Note: Descriptions are shown in the official language in which they were submitted.
CA 02235093 2007-05-02
METHOD AND APPARATUS FOR
IMPROVED IDENTIFICATION OF PROBABLE DEFECTS IN A WORKPIECE
Technical Field
This invention relates to a method and apparatus for improved identification
of probable
defects in a workpiece, and in particular to automated high speed linear x-ray
lumber grading.
Background of the Invention
It has become a serious consideration in the lumber industry to improve
grading of lumber
and therefore improve secondary breakdown decisions. By optimizing the
recovery of "good
wood" against a slate of desired products, the value of the lumber can be
increased. "Good wood"
refers to wood which meets prescribed criteria. For different uses, what is
considered "good
wood" can vary. For example, for fine furniture it may be unacceptable to have
any knots in the
wood. However, for furniture intended to have a more rustic appearance, a
certain number of
knots can in fact be desirable. In general though it is desirable to identify
certain "defects" in the
lumber and to locate them with respect to a spatial reference system. One
method of doing this is
to have a human visually inspect each piece of lumber prior to it being cut
into secondary boards.
This is slow and prone to error. Further, even if the defect is identified,
the information must still
somehow be communicated to a saw operator in a meaningful manner to allow the
defect to be
isolated, yet allow wood recovery to be optimized against a desired product
slate.
There have been some improvements in the area of grading lumber, for example
Lumber
Optimizer (US Patent 4,879,753 to Aune et al) using x-ray, Method of
Estimating the Strength of
Wood (US Patent 4,941,357 to Schajer) also using x-ray, Dielectric Sensor
Apparatus (US Patent
5,654,643 to Bechtel et al) Detector for Heterogeneous Materials (US Patent
5,585,732 to Steele et
al) also a dielectric sensing device, and Flaw Detection System Using
Microwaves (US Patent
4,514,680 to Heikkild et al) which uses microwaves to measure lumber flaws.
1
CA 02235093 2007-05-02
Defects include features such as knots, rot, splits, sap, holes, and the like.
Defects can be
further subclassified, for example a knot can be a sound knot or an unsound
knot. Most defects
have some attribute which allows them to be detected by automated scanning,
for example
reflective inspection (laser or gray-scale video) can detect stain and sap in
wood. Transmissive
inspection techniques (such as x-ray) can detect density variations, and thus
knots and rot and the
like. However, past efforts to identify defects automatically, particularly
with transmissive energy
inspection techniques, have met with limited success due to the natural
variation in wood density
across a board. This is due to the inherent randomness of wood properties,
being a naturally
occurring, non-homogeneous substance. Thus, what might appear to an automatic
density sensor
to be a large area of rot (low density) in a board, might in-fact just be an
area ofnaturally occurring
low density within the board.
Some past methods of automated lumber grading have had limited success in
identifying
defects. However, the techniques used often result in more good wood being cut
from the board
than is necessary. For example, prior methods might identify defects in a
board, but would bracket
them with zones thereby leaving a large section as waste and only a small
section as "good wood".
It is desirable if a method and apparatus could be provided which would
confine defects to a small
waste section, leaving a larger section as "good wood".
In addition to the difficulty of identifying defects and isolating them to a
reasonable area
within the board, variations in the equipment used to sense the board have
contributed to errors in
detecting and identifying defects. For example, x-ray scanners, used to
measure board density, are
inherently "noisy" with scatter, and also tend to "drift" in power output
during their warm-up
period of operation. Similarly, the sensors or detectors used to measure the x-
rays passing through
a board are prone to drift.
Therefore, what is needed is an improved method and apparatus for detecting
defects
within a board. Further, there is a need for improved methods of localizing
defects once they are
2
CA 02235093 2007-05-02
detected within a board to increase yield of "good wood" from the board. There
is also a need for
an improved x-ray board defect detecting apparatus which accounts for drift
and variation in the
source and detectors used in x-ray imaging of the board. Preferably, it is
desirable to have an x-ray
system that allows high speed and accurate lumber grading, improving the
location and
identification of defects in lumber, thereby improving yield from boards.
Summary of the Invention
A method for identifying probable locations of defects within a workpiece is
disclosed.
The method is particularly applicable to workpieces comprising wood, lumber,
or other media
wherein a natural variation in a property such as density is found in the
medium which can obscure
: defects which are identifiable by the same naturally varying characteristic.
The method comprises
the following steps: providing a workpiece to be surveyed for probable
locations of defects,
wherein the workpiece is capable of being identified by selected regions;
providing an energy
source which can impinge energy on the workpiece to produce a resulting energy
signal indicative
of a physical property of the workpiece; and providing a plurality of
detectors capable of receiving
at least a portion of the energy signal which is produced by the workpiece and
generating a
detector signal in response to the energy signal. The method includes the step
of positioning the
workpiece relative to the energy source and the detectors to allow the
detectors to receive the
resulting energy signals. In the method the workpiece and the detectors are
moved relative to one
another while the energy from the energy source is impinged onto ~,he
workpiece to produce a
series of detector signals which are representative of the physical property
of the workpiece at
different spatial positions along the workpiece. The detector work signals are
stored in a memory
device in a manner such that the signals are identifiable by the associated
spatial position on the
workpiece.
According to a first embodiment of the present invention there is disclosed
method for
inspecting a workpiece to determine the probable existence'and locations of
defects within the
workpiece. The method comprises:
3
CA 02235093 2007-05-02
surveying the workpiece with an energy source to generate a response map of
resulting
signals, the signals being identifiable with locations within the workpiece,
sorting the response map into zones,
determining, for each zone, a bulk average of those signals associated with
the zone,
determining, for each zone, a moving long term average for localized positions
within the
zone, wherein the moving long term average is a function of the bulk average
of those
signals associated with the zone and the average of the signals associated
with the
localized position,
establishing a moving defect range for each zone, the moving defect range
being within a
predetermined percent of the moving long term average for the zone; and
for each zone, identifying those signais which are beyond the moving defect
range. Such
signals are indicative of the probable existence of.a defect in the workpiece
at the location
associated with the identified signal.
