PAPER PRESENTATION

ON

“3D MACHINE VISION SYSTEMS

AS
SHOP FLOOR
METROLOGY TOOL”

BY

SREEJA REDDY
III BTECH
ELECTRONICS AND COMMUNICATION ENGG
[email protected]

SREE VIDYANIKETHAN ENGINEERING COLLEGE

Sree Sainath Nagar, Chandragiri Mandal, Tirupati.
Chitoor Dist. A.P.
ABSTRACT years, moving from purely visualization
Machine vision refers to tools that generate attractive color
applications in which the PC pictures, to serious measurement tools.
automatically makes a decision based on These 3D systems include laser
visual input from a camera. Machine scanning, structured light, stereo
vision is a term typically used in viewing, and laser radar just to name a
industrial manufacturing, where few.
applications range from culling INTRODUCTION
blemished oranges from a conveyor belt Modern day durable goods
to saving lives by inspecting to ensure manufacturing have begun to embrace
that the correct drug capsule has been the concepts of digitization as a means to
placed in the package before the product improve productivity and quality.
is shipped to the pharmacy. Three Moving away from expensive hard
dimensional vision based measurement gages made for specific parts;
systems have made their presence into manufacturers are seeking the means to
production metrology applications, measure parts in a flexible manner, and
notably in the electronics field. capture the resulting measurements by
However, in the more traditional fields digital means. For higher volume parts,
of durable goods now dominated by hard such as from forging or fast machining
gauges and CMMs, 3D optical systems operations, speed of gauging is still an
has been hindered by perceptions and issue.
real limitations. This paper will review This is the area where machine
where 3D vision is today, and what vision based tools start to excel.
advances have been made to enable Machine vision in general has been used
more quantitative, shop floor metrology for everything from guiding the insertion
applications. The field of 3D machine of electronic chips on circuit boards to
vision is a less established field, but one inspecting bottles at several per second
that is actively growing today. Three in bottling lines. A natural extension of
dimensional vision based measurements machine vision inspection is to provide
have come a long way in the past few programmable measurements for
machined parts. In many applications, single line of points in the form
these measurements can be made in two of a cross section of the contour
dimensions for which there is an of interest,
established based of machine vision  Full-field sensors provide an X,
tools working in the sub-thousandth of Y, Z map of all the points in the
an inch range at multiple measurements scene, which must then be
per second. Each of these methods has analyzed down to the information
their strong points and weak points for a of interest.
given application. Each of these types of sensors has
The key performance parameters needed developed through technology which is
for durable good manufacturing include: suited to the application. In some cases,
• Resolution in the mil, and sub-mil the technology is capable of multiple
range modes of operation (finding point on a
• Speeds sufficient to complete all surface, or defining the full surface) as
measurements in a few seconds well, but this often stretches the
• Ability to look at a wide range of technology into a field overlapping other
surface types and finishes technologies. There has not to date been
This last point, the ability to look at a any single sensor in industrial
wide range of surface finishes has applications which does everything.
perhaps been the biggest limitation of DISCUSSION OF TECHNOLOGIES
3D machine vision technology. In many There are currently three basic
cases, the surface needs to be diffuse to approaches to three-dimensional
permit reflected light to be easily seen to machine vision:
achieve a good signal to noise ratio. • range finding including
Three dimensional optical sensors can structured lighting
perhaps be broken into a few basic • stereo or binocular vision,
types:- • gray scale or range finding
 Point scanning sensors measure methods.
only the specific points of TRIANGULATION
interest, typically in a serial The most popular commercial
fashion,Line sensors provide a versions of range finding use the
triangulation method where a beam of part. Depending on how random or
light is projected onto the object's directional the pointing of these micro-
surface at some angle and the image of mirrors may be, the apparent spot seen
this spot or line of light is viewed at on the surface will not be a direct
some other angle (see Figure 1). As the representation of the light beam as
object distance changes a spot of light on incident on the part surface. The effects
the surface will move along the surface that may be seen from a laser beam
by: reflecting off a rough surface include:
(change in spot position) = (change in • directional reflection due to
distance) x (tan(incident angle) + surface ridges
tan(viewing angle)) • skewing of the apparent light
distribution due to highlights
• expansion of the incident laser
spot due to micro surface piping
The result of this type of laser reflection
or "speckle" is a noisy signal from some
surfaces. In like manner, there can be a

