Description
System and method for detecting irregularities on a three-
dimensional surface
The present invention relates to inspection of irregularities
on three-dimensional surfaces, including, arbitrarily curved
surfaces.
Inspection of irregularities, such as embossments, on various
items such as machine parts is a problem of wide interest
since it affects quality control and sorting. The
aforementioned problem can be solved by a three dimensional
profiling system (for example, as disclosed in U.S Patent No.
5193120) but such a system is prohibitively slow for machine
vision inspection purposes. The problem has been solved using
high speed acquisition system and a triangulated laser beam.
Various solutions have been implemented to address each issue
with different levels of success. All prior-art inspection
systems require a high speed area scan camera and a laser.
The laser line obtained from a beam shaper or triangulation
using optical assemblies is made incident on the inspection
surface (U.S. Patent No. 5054918). A camera scans the laser
profile as the object under inspection is made to pass under
the incident laser line. The distortion in the laser pattern
provides depth information which is subsequently compared
with a standard template for inspection purposes. Such a
solution can be adapted to cases where the item under
inspection moves rapidly. This is done by using high speed
area scan cameras. High speed area scan cameras are expensive
and demand huge data transfer or processing rates. Line scan
cameras overcome this disadvantage since a lot of unnecessary
information is removed from transfer or processing. However,
with line scan cameras, the surface needs to be flat so that
depth artifacts can be mapped to ridges that represent
embossments. Many scenarios of inspection of such
embossments involve cases where the surface of the inspected

item is irregular and possesses significant variable
curvature.
Adaptive lens steering (U.S. Patent No. 5576948) has also
been used where the triangulated laser beam translates
adaptively over a test surface to compensate for angular
deflections. Such a system is not just complex, slow and
expensive, but cannot be rendered for a machine vision system
where space constraints prohibit such motion.
The object of the present invention is to provide an improved
system and method for detecting an irregularity on a three-
dimensional surface.
The above object is achieved by a method for detection of an
irregularity on a three-dimensional test surface, comprising:
- a first step, comprising synthesizing a curved beam profile
for each strip of a three-dimensional training surface, such
that the projection of a radiation beam having said curved
beam profile on said strip on said three-dimensional training
surface appears as a straight line from the position of a
training camera unit, and storing synthesized curved beam
profiles for successive strips of said three-dimensional
training surface, and
- a second step, comprising dynamically projecting a test
radiation beam on successive strips of said three-dimensional
test surface, wherein the test radiation beam projected on
each strip of said successive strips of said three-
dimensional test surface has a curved radiation beam profile
as stored in said first step for a corresponding strip of
said three-dimensional training surface, said second step
further comprising capturing a series of images of the
projections of the test radiation beam on successive strips
of said three-dimensional test surface by a line-scan camera,
each of said projections appearing as a straight line from
the position of said line scan camera, wherein an
irregularity on said three-dimensional test surface is

identified based upon a drop in intensity in one or more
images captured by said line scan camera.
The above object is further achieved by a system for
detection of an irregularity on a three-dimensional test
surface, comprising:
- means for synthesizing a curved beam profile for each strip
of a three-dimensional training surface, such that the
projection of a radiation beam having said curved beam
profile on said strip on said three-dimensional training
surface appears as a straight line from the position of a
training camera unit, and for storing synthesized curved beam
profiles for successive strips of said three-dimensional
training surface,
- dynamic beam projection means adapted for dynamically
projecting a test radiation beam on successive strips of said
three-dimensional test surface, wherein the test radiation
beam projected on each strip of said successive strips of
said three-dimensional test surface has a curved radiation
beam profile as stored for a corresponding strip of said
three-dimensional training surface,
- a line scan camera for capturing a series of images of the
projections of the test radiation beam on successive strips
of said three-dimensional test surface, each of said
projections appearing as a straight line from the position of
said line scan camera, and
- means for identifying an irregularity on said three-
dimensional test surface based upon a drop in intensity in
one or more images captured by said line scan camera.
The underlying idea of the present invention is to synthesize
a curve which on incident on a section of the test surface
appears as a straight line when seen from the camera. Once
the line scan camera sees the straight line, the synthesized
curve now represents information about the surface. This
process is repeated for different strips on the surface and
combined to form a representative shape descriptor for the
particular surface. The present invention thus obviates the