The method may furkher comprise the step of smoothing the signals prior to
determining
the bulk average of the signals for the zones to remove signal noise from the
signals. The method
may fizrtlier comprise, for those identified signals which are beyond the
moving defect range, the
step of interpolating those signals which are within a predetermined distance
of one another and
are in adjacent zones. The method may further comprise interpolating the
moving defect range
between adjacent zones for corresponding positions within the zones.
According to a further embodiment of the present invention there is disclosed
a method for
identifying probable locations of defects within a workpiece. The method
comprises:
4
CA 02235093 2007-05-02
providing a workpiece having a region to be surveyed for probable locations of
defects.
The region having an initial section and a terminal section, and is
characterized by
intermediate sections subsequent to the initial section and prior to the
terminal section,
exposing the initial section to an energy source of sufficient intensity that
at least a portion
of the energy penetrates the workpiece,
receiving at a plurality of detectors received energy comprising at least a
portion of the
energy which penetrates the workpiece at the initial section, wherein the
detectors generate
a signal in response to the energy received thereat,
for at least some of the intermediate sections of the workpiece, repeating the
steps of
exposing the intermediate section to the energy source and receiving at the
plurality of
detectors the received energy to generate a plurality of received energy sets
corresponding
to the exposed intermediate sections of the workpiece. Each energy set has a
section
average value corresponding to the average of the signals generated for the
energy set by at
least some of the detectors, and each signal in ari energy set corresponds to
a spatial
position on the workpiece.
Determining an average of at least some of the signals generated for at least
some of
sections to establish an overall workpiece average value,
establishing defect thresholds, the defect thresholds being a predetermined
percent greater
than and less than the overall workpiece average value,
determining a moving workpiece average value for at least some of the sections
of the
workpiece. The moving workpiece average value for a section is a function of
the moving
workpiece average value for a prior workpiece section and the section average
value,
5
CA 02235093 2007-05-02
wherein the section average value for the initial section is a function of the
overall
workpiece average value and the initial section average value,
determining moving defect thresholds for at least some of the sections of the
workpiece,
the moving defect thresholds for a section being the predetermined percent
greater than
and less than the moving workpiece average value for the section to produce a
threshold
range about the moving workpiece average for the section; and
for at least some of the sections, identifying those spatial positions on the
workpiece
having corresponding signals which are beyond the threshold range, the
identified spatial
positions corresponding to probable locations of defects.
According to a further embodiment of the present invention there is a method
for
identifying probable locations of defects within a workpiece. The method
provided comprises:
providing a workpiece having a region to be surveyed for probable locations of
defects, the
region having an initial position and a terminal position, the region being
characterized by
a plurality of zones, each zone being defined by a section of the region
between the initial
position and the terminal position,
providing an energy source of sufficient intensity that at least a portion of
the energy can
penetrate the workpiece,
providing a plurality of detectors in an essentially aligned configuration,
the detectors
being capable of receiving the portion of the energy that can penetrate the
workpiece,
positioning the workpiece proximate the detectors such that the zones are
essentially
perpendicular to the essentially aligned detectors, and such that workpiece is
disposed
between the energy source and the detectors,
6
CA 02235093 2007-05-02
moving the workpiece relative to the detectors from the initial position to
intermediate
positions subsequent to the initial position and prior to the terminal
position, and to the
terminal position, while exposing at least some of the region to the energy
source,
for a plurality of positions of the region, receiving at the detectors
received energy
comprising at least a portion of the energy which penetrates the workpiece,
wherein the
detectors generate signals in response to the energy received thereat, each
the signal being
representative of a physical characteristic present at a distinct spatial
position on the
workpiece,
storing the signals in a memory device in a manner such that the signals are
identified by
zone and by the position within the zone,
determining an average of at least some of the signals for at least some of
the zones to
generate overall zone signal averages for the at least some zones,
determining a moving zone average for at least some of the zones of the
workpiece, the
moving zone average for a position within a zone being a function of the
moving zone
average for a prior zone position and average of the signals for the position,
wherein the
position average value for the initial position is a function of the overall
zone signal
average and the average of the signals for the initial position,
determining moving defect thresholds for at least some of the zones, the
moving defect
thresholds for a zone being a predetermined percent greater than and less than
the moving
zone average for the zone, the moving defect thresholds defining a threshold
range beyond
which a signal can be attributable to a defect in the workpiece; and
7
CA 02235093 2007-05-02
for at least some of the zones, identifying at least some of the spatial
positions on the
workpiece having corresponding signals which are outside of the threshold
range, the
identified spatial positions corresponding to probable locations of defects.
The energy source may be an x-ray source. The method may further comprise the
steps of
exposing the detectors directly to the energy source to generate a gain,
isolating the detectors from
the energy source to generate an offset, and generating calibrated signals,
wherein the calibrated
signals are a function.of the gain and the offset. The method may also
comprise, smoothing at
least some of the calibrated signals by low-pass filtering the calibrated
signals signal noise from
the signals prior to. determining the moving zones average.
-The method may further comprise, prior to the step of identifying the spatial
positions on
the workpiece having corresponding signals which are outside of the defect
range, the steps of
interpolating the moving zone average between adjacent zones to generate
interpolated zone
averages, and using the interpolated zone averages.in place of the moving zone
averages.to
generate the moving defect thresholds. The moving zone average for zones adj
acent an edge of the
workpiece are interpolated with one another.
The signals which are above the threshold range may be indicative of areas of
high density
within the workpiece, and signals which are below the threshold range may be
indicative of areas
of low density within the workpiece. The method may further comprise the step
of grouping
candidate objects representative of potential defects within the workpiece.
