is demonstrated to be capable of reducing the number of input images required by PSP from 16 (when best contemporary unwrapping algorithm is used) to 5. (for measurements carried out with the pattern containing 8 fringes. i.e., S=8).
(b) Proposed unwrapping algorithm when used with PATTERN-II is demonstrated to be capable of reducing the number of input images required by PSP from 24 to 5, and by FTP from 6 to 2 respectively. (for S=32).
2. Novel single-shot structured light technique for real-time measurements: novel design of the color-coded pattern and associated processing mechanisms are proposed which facilitated in obtaining accurate, reliable and high resolution 3D data from a single input image.
3. Investigations on Fringe Analysis Techniques (FAT): It has been observed that not all fringe analysis techniques are suitable for realizing an efficient single-shot method if the object under investigation possesses certain features in its geometry and/or color texture variations. The quality of the generated wrapped phase map affects the overall accuracy of reconstructed profile. Therefore, investigations are carried out to study and evaluate errors introduced by different fringe analysis techniques in wrapped phase generation and thereby identify the best one in these case studies:
a) The object under investigation possesses surface discontinuities and/or has non-periodic height distribution with in the imaging window (Issues related to Energy leakage)
b) The Object has large color texture variations: Experimental investigations on 3D shape measurement of objects with large color texture variations (Human faces with and with out facial hair) revealed superiority of wavelet transform analysis over FFT analysis. Result of this investigation has led to an important contribution that has served as a predecessor for the development of beard, mustache invariant 3D face recognition system.
These case studies helped in identifying wavelet transform analysis as the best choice for calculating wrapped phase maps. Using wavelet transform analysis together with proposed phase unwrapping algorithm has led to the development of efficient single-shot 3D shape measurement technique.