aforementioned problems of the prior art and simultaneously
provide a system which provides high speed, efficiency (in
terms of information transfer and processing) and cost
effectiveness.
In one embodiment, synthesizing a curved beam profile for
each strip of the three-dimensional training surface further
comprises:
- projecting a training radiation beam having an initial beam
profile on that strip,
- capturing an image of a projection of said training
radiation beam on said strip by said training camera unit,
and obtaining a resultant profile from the image of the
projection of said training radiation beam on said strip, and
- iteratively modifying the beam profile of the training
radiation beam to arrive at an inverse curve that defines a
final curved beam profile for said training radiation beam
for which the resultant profile obtained from said image of
the projection of said training radiation beam is a straight
line, based upon a feedback relationship between the beam
profile of the training radiation beam and the resultant
profile obtained from the image of the projection of the
training radiation beam on said strip.
This involves mathematical computations that can
advantageously be carried out with appropriate computer
software.
In a further embodiment, said training camera unit comprises
an area scan camera. In an alternate embodiment, said
training camera unit comprises a line scan camera, wherein
obtaining a resultant profile from the image of the
projection of said training radiation beam on said strip
further comprises:
- obtaining an intermediate profile from the image of the
projection of said training radiation beam on said strip
captured by said line scan camera, and
- spline fitting discontinuous portions of said intermediate
profile to obtain a continuous resultant profile.

In one embodiment, said dynamic beam projection means
includes a programmable liquid crystal display (LCD)
projector.
In one embodiment, said training radiation beam and said test
radiation beam comprise laser beams. Using a laser beam is
useful in inspection of irregularities dark surfaces having
very little contrast in intensity.
In an alternate embodiment, said training radiation beam and
said test radiation beam comprise x-ray beams. An x-ray beam
has lower wavelength and is useful in measuring surface
irregularities of extremely small dimensions.
The present invention is further described hereinafter with
reference to illustrated embodiments shown in the
accompanying drawings, in which:
FIG 1 is a schematic diagram of a system for detecting an
irregularity in a three-dimensional test object,
FIG 2 is a schematic drawing illustrating successive
deformations of the training radiation beam profile and the
corresponding resultant profile of the projection of the beam
in the image captured by the training camera unit,
FIG 3 is a flowchart illustrating a method for training
according to one embodiment of the present invention,
FIGS 4A, 4B and 4C respectively illustrate successive
deformations of the training radiation beam profile, the
corresponding intermediate profiles of the projection of the
beam as seen by a line scan cameras and resultant profiles
after spline fitting of discontinuous portions in the
intermediate profiles

FIG 5 is a flowchart illustrating a method for testing
according to one embodiment of the present invention, and
FIG 6 is a schematic diagram of an exemplary arrangement of a
line scan camera for testing.
Embodiments of the present invention provide an efficient and
low cost surface matching system for machine vision
inspection. The surface can be arbitrary, having no
regularity constraints. In the training phase the machine
would learn the surface (through a training surface that has
a generally similar profile to the test surface) to aid
deployment and accommodate any 'on the field' changes in a
realistic time frame.
The proposed system integrates a dynamic light projection
means such as a programmable liquid crystal display (LCD)
projector, a line scan camera, an area scan camera and an
adaptive beam reshaping algorithm. The said integration is
done in a manner to benefit inspection at high speeds, done
at low cost to accommodate surface inspection and detect
embossments. Conventional systems project a straight light
sheet, capture the profiles using area scan cameras and use
them for surface inspection. The drawback of such systems is
the unnecessary time consumed for acquisition of the whole
image and high data transfer rates. Such drawbacks decrease
the speed of execution and increase the cost of the system.
The proposed system is conceptualized inverse to the above
paradigm. The underlying idea here is to synthesize a curve
which on incident on a section of the surface appears as a
straight line when seen from the camera. Once the line scan
camera sees the straight line, the synthesized curve now
represents information about the surface. This process is
repeated for different strips on the surface and combined to
form a representative shape descriptor for the particular
surface. The purpose of the dynamic beam projection means is
to iteratively structure the incident light in a feedback
loop with the training camera unit.