The step of grouping
candidate objects comprises grouping identified spatial positions
corresponding to signals which
are above the threshold range and which are within a predetermined distance of
one another to
identify high density candidate obj ects; and grouping identified spatial
positions corresponding to
signals which are below the threshold range and which are within a
predetermined distance of one
another to identify low density candidate objects.
8
CA 02235093 2007-05-02
The signals may be analog signals. The method may and further comprise the
step of
converting the analog signals into digital signals. The method may further
comprise the step of,
for at least some of the candidate objects, generating a feature vector
containing characteristics of
the candidate objects. A candidate object may be characterized by candidate
object characteristics
comprising a maximum width, a maximum height, a perimeter length, an area, and
an aspect ratio
of the maximum width and the maximum height, of the candidate object. The
feature vector for a
candidate object may contain at least some of the candidate object
characteristics. The method
may further comprise the step of comparing the feature vectors against a rule
base to determine
whether a candidate object is likely to be an actual defect, and if so, the
type of defect.
The method may further comprise the steps of exposing the detectors directly
to the energy
source to generate a gain, isolating the detectors from the energy source to
generate an offset,
generating calibrated signals, wherein the calibrated signals are a function
of the signals, the gain
and the offset. The method may further comprise the step of smoothing at least
some of the
calibrated signals by low pass filtering the calibrated signals to signal
noise from the signals. The
method may further comprise the steps of disposing a material of known density
between the
energy source and the detectors, exposing the material to the energy source to
generate an
intermediate calibration signal, and using the . intermediate calibration
signal to calibrate the
signals.
The detector signals are processed in the following manner to produce a defect
map
indicative of the locations and types of probable defects within the
workpiece. An average is
determined for at least some of the detector signals to generate a bulk
average detector signal for
the overall workpiece. The bulk average detector signal is then adaptively
filtered for selected
regions on the workpiece to establish a localized average detector signal,
wherein the adaptive
filtering is configured to account for variance in the average of the detector
signals as compared to
the bulk average detector signal. Such a filter can also be known as a "moving
average" since the
average is adjusted relative to movement along the workpiece. After a moving
average for the
workpiece has been established, a defect threshold range is established. The
defect threshold range
9
CA 02235093 2007-05-02
is a function of the localized average detector signal (the moving average)
such that signals outside
of the threshold range can be attributed to a defect in the workpiece.
Finally, for at least some of
the regions, spatial positions along the workpiece are identified as having
corresponding detector
signals which are outside of the threshold range, those identified positions
corresponding to
probable locations of defects.
More particularly, the workpiece is divided into a plurality of zones which
are oriented in
the direction in which the workpiece is moved relative to the detectors.
Rather than establish a
bulk average for the entire workpiece, bulk averages are established zone by
zone. This reduces
the effect that a variance in a physical property such as density across the
workpiece will have on
the bulk average. Moving averages are established for each zone rather than
for the overall board.
Likewise, defect thresholds are established for each zone rather than for the
overall board. In the
step wherein the signals are compared to the threshold, defect thresholds
between adjacent zones
are interpolated to account for variations in physical property across the
workpiece. For those
zones which are adjacent to an edge of the board, the defect threshold is
extrapolated to the edge of
the workpiece.
The method can further include a step of smoothing the detector signals prior
to
establishing the long-term average. The step of smoothing the signals is
performed to reduce the
effects that noise and inherent signal randomness can have on the data.
Smoothing is
accomplished by filtering the raw data. The filter can comprise a 3 x 3 low
pass filter wherein the
gain and offset are determined by obtaining detector readings with the
detectors first fully exposed
to an energy source, and then isolated from the energy source to establish a
two-point gain/offset
correction (respectively).
_
The invention also includes an apparatus for perforrning the above-described
method. The
apparatus comprises an energy source to impinge energy on a workpiece,
detectors configured to
detect energy signals resulting from the impinged energy on the workpiece, a
mechanism for
CA 02235093 2007-05-02
causing relative motion between the workpiece and the detectors, and a
computer having a
memory configured to store the detector signals and to perform the steps of
the described method.
According to a further embodiment of the present invention there is disclosed
an apparatus
for identifying defects within a workpiece. The apparatus comprises:
an energy source configured to impinge energy on a workpiece and produce an
energy
signal in response.thereto,
detectors configured to detect energy signals produced by a workpiece as a
result of energy
being impinged thereon by the energy source, and produce a detector signal in
response
thereto,
a mechanism for causing relative motion between a workpiece and the detectors,
a computer having a memory configured to store detector signals, the computer
being
configured to perform the steps of:
categorizing stored detector signals based on their associated spatial
position on a
workpiece;
determining a bulk average of at least some stored detector signals to
approximate
an overall average detector value for a workpiece;
adaptively filtering the bulk average to establish a localized average
detector signal
for selected localized regions on the workpiece, wherein the localized regions
are
identifiable by spatial positions on a workpiece, and the adaptive filtering
being
configured to account for variance between an average of detector signals near
the
region and the bulk average;
11
CA 02235093 2007-05-02
establishing a defect threshold range as a fanction of the localized average
detector
signal such that signals outside of the threshold range can be attributable to
a
defect in a workpiece;
identifying at least some of the spatial positions on the workpiece having
corresponding detector signals which are outside of the threshold range, the
identified spatial positions corresponding to probable locations of defects;
and
storing the identified spatial positions in a computer readable medium,
wherein the computer is further configured to sort stored detector signals
into zones based
on their associated spatial position on a workpiece, and wherein the bulk
average and the
adaptive filtering are perfonned based on zones,
wherein the computer is further configured to interpolate the defect threshold
range
between zones.
