High-quality 3D shape measurement with binarized dual phase-shifting method
Is Version Of
3-D technology is commonplace in today's world. They are used in many dierent aspects
of life. Researchers have been keen on 3-D shape measurement and 3-D reconstruction
techniques in past decades as a result of inspirations from dierent applications ranging from
manufacturing, medicine to entertainment. The techniques can be broadly divided into contact
and non-contact techniques. The contact techniques like coordinate measuring machine
(CMM) dates way back to 1950s. It has been used extensively in the industries since then.
It becomes predominant in industrial inspections owing to its high accuracy in the order of
m. As we know that quality control is an important part of modern industries hence the
technology is enjoying great popularity. However, the main disadvantage of this method is
its slow speeds due to its requirement of point-by-point touch. Also, since this is a contact
process, it might deform a soft object while performing measurements.
Such limitations led the researchers to explore non-contact measurement technologies
(optical metrology techniques). There are a variety of optical techniques developed till now.
Some of the well-known technologies include laser scanners, stereo vision, and structured
light systems. The main limitation of laser scanners is its limited speed due to its point-by-point
or line-by-line scanning process. The stereo vision uses two cameras which take pictures
of the object at two dierent angles. Then epipolar geometry is used to determine the 3-D
coordinates of points in real-world. Such technology imitates human vision, but it had a
few limitations too like the diculty of correspondence detection for uniform or periodic
textures. Hence structured light systems were introduced which addresses the aforementioned
limitations. There are various techniques developed including 2-D pseudo-random codication, binary codication, N-ary codication and digital fringe projection (DFP). The
limitation of 2-D pseudo-random codication technique is its inability to achieve high spatial
resolution since any uniquely generated and projected feature requires a span of several
projector pixels. The binary codication techniques reduce the requirement of 2-D features
to 1-D ones. However, since there are only two intensities, it is dicult to differentiate
between the individual pixels within each black or white stripe. The other disadvantage is
that n patterns are required to encode 2n pixels, meaning that the measurement speeds will
be severely affected if a scene is to be coded with high-resolution. Dierently, DFP uses
continuous sinusoidal patterns. The usage of continuous patterns addresses the main disadvantage
of binary codication (i.e. the inability of this technique to differentiate between
pixels was resolved by using sinusoid patterns). Thus, the spatial resolution is increased up
to camera-pixel-level. On the other hand, since the DFP technique used 8-bit sinusoid patterns,
the speed of measurement is limited to the maximum refreshing rate of 8-bit images
for many video projectors (e.g. 120 Hz). This made it inapplicable for measurements of
highly dynamic scenes. In order to overcome this speed limitation, the binary defocussing
technique was proposed which uses 1-bit patterns to produce sinusoidal prole by projector
defocusing. Although this technique has signicantly boosted the measurement speed up to
kHz-level, if the patterns are not properly defocused (nearly focused or overly defocused),
increased phase noise or harmonic errors will deteriorate the reconstructed surface quality.
In this thesis research, two techniques are proposed to overcome the limitations of both
DFP and binary defocusing technique: binarized dual phase shifting (BDPS) technique and
Hilbert binarized dual phase shifting technique (HBDPS). Both techniques were able to achieve
high-quality 3-D shape measurements even when the projector is not sufficiently defocused.
The harmonic error was reduced by 47% by the BDPS method, and 74% by the HBDPS
method. Moreover, both methods use binary patterns which preserve the speed advantage of the binary technology, hence it is potentially applicable to simultaneous high-speed and
high-accuracy 3D shape measurements.