The proposed invention would thus provide a machine vision
system that can map, detect surfaces, inspect embossments and
store the surface data in a manner that can be later on
compared for inspection purposes is developed.
Referring to FIG 1, a system 10 is shown for detecting an
irregularity on a three-dimensional surface, which may be any
arbitrarily curved surface. The first step of training uses a
training surface 14, which has a generally similar surface
profile as the test surface 15. The training step further
uses means 12 for synthesizing the' curved radiation beam
profiles along with a training camera unit 30, which may
include an area scan camera and/or a line scan camera. The
means 12 includes a dynamic beam projection means 16, for
example an LCD projector, which is coupled to a computing
device 18 (for example, a PC) that executes the algorithm for
synthesizing of the curve beam profiles for each of
successive strips of the three-dimensional training surface
14. During training, the LCD projector 16 projects a
radiation, beam 20 (in this case, a light beam) on a strip of
the training surface 14. The training surface 14 is a three-
dimensional surface, and hence, if the radiation beam 20 has
a beam profile 22 (i.e, beam cross-section) in the form of a
straight line, the projection 26 of the beam 20 on the strip
of the 3D training surface 14 would appear as a curved line
to the training camera unit 30. The idea of the present
invention is to modify or re-shape the beam profile of the
beam 20 to a curved beam profile 34, such that when the beam
having this curved beam profile 34 is projected on that strip
of the 3D training surface 14, the projection 28 as seen from
the position of the training camera unit 30 is a straight
line. The curved beam profile 34 is stored. Similarly, curved
beam profiles are synthesized and stored for successive
strips of the 3D training surface 14.
In one embodiment, the training camera unit 30 comprises two
cameras, namely an area scan camera and a line scan camera.

The area scan camera is required at the first step only. In
the second step (test step, for inspection and embossment
detection, only the line scan camera is required. This is
ensures the highest speed at inspection time, where real-time
operations are performed. During training, the centre of the
area scan camera is set up such that the line strip image
acquired by the line scan camera coincides with the centre
row of the area scan camera.
An exemplary process of synthesizing curved beam profiles for
each strip of successive strips of the training surface is
now discussed. A training radiation beam 20 having an initial
beam profile is projected on to the strip from the LCD
projector 16. An image of the projection of the beam 20 is
captured by the area scan camera in the training camera unit
30, and a resultant profile obtained from the image of that
projection as captured by the area scan camera. The beam
profile of the radiation beam 20 is then iteratively modified
using feedback based mechanism between the LCD projector 16
and the area scan camera of the training camera unit, which
reshapes the projected radiation beam 20 from the LCD
projector 16 such that the image of the projection of the
radiation beam as captured by the area scan camera has a
resultant profile which is a straight line. The beam profile
projected by the LCD projector which gives a resultant
profile (captured by the area scan camera) as a straight line
is referred to as an inverse curve profile for that
particular strip of the 3D training surface.
FIG 2 shows the successive deformations of the beam profiles
vky (x) projected by the LCD projector and the corresponding
resultant profiles fky (x) seen by the area scan camera for a
particular strip of the 3D training surface as the iterations
progress. The arrow 40 indicates the direction of iterations.
The iterations stop when the area scan camera sees a
sufficiently straight resultant profile. This deformed curve
beam profile is stored as a template for a particular strip
of the 3D training surface. The entire surface profiling is