According to a fin-ther embodiment of the present invention there is disclosed
an apparatus
for identifying defects within a workpiece. The apparatus comprises:
an energy source configured to impinge energy on a workpiece and produce an
energy
signal in response thereto,
detectors configured to detect energy signals produced by a workpiece as a
result of energy
being impinged thereon by the energy source, and produce a detector signal in
response
thereto,
a mechanism for causing relative motion between a workpiece and the detectors,
12
CA 02235093 2007-05-02
a computer having a memory configured to store detector signals, the computer
being
configured to perform the steps of:
categorizing stored detector signals based on their associated spatial
position on a
workpiece,
determining a bulk average of at least some stored detector signals to
approximate an overall average detector value for a workpiece,
adaptively filtering the bulk average to establish a localized average
detector
signal for selected localized regions on the workpiece, wherein the localized
regions are identifiable by spatial positions on a workpiece, and the adaptive
filtering being configured to account for variance between an average of
detector
signals near the region and the bulk average,
establishing a defect threshold range as a function of the localized average
detector signal such that signals outside of the threshold range can be
attributable
to a defect in a workpiece,
identifying at least some of the spatial positions on the workpiece having
corresponding detector signals which are outside of the threshold range, the
identified spatial positions corresponding to probable locations of defects;
and
storing the identified spatial positions in a computer readable medium,
wherein the computer is further configured to sort stored detector signals
into zones based
on their associated spatial position on a workpiece, and wherein the bulk
average and the
adaptive filtering are performed based on zones,
13
CA 02235093 2007-05-02
wherein the computer is further configured to interpolate the localized
average detector
signals between zones. "
According to a further embodiment of the present invention there is disclosed
an apparatus
for identifying defects within a workpiece. The apparatus comprises:
an energy source configured to impinge energy on a workpiece and produce an
energy
signal in response thereto,
detectors corifigured to detect energy signals produced by a workpiece as a
result of energy
being impinged thereon by the energy source, and produce a detector signal in
response
thereto,
a mechanism for causing relative motion between a workpiece and the detectors,
a computer having a memory configured to store detector signals, the computer
being
configured to perform the steps of
2 0 categorizing stored detector signals based on their associated spatial
position on a
workpiece,
determining a bulk average of at least some stored detector signals to
approximate
an overall average detector value for a workpiece,
adaptively filtering the bulk average to establish a localized average
detector signal
for selected localized regions on the workpiece, wherein the localized regions
are
identifiable by spatial positions on a workpiece, and the adaptive filtering
being
14
CA 02235093 2007-05-02
configured to account for variance between an average of detector signals near
the
region and the bulk average,
establishing a defect threshold range as a function of the localized average
detector
signal such that signals outside of the threshold range can be attributable to
a
defect in a workpiece,
identifying at least some of the spatial positions on the workpiece having
corresponding detector signals which are outside of the threshold range, the
identified spatial positions corresponding to probable locations of defects;
and
storing the identified spatial positions in a computer readable medium,
wherein the detectors are configured to produce analog detector signals, and
the apparatus
further comprises an analog to digital converter configured to convert analog
detector
signals to digital signals to be received by the computer memory.
According to a further embodiment of the present invention there is disclosed
an apparatus
for inspecting a workpiece to determine the probable existence and locations
of defects within the
workpiece. The apparatus comprises:
means for the workpiece with an energy source to generate a response map of
resulting
signals, the signals being identifiable with locations within the workpiece,
means for sorting the response map into zones,
means for determining, for each zone, a bulk average of those signals
associated with the
zone,
CA 02235093 2007-05-02
means for determining, for each zone, a moving long term average for localized
positions
within the zone, wherein the moving long term average is a function of the
bulk average of
those signals associated with the zone and the average of the signals
associated with the
localized position,
means for establishing a moving defect range for each zone, the moving defect
range being
within a predetermined percent of the moving long term average for the zone;
and
means for each zone for identifying those signals which are beyond the moving
defect
range, such signals being indicative of the probable existence of a defect in
the workpiec,e
at the location associated with the identified signal.
The invention further includes a computer readable medium having computer
executable
instructions for performing the described method.
The invention also includes an apparatus for projecting an energy beam onto a
workpiece.
The apparatus is preferably used to perform the above-described method. The
apparatus for
projecting an energy beam 'comprises an energy coriduit to which at a first
end is attached an
energy source. Energy from the energy source can exit at a second end of the
energy conduit. The
apparatus further includes a first aperture device for collimating energy from
the energy source
into the energy conduit. The apparatus also includes a plurality of
calibration shutters, each of the
calibration shutters being configured to be operably positionable within the
energy conduit to
intersect energy within the conduit and produce an attenuated energy beam
which can be impinged
upon a detector array.
The apparatus preferably further includes a second aperture device disposed
within the
energy conduit between the first aperture device and the plurality of
shutters. The apparatus can
further comprise a primary shutter disposed within the energy conduit between
the energy source
and the first aperture device to allow energy produced by the energy source to
be isolated from the
16
CA 02235093 2007-05-02
energy conduit. The apparatus may further comprise a plurality of detectors
held in selective
position a distance from the energy conduit outlet, and is further provided
with a detector aperture
device to focus energy from the energy source onto the detectors. Preferably,
the aperture devices
comprise a plate having a hole disposed therethrough to act as the aperture,
wherein the edge of
the hole is chamfered on the side of the plate which is facing the energy
source.
The invention provides other advantages which will be made clear in the
description of the
preferred embodiments.
Brief Description of the Drawings
Preferred embodiments of the invention are described below with reference to
the
following accompanying drawings.
The invention will be better understood by reference to drawings, wherein:
Figure 1 is a plan view of a board showing defects and how defects can be cut
away from
the board.
Figure 2 shows the board of Fig. 1 but with a preferred defect removal
schedule to increase
yield of "good wood" from the board.