Figure 1. A triangulation based system problem with translucent surfaces, as the

using a point of light to obtain distance. laser light will scatter through the

Good resolution of a few tens of medium and produce a false return

microns have been realized with these signal. For a laser based sensor, a

systems. Most triangulation gages today smooth, non-mirror like, opaque surface

use laser light. When a laser beam is produces the best results. An active

incident on an opaque, rough surface, the variation of restricting the view uses

micro-structure of the surface can act as synchronized scanning. In the

though it is made of a range of small synchronized scanning approach (see

mirrors, pointing in numerous directions. figure 2), both the laser beam and

These micro-mirrors may reflect the viewing point is scanned across the field.

light off in a particular direction, as In this manner, the detector only looks at

generally machine marks do, or may where the laser is going.

direct the light along the surface of the

 Reflections that do not go along the view
axis are not seen at all. The limitations
of this approach can be more time
consumed in seeking each point, and low
light collection to maintain high angle
separate. If as opposed to a single spot of

Figure 2. A synchronized scanning light, a line is projected onto the surface

system, limited the range of view. by imaging or by scanning a beam, as

A more extreme variation of the shown in Figure 1, the line will deform

synchronized scanning type of approach as it moves across a contoured surface as

is to use an active point seeking each point of the line move as described

triangulation system (see figure 3). With above. The effect is to provide an image

an active point seeking triangulation of a single profile of the part (see figure

system, a single point detector is limited 4). In applications requiring only a

to a very narrow view, at a constant profile measurement, these techniques

angle, which is then translated across the offer good speed of measurement.

path of the laser beam.

Figure 4. Laser radar sensor scanning a

Figure 3. Active triangulation system, box from a single view point

seeking laser point, providing extended The individual laser probes have

range. seen a near 10 fold improvement in

The detector now has the resolution in the past few years. The

advantage that it can resolve the application of such probes in the

distribution of light seen along that electronics industry has been a great

particular angle, and potentially decide benefit to making high speed electronics