completed when all strips are traversed and their
corresponding inverse curve beam profiles are stored.
FIG 3 is a flowchart showing an exemplary method 50 of
training the proposed system, using a beam re-shaping
algorithm. The method 50 begins at block 52 when a target
strip is selected on the 3D surface. Let 'y' be a strip in
consideration at a certain point of time. The strip at 'y'
denotes the position along the scanning axis. For example, if
the object is placed on a conveyor, as the object travels,
the conveyor belt movement represents the scanning axis; as
the machine part traverses, the 'y' value (or position)
increments. The iteration carried for each 'y' is as follows.
The iterative beam deformation algorithm is carried out as
explained in FIG 2. The objective profile is what is sought
after. In this case, the objective function is Φy(x) = h/2 ,
where 'h' is the height in pixels of the area scan camera.
At block 54, a radiation beam having an initial beam profile
is projected on to the selected strip of the training
surface. The initial beam profile is v°y (x), where 'x' runs
along the line of pixels, represented by the line scan camera
(for a standard line scan camera available, x would be a
vector from 1:640 or 1:1024). The centre row of the area scan
camera sees exactly at the same location and orientation as
the line scan camera. At block 56, the resultant profile of
the projection of the beam on the training surface is
captured by the area scan camera. The area scan camera now
sees the resultant profile on the surface as f°y (x). This
information is fed to the iterative beam re-shaping algorithm
(block 58).
In the iterative algorithm (block 58), the new profile to be
fed to the LCD projector is iterated by
The profile v1y (x) when projected on
to the training surface would be seen by the area scan camera
as f1y (x). This iteration is carried out for every time step


minimized. The parameter 'k' is a relaxation parameter
enhancing the convergence rates and incorporates the mapping
between the physical space and the image plane.
Next, at block 60, the synthesized inverse curve beam profile
vfinaly (x) is stored. At block 62, the next adjacent strip of
the training surface is selected as the target strip and the
above steps (blocks 52 to 60) are repeated. The method 50
ends at block 64 by storing all the inverse curve beam
profiles that are synthesized.
In an alternate embodiment, instead of a combination of an
area scan camera and a line scan camera, the training camera
unit may simply comprise a line scan camera. During training,
a line scan camera would not be able to image complete
profile of the projection of the radiation beam on a given
strip of the training surface, and would see only a part of
the profile. Herein an intermediate profile is obtained from
the image of the projection of the training beam as captured
by the line scan camera, and then, discontinuous portions of
this intermediate profile are spline fitted to obtain the
resultant profile. This is explained referring to FIGS 4A, 4B
and 4C. FIG 4A shows the deformation of the beam profile
vky(x) along a direction of iteration indicated by the arrow
70. FIG 4B shows the corresponding intermediate profiles 72
from the images of the projection of the beam having the
corresponding beam profiles as shown in FIG 4A. As can be
seen, the intermediate profiles are discontinuous, since the
line scan camera can capture only portions of the projection
of the beam on a given strip of the training surface. FIG 4C
shows the resultant profiles fky(x) obtained by spine fitting
of discontinuous portions in the intermediate profiles shown
in FIG 4B. The beam-reshaping algorithm is carried out based
on a feedback relationship between the beam profile vky(x)
shown in FIG 4A and the resultant profile fky(x) obtained
from the line scan camera as shown in FIG 4C.

Referring back to FIG 1, in the testing step, the 3D training
surface 14 is replaced by a 3D test surface 15, on which the
irregularity has to be detected. Herein, the LCD projector 16
dynamically projects a test radiation beam 21 on successive
strips of the 3D test surface 15. The test radiation beam 21
that projected on each strip of said successive strips of the
test surface 15 has a curved radiation beam profile 34 as
stored in the training step for a corresponding strip the
training surface 15. A line scan camera 31 captures a series
of images of the projections 29 of the test radiation beam 21
on successive strips of the 3D test surface 15, each of the
projections 29 appearing as a straight line from the position
of said line scan camera 31. Means 32 are provided for
identifying an irregularity on the test surface 15 based upon
a drop in intensity in one or more ,images captured by said
line scan camera 31.
FIG 5 is a flowchart illustrating an exemplary method 8 0 for
testing. At block 82, a particular strip 'y' of the test
surface is selected. As the 3D test surface is translated
along the scanning axis, the synthesized profile vfinaly (x)
for any strip 'y' is loaded into the LCD projector and
projected onto the 3D test surface. For the corresponding
strip on the test surface, the projection ffinaly (x) of the
beam will appear as a straight line as imaged by the line
scan camera (block 84). Block 86 includes identifying any
changes in the test surface profile based on drop in
intensity as seen by the line scan camera or breakages in the
line. At block 88, this information on irregularity is
stored. Next, at block 90, the next adjacent strip is
selected as the next strip until all the desired strips are
captured. Finally, at block 92, a pattern of irregularity is
obtained after scanning successive strips of the test
surface, and this test pattern is now converted to an error
profile (in the case of surface inspection) or embossment
information (in the case of embossment inspection). The
information is ensembled and a final decision is made for