Figure 3 is an isometric schematic diagram showing a modern lumber mill and
the system
for determining the presence of defects in a board and making optimiza.tion
decisions based
thereon.
Figure 4 is a side elevation view of a preferred embodiment of an x-ray
scanner which can
be used to detect defects in a workpiece.
17
CA 02235093 2007-05-02
Figure 5 is a schematic plan view of a board defect image map showing grouped
defects.
Figure 6 is a graph showing cross-board density variations.
Figure 7 is a plan schematic diagram of a board model segmented into zones for
localized
long term average density determination.
Figure 8 is a graph showing the moving long term average density variation in
a board, the
defect thresholds, and two defects.
Figure 9 is a flowchart for performing the method described herein.
Figure 10 is a block diagram of a variation of the system shown in Fig. 3,
further including
a plurality of sensing systems and a defect assembler.
Detailed Description of a Preferred Embodiments of the Invention
The method and apparatus for automated lumber grading, preferably high. speed
linear
x-ray lumber grading, provides improved identification and location of defects
in wood such as
2 0 rot, knots, sap pockets and voids, and provides higher recovery data
(yield of "good wood") for
cutting lumber.
Referring to Figure 1, prior methods of automated lumber grading might
identify the
defects 2, 3 and 4 in board 1, but would bracket them with zones 5 and 6,
leaving section 9 as
waste and only section 7 as "good wood". Turning now to Figure 2, the method
and apparatus of
the present application give the preferred results where the defects 2, 3 and
4 are confined to a
waste section 11, leaving section 8 as "good wood".
18
CA 02235093 2007-05-02
Overview of the system
Fig. 3 shows an overview of the system 10 for grading and handling graded
lumber,
including the cutting of boards, which is generally found in a sawmill or more
generally, a plant.
Lumber or a workpiece 1 generally advances into the plant from a planer 14 on
a feed belt 13
adjacent a fence 18. The workpiece 1 enters a scanner system 100, whereby the
workpiece 1 is
scanned to detect defects. Although in the present description the invention
is described
particularly with respect to an x-ray scanner, it is to be appreciated that
other types of scanners can
also work with the present invention. Although the terms "x-ray scanner can be
used when
generally describing the invention, this should not be considered as
precluding the use of other
types of scanners, particularly when describing the method and apparatus for
determining the
probable presence of defects in the workpiece. Other types of scanners
include, without by way of
limitation, a laser scanner, microwave scanners, optical scanners, and
computer aided tomogra.phy
scanners.
When a workpiece 1 enters the scanner station 100 its presence can be detected
by a barrier
photoeye, although the scanner itself can also be used for this purpose, as
will become apparent in
the detailed description below. The speed of the workpiece 1 can be tracked
using encoder wheels
or other means such as a doppler-effect device. Within the scanner 100, image
maps (also referred
to herein "frames") of the workpiece are collected from the scanner one frame
at a time as the
workpiece passes through the scanner. The frames are represented by data
collected from the
scanner which is recorded on a computer readable medium. When a frame has been
recorded it is
processed and evaluated in a computer system 200 for features and defects of
the workpiece.
Probable defects are identified, as well as their location on the workpiece,
in a data modeler. The
data modeler can comprise either the computer system 200, or the host computer
330. In a system
having a plurality of sensing systems, for example, an x-ray system 100, a
profile system 102, and
a visual scanning system 104, the data modeler can be replaced with a defect
assembler (DFA) 325
as shown in Fig. 10. The defect assembler combines the three models which
result from the three
sensor systems to produce a more accurate and complete representation of the
workpiece. In either
19
CA 02235093 2007-05-02
case, the resulting data, which is representative of a workpiece such as a
board, can properly be
described as a "virtual board", since it is a computer model of the board,
including those defects
which the scanner is-capable of identifying. This data representation of the
board can also be
properly called the "board model".
As indicated in Fig. 10, which shows in block diagram form a variation on the
configuration of the system shown in Fig. 3, the x-ray computer 200
communicates with the rest of
the system 10 through a communication system 300 which links together the
various components
of the system. The system 10 can further comprise a control subsystem 310
which can be used to
track workpieces as they move through the system, and can also be used to
control computer
controllable saws 400 and 410. The control subsystem 310 can also be used to
control a
programmable logic control (PLC) device 315 which controls field devices such
as workpiece
routers for routing the workpiece through the system. The system 10 can also
include a network
server or host 330 which can be used for, among other things, storing the
board model and other
data, as well as various computer.programs utilized by the various other
subsystems in the system,
on computer readable medium such as a hard drive 332 or a tape drive (not
shown). The network
server can support human/machine interfaces such as the graphical user
interfaces or termin.als
340. The network server can also perform other known functions such as
commuiiication with
remote devices through modem 350 (Fig. 3), and interfacing with other computer
systems in the
mill or other facilities through line 360. For example, the network server 330
can provide
production information to an accounting or shipping office. It should be
appreciated that
collection of processors in the overall system 10, including the DFA 325 when
used, can be
considered as a "computer", even though it is comprised of various components.
Alternately, each
of the components 'such as the control unit 310, the DFA 325, the PLC 315, the
x-ray computer
200, and the network server 330 can also properly be considered separately as
computers. While
the system 10 is shown and described herein in.particular configurations in
Figs. 3 and 10, it
should be appreciated by one skilled in the art that the necessary functions
of the invention, as
described fuller herein below, can be accomplished with any one of a variety
of system
configurations.
CA 02235093 2007-05-02
The system 10 further preferably includes an optimizer/decision processor 320,
known
herein as the "optimizer" for brevity. The optimizer is provided the board
model from x-ray
computer 200 or, when more than one sensor subsystems are employed, from the
defect assembler
325. Based on the board model, the optimizer can make decisions on the best
way to cut the
workpiece to optimize yield, consistent with the probable existence of defects
in the workpiece.