on which signals are the correct ones. assembly possible. Scanning and fixed
triangulation systems have been used to The techniques described so far
contour large structures such as are based on some form of active
airplanes, logs or airfoil shapes, and illumination of the surface, and active
metal rolling mill flatness control. The sensing of that specific illumination.
large area systems primarily have used When people view an object, they
one or multiple lines of light to obtain a receive 3-D information based upon the
cross sectional profile at a time. In many principles of parallax and shadings seen
cases, these line sensors are connected on the surface.
with a CMM type device to extend the Stereo or binocular machine
working range of the sensor. The vision methods work on the same
resolutions of such systems need principle as human vision by obtaining
typically be less than a millimeter, and parallax information by viewing the
more typically is around 2.5 microns object from two different perspectives as
(0.0001 inch). Full-field structured light is shown in Figure 5. (as our two eyes
systems, based upon projected grids by do).
direct sensing of the grid or related to
moire are also commercially available.
The primary application of this type of
sensor has been the contouring of
continuously curved non-prismatic parts
such as turbine airfoils, sheet metal, clay
models, and human anatomy. Special
Figure 5. Two cameras see different
compact sensors for use on CMMs are
perspectives to provide stereo parallax
also available with this technology. Most
information.
of the applications of this technology has
The two views must be matched
been on applications requiring dense
up and then the difference in the position
data, but have also been engineered to
of a correlated feature gives the depth
enhance video data for the purpose of
information from the geometry of the
3D "comparator" type measurements on
two view points(similar to the
objects ranging from rolled sheet metal
triangulation methods). These methods
(for flatness) to car bodies.
can be full-field, thereby keeping the
data acquisition time to a minimum, and touch probes, there are certain types of
they do provide all the 2-D feature features or surfaces that optical 3D
information. In fact, stereo depends on methods can be expected to work good
the surface features to the extent that a on, and others where there may be
featureless, smooth surface (such as the problems. If has been pointed out that
turbine blade), can not be contoured shiny, but not specular surfaces have
directly with stereo views. offered one of the biggest challenges. In
Many machine vision systems like manner, when a surface changes
have a simple stereo matching from a shiny area to a dull, many sensors
capability, made for tracking individual may generate a bias error. In the simple
points in space, but not for providing full case of triangulation, the measurement is
surface contours as would be desired for based upon finding the centroid of a
reverse engineering applications. light spot of some finite size. If half that
Resolutions of a few millimeters at spot is on an area that reflects back to
certain points, and processing times of a the sensor well, and the other half is not,
few seconds has been typical for the the center of brightness of the spot will
current industrial applications. There not be the geometric center, but rather
have been dedicated hardware systems weighted toward the brighter region.
made, but stereo has not be able to Testing the sensor on edge and surface
provide the flexibility of scanning type transition features is a valuable first test
systems for CAD data generation to consider (such as in Figure 6).
applications.
APPLICATION TESTING
As has already been stated, the key
operational parameters needed for
production machine vision include
Figure 6. Measuring the effect across an
speed, resolution, and robustness
edge boundary.
especially to changing part surface
The next area of concern is the surface
conditions. Many systems that provide
texture itself. A surface with machining
the best resolution are not the fastest, so
marks has a texture which may scatter
a tradeoff must be made. Just as with
light into long lines that may confuse the
sensor. In a similar manner, if the Anytime one is trying to improve a
surface is translucent, the spot of light process, one encounters the challenge of
may spread out in an unpredictable demonstrating a capability against a tool
manner, again offsetting the spot center, with less capability. The very high
and hence the measurement made. speeds and data densities available from
Therefore, testing the sensor on a surface optical gages offer some significant
texture that matches the one to be advances. Yet trying to compare those
measured is important. A final important advantages against a gage which stakes
feature consideration for many optical everything on a half dozen points is a
gages is the slope of the surface. A difficult road. However, the comparisons
standard way to test such effects is to and qualification of the new process
scan a diffuse sphere and fit the data to must be done. Comparisons against
the known diameter (see Figure 7).At inadequate measurement tools can prove
some angle around the sphere, the very frustrating, so at least one
surface can be expected to lift off the independent means of comparison is
true surface, before the signal is lost desirable. The use of reference artifacts
entirely. As with any gage, comparison that can be verified by independent
of the optical gage against other means, preferable with primary standard
measurement tools provides valuable reference (e.g., the laser distance
information regarding confidence of the interferometer is the international
measurements. This is not always an standard of length measurement), is a
easy comparison, as the repeatability of valuable aid in these comparison.
many traditional gages may not be as APPLICATIONS
good as the optical gage. Machine vision systems can be used in
various fields for numerous purposes as
stated below:
• Industrial inspection (Inspecting
machine parts, Adaptive
inspection systems)
Figure 7.Measurement on a sphere
• Vision for Autonomous Vehicles
(Detecting obstructions,
 Exploring new surroundings, AV resolution with lower noise, and better
surveillance) light sources such as higher power laser
• Transport (Traffic monitoring, diodes well as higher brightness and
Aerial navigation, Transport resolution LCD projectors. The
Safety) consumer market largely influences all
• Surveillance(Intruder monitoring, of these technologies, which is currently
Number plate identification, a much bigger driver than any
Tracking people) manufacturing support system.

• Vision and Remote Sensing However, as system prices decrease and

(Land Management, Crop performance improves, there is a wide

classification, Surveying by range of new markets these systems will

satellite) likely address ranging from dentistry to

• Medical (Planning radiation home 3D pictures.

therapy, Chromosome analysis,

Merging medical images) REFERENCES
1. ASME Journal of Machine Design
2. "Machine Vision Based Gaging of

CONCLUSION Manufactured Parts," K. Harding, Proc.

As with any technology of this SME Gaging '95, Detroit, Nov. 1995.

nature, the performance changes with the 3. “The Promise of Machine Vision,” K.

component technology. The primary Harding, Optics and Photonics News, p.

advance that has made machine vision 30 May 1996 Optical Society of

systems feasible for shop floor gauging America

applications has been the speed up in 4. "Light Source Design for Machine

computing power. This has brought the Vision", Eric J. Sieczka and Kevin G.

processing times from 15 or 20 minutes Harding, SPIE Proceedings Vol. 1614,

on an expensive workstation to seconds Optics, Illumination and Image Sensing

on a standard PC. The other technologies for Machine Vision VI, Boston, Nov.

that are influencing performance today 1991.

include lower cost, digital cameras than

provide better light range and pixel
5. L. Bieman, "Three-Dimensional
Machine Vision," Photonics Spectra,
May 1988, p. 81.
6. E. L. Hall and C. A. McPherson,
"Three Dimensional Perception for
Robot Vision," SPIE Proceedings Vol.
442, p. 117 (1983).