pass or fail (inspection) or sorting (embossment based
character reading for sorting).
FIG 6 shows an exemplary embodiment illustrating the
mechanics of the line scan camera 31 for the testing step. A
rotating mirror (rotatable from position 106 to position 108)
sweeps for a range of angles (represented by numerals 102 and
104) to enable vertical scan over a 3D test surface 15. The
projector 16 projects a radiation beam 21 at a particular
designated strip which is the focus strip as deflected by the
scanning mirror at any instant of time such that the line
scan sensor 31 is focused on the strip.
In alternate embodiments, instead of light beams, other forms
of radiation can be used in the training and testing steps.
For example, the training radiation beam and the test
radiation beam may comprise laser beams. Using a laser beam
is useful in inspection of irregularities dark surfaces
having very little contrast in intensity. Still alternately,
the training radiation beam and the test radiation beam may
comprise x-ray beams. X-ray beams have lower wavelength and
is useful in measuring surface irregularities of extremely
small dimensions.
The present invention is advantageous in a number of ways.
Firstly, re-shaping the radiation beam will remove any
referencing issues for depth variation search for any
arbitrarily shaped surfaces. Secondly, all training can be
done at high speeds. Further, using the line scan camera, the
testing phase will elapse very little time. Still further,
embossment information can be easily obtained for sorting and
identification purposes. Yet another advantageous feature is
that redundant data will not be transferred or processed,
hence the efficiency of the system increases, thereby
allowing the system to be ported to low end processing chips.
Moreover, the proposed system is invariant to ambient
illuminations. Finally, the system is low cost and can be
quickly deployed.

Summarizing, the present invention provides a system and a
method for detection of an irregularity on a three-
dimensional test surface. The proposed system comprises means
for synthesizing a curved beam profile for each strip of a
three-dimensional training surface, such that the projection
of a radiation beam having said curved beam profile on said
strip on said three-dimensional training surface appears as a
straight line from the position of a. training camera unit.
Synthesized curved beam profiles for successive strips of
said three-dimensional training surface are stored. The
proposed system further includes dynamic beam projection
means adapted for dynamically projecting a test radiation
beam on successive strips of said three-dimensional test
surface, wherein the test radiation beam projected on each
strip of said successive strips of said three-dimensional
test surface has a curved radiation beam profile as stored
for a corresponding strip of said three-dimensional training
surface. A line scan camera captures a series of images of
the projections of the test radiation beam on successive
strips of said three-dimensional test surface, each of said
projections appearing as a straight line from the position of
said line scan camera. The proposed system further includes
means for identifying an irregularity on said three-
dimensional test surface based upon a drop in intensity in
one or more images captured by said line scan camera.
Although the invention has been described with reference to
specific embodiments, this description is not meant to be
construed in a limiting sense. Various modifications of the
disclosed embodiments, as well as alternate embodiments of
the invention, will become apparent to persons skilled in the
art upon reference to the description of the invention. It is
therefore contemplated that such modifications can be made
without departing from the spirit or scope of the present
invention as defined.

We claim,
1. A method for detection of an irregularity on a three-
dimensional test surface (14), comprising:
- a first step, comprising synthesizing a curved beam
profile (34) for each strip of a three-dimensional
training surface (14), such that the projection (28)
of a radiation beam (20) having said curved beam
profile (34) on said strip on said three-dimensional
training surface (14) appears as a straight line
from the position of a training camera unit (30),
and storing synthesized curved beam profiles for
successive strips of said three-dimensional training
surface (14), and
- a second step, comprising dynamically projecting a
test radiation beam (21) on successive strips of
said three-dimensional test surface (15), wherein
the test radiation beam (21) projected on each
strip of said successive strips of said three-
dimensional test surface (15) has a curved
radiation beam profile (34) as stored in said first
step for a corresponding strip of said three-
dimensional training surface (15), said second step
further comprising capturing a series of images of
the projections (29) of the test radiation beam
(21) on successive strips of said three-dimensional
test surface (15) by a line-scan camera (31), each
of said projections appearing as a straight line
from the position of said line scan camera (31),
wherein an irregularity on said three-dimensional
test surface (15) is identified based upon a drop
in intensity in one or more images captured by said
line scan camera (31).
2. The method according to claim 1, wherein synthesizing a
curved beam profile for each strip of the three-
dimensional training surface further comprises:

- projecting a training radiation beam having an
initial beam profile on that strip,
- capturing an image of a projection of said training
radiation beam on said strip by said training camera
unit, and obtaining a resultant profile from the
image of the projection of said training radiation
beam on said strip, and
- iteratively modifying the beam profile of the
training radiation beam to arrive at an inverse
curve that defines a final curved beam profile for
said training radiation beam for which the resultant
profile obtained from said image of the projection
of said training radiation beam is a straight line,
based upon a feedback relationship between the beam
profile of the training radiation beam and the
resultant profile obtained from the image of. the
projection of the training radiation beam on said
strip.
3. The method according to claim 2, wherein said training
camera unit comprises an area scan camera.
4. The method according to claim 2, wherein said training
camera unit comprises a line scan camera, wherein
obtaining a resultant profile' from the image of the
projection of said training radiation beam on said
strip further comprises:
- obtaining an intermediate profile from the image of
the projection of said training radiation beam on
said strip captured by said line scan camera, and
- spline fitting discontinuous portions of said
intermediate profile to obtain a continuous
resultant profile.
5. A system (10) for detection of an irregularity on a
three-dimensional test surface (14), comprising:
- means (12) for synthesizing a curved beam profile
(34) for each strip of a three-dimensional training

surface (14), such that the projection (28) of a
radiation beam (20) having said curved beam profile
(34) on said strip on said three-dimensional
training surface (14) appears as a straight line
from the position of a training camera unit (30),
and for storing synthesized curved beam profiles for
successive strips of said three-dimensional training
surface (30),
- dynamic beam projection means (16) adapted for
dynamically projecting a test radiation beam (21)
on successive strips of said three-dimensional test
surface (15), wherein the test radiation beam (21)
projected on each strip of said successive strips
of said three-dimensional test surface (15) has a
curved radiation beam profile (34) as stored for a
corresponding strip of said three-dimensional
training surface (14),
- a line scan camera (31) for capturing a series of
images of the projections (29) of the test
radiation beam (21) on successive strips of said
three-dimensional test surface ..(15), each of said
projections (29) appearing as a straight line from
the position of said line scan camera (31), and
- means (32) for identifying an irregularity on said
three-dimensional test surface (15) based upon ,a
drop in intensity in one or more images captured by
said line scan camera (31).
6. The system according to claim 5, wherein said dynamic
beam projection means (16) includes a programmable
liquid crystal display (LCD) projector.
7. The system according to claim 5, wherein said training
radiation beam and said test radiation beam comprise
laser beams.

8. The system according to claim 5, wherein said training
radiation beam and said test radiation beam comprise
x-ray beams.
9. A system or method substantially as herein above
described in the specification with reference to the
accompanying drawings.

The present invention provides a system and a method for detection of an irregularity on a three-dimensional test surface. The proposed system (10) comprises means (12) for
synthesizing a curved beam profile (34) for each strip of a three-dimensional training surface (14), such that the projection (28) of a radiation beam (20) having said curved
beam profile (34) on said strip on said three-dimensional training surface (14) appears as a straight line from the position of a training camera unit (30). Synthesized curved
beam profiles for successive strips of said three-dimensional training surface are stored. The proposed system (10) further
comprises dynamic beam projection means (16) adapted for dynamically projecting a test radiation beam (21) on successive strips of said three-dimensional test surface
(15), wherein the test radiation beam projected (21) on each strip of said successive strips of said three-dimensional test surface (15) has a curved radiation beam profile (34) as
stored for a corresponding strip of said three-dimensional training surface. A line scan camera (31) captures a series of images of the projections (29) of the test radiation beam
(20) on successive strips of said three-dimensional test surface (14), each of said projections (29) appearing as a
straight line from the position of said line scan camera (30). The proposed system further includes means (32) for identifying an irregularity on said three-dimensional test
surface (15) based upon a drop in intensity in one or more images captured by said line scan camera (31).