The optimizer is more preferably provided with a rule set.to compare against
the board model, the
rule set being a standard by which defects in the board model can be compared
to determine if the
defect identified in the board model rises to the level where the defect
should be excluded from
final product. More preferably, the optimizer is provided with a product slate
of desired products
and quantities. The optimizer can be configured to compare the board model
against the product
slate to determine if the board can be cut so as to generate a product which
is indicated as being
required by the product slate. Once a board fitting product criteria is
identified, the product slate
can be updated by the control system 300 to revise the quantity of product
needed to fill a product
order. All of the optimizer functions can be accomplished by a computer
program or programs.
Optimizers and optimizer programs are known in the art and will not be
described further herein.
Once'the optimizer has made a decision regarding a particular workpiece, the
decision is
preferably communicated to the programmable logic controller 315 for
execution. The PLC 315
preferably includes computer readable medium for storing computer programs.
Examples of
computer readable medium include, without limitation, ROM, RAM, a hard drive,
a diskette and
diskette drive, a CD ROM and CD drive, a tape and tape drive, and an EPROM.
The controller
315 can decide whether the board is to be cut or not, and if so, by which saw.
Typically a plant has
a plurality of computer controllable saws 400 and 410. Such ensures that the
sawing of the boards
do not become a bottleneck for throughput through the plant. Further,
depending on the type of cut
to be made (for example, rip or cross-cut), one saw can be preferably
configured over the other.
Boards can be routed to a selected saw via a conveyor interchange 500 which
can be for example a
pneumatically actuated conveyor interchange system actuated by the controller
315. The conveyor
interchanger 500 can move a workpiece from first conveyor 13 to second
conveyor 510. The
21
CA 02235093 2007-05-02
programmable logic controller 315, preferably in conjunction with the network
server 330, can
track each board as it progresses through the plant.
The system 10 is preferably provided with sensors (not shown) to allow the
tracking and
detection of workpieces as they move through the system. 'Workpieces can be
marked with bar
code or other marking as they exit the scanner 100, allowing the board to be
tracked as it moves
through the plant. For workpxeces such as lumber where it can be undesirable
to visibly mark
boards, bar coding can be accomplished by using ink which can only be read by
non-visible light
frequencies (such as ultraviolet). Methods and apparatus for tracking
workpieces as they move
through a plant .are known, and will not be described further herein.
In a preferred embodiment of the invention, a workpiece (lumber or a board) is
transported
through the x-ray scanner or imager 100 longitudinally at a relatively high
speed (approximately
3.6 meters per second (700 feet per minute) in one example). The conveyor or
feed belt 13 can be
angled at. approximately 20 degrees to the direction of travel of the board to
help insure that the
lumber keeps in constant contact with the fence 18 as the lumber moves through
the scanner 100.
- Such ensures that a corner reference point for the board remains along a
constant, known reference
line, being the fence 18. As boards enter the x-ray scanner station 100, the
feed belt 13 can briefly
separate at the scanner 100 so as to provide a better image quality,
especially when the scanner
scans wood species that can cause pitch build up on the feed belt 13.
The method for identifying probable types and locations of defects within a
workpiece will
now be described in detail, as well as a preferred embodiment of an x-ray
scanner.
Preferred embodiment of an x-ray scanner
Referring to Figure 4, a preferred embodiment of an x-ray scanner 100 is
shown. The x-ray
scanner 100 can be used with the method and apparatus for defect detection
described fuller
herein. The x-ray scanner 100 comprises a electron source 20 (which can be
described more
22
CA 02235093 2007-05-02
generally as an energy source) which projects electrons onto a target 21 which
in turn generates a
stream of photons or x-rays directed towards sensor array 40. Workpieces such
as boards 1 are
moved by the conveyor 13 into the path of the x-rays. X-rays penetrating the
workpiece 1 are
detected by detectors or sensors 44. The sensors produce a signal in response
thereto which can
be recorded or further processed as more fully described below.
The x-ray source 20 is preferably a ceramic x-ray tube assembly or similar
type where x-
rays are produced and directed by the source 20 towards the sensor array
below. A reflector 21 can
be incorporated to reduce the height of the apparatus 100. The source 20 and
reflector 21 are
preferably housed in a box 120 comprising lead to reduce spurious x-rays from
escaping. X-rays
can exit the box 120 through opening 122 and are projected into the energy
conduit 22 towards
sensors 44 The conduit 22 can also act as a frame to support the components of
the apparatus 100.
Directly below opening 122 is a primary shutter 24. Shutter 24 is preferably
fabricated from
tungsten or similar material resistive to long-term x-ray damage. The primary
shutter 24 is
provided with an opening 123 to allow x-rays to.pass through opening 122 and
into the energy
conduit 22. The primary shutter 24 can be translated from side to side by
primary shutter actuator
124 to cause the plate 125 to block the stream of x-rays exiting opening 122.
Actuator 124 can be a
solenoid or a pneumatic actuator.. Primary shutter 24 can be used to block the
x-ray source from
the sensors 44 for purposes such as calibration of the sensors or where it is
desirable to isolate the
x-ray source from the energy conduit 22 for human health reasons during
maintenance of the
apparatus 10 and the like. Another purpose for the primary shutter 24 is that
the source 20 life
expectancy can be lengthened if the power to the source remains on during
temporary shut downs
when there is no need to shut the power off to the source.
Immediately below the primary shutter 24 within the conduit 22 is a first
collimator or
aperture device 26. Collimator 26 is preferably fabricated from lead or other
similar material
impenetrable to x-rays. Disposed through collimator 26 is aperture 126. An
exemplary dimension
for aperture 126 is approximately 6 mm by 50 mm (1/4 of an inch by 2 inches)
to limit the x-rays
projected onto the linear detector array 40. First collimator 26 provides a
reduction in both scatter
23
CA 02235093 2007-05-02
and quantity of x-rays projected towards the workpiece. Below the first
collimator 26 within the
conduit 22 is a second collimator or aperture device 28. An_ exemplary
dimension between the
first and second aperture devices is approximately 45 cm (18"). Preferably,
the second aperture
device 28 can be translationally positioned by positioners 130 to allow
positioning of the x-ray
beam onto the sensors 44. Disposed through collimator 28 is aperture 128. . An
exemplary
dimension for aperture 128 is approximately 2.5 mm by 330 mm (1/10 of an inch
by 13 inches) to
project the x-rays onto a similarly dimensioned detector array 40. Preferably,
the top edge of each
collimator's aperture 126 and 128 has a chamfer. 127 and 129 (respectively) to
iinprove scatter
reduction of the x-ray beam onto the detector array 40.
The sensor array 40 (as well as the x-ray source 20) can be calibrated by
exposing the
sensors 44 to the x-rays without a workpiece in place, as shown in Figure 4.
The x-rays can also
be blocked by first shutter 24 to provide. a"black out" reading for
calibration. Calibration is
desirable due to drift in the intensity of the x-ray source as the source 20
warms up, drift in the
response of the detectors 44 due to a variety of factors, and differences in
the base response of
detectors 44 due to manufacturing variances and environmental conditions (dust
on a sensor, etc).
More preferably the apparatus is provided with a device allowing intermediate
calibration
levels beyond fully exposed and no exposure. This can be accomplished by the
calibration device
26 30 shown in Fig. 4. Disposed between the second aperture device 28 and the
detector array 40 is a
multi-stage calibration shuiter 30 having a plurality of calibrations shutters
32, 34 and 36 which
can be positioned within the energy conduit 22 to partially block x-rays from
the sensors 44. The
shutters 32, 34 and 36 are positionable within the energy conduit 22 by
respective calibration
shutter actuators 33, 35 and 37, which can be solenoids or pneumatic
actuators. Preferably,
actuation of the calibration shutters is computer controlled to allow frequent
calibration of the
apparatus during a warm-up period, and less frequent calibration thereafter.
The shutters. 32, 34
and 36 are fabricated from a material which allows passage of an attenuated
amount of energy
from the source 20 to the detectors 44. In this manner, intermediate
calibrations can be
accomplished. When three calibration shutters are used as shown, 7
combinations of intermediate
24
CA 02235093 2007-05-02
calibration can be accomplished. That is, shutters 32,34 and 36 can be used
separately, shutters 32
and 34, or 32 and 36, or 34 and 36 can be used in combination, or all three
shutters can be used in
combination. This provides for 9 levels of calibration overall.(including full-
open and fully
blocked) for the apparatus 10. Preferably, the shutters 32, 34 and 36 are
fabricated from a
homopolymer polyformaldehyde, known by its trade name DelrinTm or other
suitable material
which allows partial transmission of x-ray energy.
The apparatus 100 can be further provided with a detector collimator (not
shown)
positioned just below the path of the workpieces adjacent the sensors 44. This
assists in giving a
final reduction in both scatter and .quantity of x-rays projected onto the
sensor array 40. An
exemplary dimension for the aperture in the detector collimator is
approximately 2.5 mm by 330
mm (1/10 of an inch by 13 inches). An exemplary dimension for mounting of the
detector
collimator is about 6 mm (1/4 inch) directly above the sensors 44. This
detector collimator is
preferably fabricated from lead or tungsten inserted into aluminum or a
similar material.
The apparatus 100 can be further provided with sensor array 40. In one
embodiment, the
sensor array 40 consists of 128 individual sensors in asingle line where each
sensor is
approximately 2.5 mm (1/10 inch) by 2.5 mm. (1/10 inch). Each sensor can thus
be considered as
a"pixeP'. Each sensor is mounted in the path of the projected x-rays and in
the path of the
workpiece 1, where the workpiece is moved over the array 40 and through the
path of the x-rays by
conveyor 13 having. conveyor belt 16 and conveyor pulley 140. The conveyor
pulley 140 can
further comprise a measurement device 142 such as an encoder wheel for
determining the lineal
distance that the workpiece 1 has moved past the sensor array 40. The encoder
can also be used to
determine the arrival of a workpiece by detecting the leading edge of the
workpiece as it enters the
scanner. This event can then be used to trigger the scanning of the workpiece
by the sensor
subsystem. In a less preferred embodiment, the workpiece 1 can be temporarily
held stationary
while the apparatus 10 moves over the workpiece. The sensors typically
generate an analog signal
which can be communicated by signal line 52 to an analog-to-digital (A/D)
converter 50. Use of
the A/D converter allows the signals to be processed by a digital computer.
CA 02235093 2007-05-02
Method for determining the probable existence and location in a workpiece of
defects
Referring to Fig. 3, a workpiece or board 1 enters the x-ray scanner 100
longitudinally and
is sensed using the scanner 100, An x-ray scannei as described above is
preferably employed to
identify defects having density variations. High density defects or variations
in a board include
knots and compression wood. Low density defects or variations in a board
include rot, voids, and
large cracks. A detector or sensor array 40 (of Fig. 4) covering the width of
the workpiece is
preferably used. The signals from the resulting scan are preferably converted
to digital signals
using A/D converter 50, and are stored in a computer readable memory for
further processing.
The workpiece or board is preferably scanned in a longitudinal direction such
that the
major dimension of the board is perpendicular to the detector array and the
short dimension of the
board (or the width) is parallel to the detector array. The board is oriented
in coordinate axes with
one corner of the board designated as (x,y) = (0,0). Boards of thickness up to
4 inches may be
accommodated by the system described herein. Greater thicknesses can be
accommodated with
higher powerful energy sources. Preferably, the board is between 1 inch and 4
inches thick. Also
preferably, the board width is less than or equal to the width of the detector
array.
The detectors typically generate an analog signal in response to energy which
is transmitted
through the board. High density regions in the board will cause less energy to
be transmitted,
tesulting in a lower signal produced by a detector within the detector array.
The analog signals
produced by the detectors can be converted to a 12-bit digital signal using an
analog-to-digital
converter. The 12-bit signal can then be provided to a video frame assembler
which assembles the
data to a line scan configuration. This can be accomplished by an image
processing card and a
computer. Thus, data from the scan is collected and organized within a
computer memory in a
configuration such that the 12-bit digital signals are organized.with respect
to their spatial position
on the board in accordance with the (x,y) coordinate. system.
26
CA 02235093 2007-05-02
In a further embodiment of the invention, detector arrays may be oriented
radially about the
board at 120 degree intervals. Each detector array so provided is preferably
provided with its own
energy source. In this manner, data may be collected and assembled into a 3-
dimensional model of
the board.
Preferably the board is scanned every 2.5 mtn (1/10 inch) as the board passes
across the
detectors. Thus, each 0.1 inch of board scanned may be properly termed as a
"line scan". The
expression "line scan" will be used herein, and is understood to designate a
single transverse scan
of the board. Thus, a complete scan of the board is made up of a plurality of
line scans as the
board progresses longitudinally across the detector array.
When the,analog signals are converted into 12-bit binary signals, a signal
range or pixel
value of 0 to 4095 can be achieved. For an exemplary board, an average overall
board density of
approximately 1500 counts is typical. Signals which exceed the normal count by
a predefined
amount are indicative of areas of high energy transmissivity, indicating the
presence of a high
density defect such as a knot. Signals which are less than the average count
by a predefined
amount are the result of areas of low density within the board and are
indicative of defects such as
rot or a hole.
In one example, a sensor array of 128 sensors or pixels spaced approximately
2.5 mm apart
is used, for a total scanned width of approximately 32.5 cm or 12.8". Typical
board lengths from 4
feet to 20 feet (1.3 m to 6 m) or more can be accommodated by the system.
Since the minimum
board length commercially encountered is typically no less than 4 feet, a
"board" length of 4 feet is
used for data processing purposes. An actual workpiece can thus comprise
several "board models"
of 48 inches. In the following example, it will be assumed that the workpiece
is a board
approximately 12" wide, although it is to be appreciated that various widths
of a workpiece can be
accommodated by changing the number of sensors or the sensor spacing. Scanning
in the
longitudinal direction (the direction of travel of the workpiece) can be
performed every 2.5 mm
(1/10 inch), giving an image of the board of 480 pixels by 128 pixels. The
rate of scanning can be
27
CA 02235093 2007-05-02
performed at 1440 Hz. This provides a signal from the sensors that can be
converted to a 12 bit
digital signal, allowing a range of 0 to 4095 counts or different density
measurements. In the
preferred embodiment of the method, a "board" or frame comprises 48" of a
first frame plus 4" of
the next frame. This provides smooth data continuity from one frame to the
next.
The method for identifying the existence and location of probable defects
within the
workpiece is shown in flowchart form in Fig. 9. The block diagram flowchart
800 can be
implemented by reducing each described step to computer program steps to
produce a computer
program which can be stored on computer readable media and executed by a
computer.
In the first step 810 of the method, the workpiece is scanned by passing a
workpiece
between an energy source and sensors to detect energy transmitted through the
workpiece to
generate raw data which is preferably converted into 12 bit data by an analog
to digital converter to
produce a set of signals which will be processed to determine the presence and
location of
probable defects in the workpiece. Next, in step 820, the sensors or detectors
which are used to
sense energy transmitted through the board, and the x-ray source are
calibrated. Calibration is
performed by exposing the detectors directly to the energy source to establish
a gain value. The
detectors are then completely blocked from the energy source to establish the
response of the
detectors in the absence of any energy from the energy source which
establishes an offset. Then
the raw, data collected in step 810 is calibrated using a two-point gain
offset correction. In the
third step 830 the raw data collected in step 820 is smoothed using a low pass
spatial filter. More
preferably, the filter is a 3x3 low pass filter, using the signal established
in step 820. In step 835
probable defects, known as "candidate objects" are generated using an adaptive
thresholding filter,
preferably on a longitudinal zone basis, as will be more fully described
herein. The resulting
product of the candidate object generation step 835 is a series of high
density and low density
candidate objects and their corresponding location on the board or workpiece.
In step 840, the candidate objects are grouped together to "grow" probable
defect objects.
This is performed by grouping candidate objects identified in step 830 which
are within a pre-
28
CA 02235093 2007-05-02
determined distance of one another. In step 850, a candidate object feature
vector is produced.
The candidate object feature is extracted from the shape or geometric
candidate objects generated
in step 840, such as the height, width, aspect ratio, area, and perimeter
length of the candidate
object. In step 860, the features extracted from the candidate object 850 are
compared against a
rule-base to reject candidate objects which are unlikely to in fact be
defects. This step can also
properly be described as "defect filtering", since the objective is to remove
phantom defects. In
step.870, the remaining candidate objects are located dimensionally on the
workpiece by
converting their digital pixel space into spatial (x,y) coordinates which can
be provided to a defect
assembler and an optimizer for scheduling the cutting of the workpiece into
smaller pieces.
Each step of the method will now be described more fully.
Calibration
Due to natural variations in equipment such as the x-ray source and the
detectors, which
can be introduced by operating temperature and the like, it is preferable to
calibrate the detectors
based on variances in the energy source and based on their own variation.
Typically, calibration is
performed more*frequently early on while the equipment is warming up, and less
frequently later.
For example, early calibration frequencies of every 5 minutes and later
calibration frequencies of
every 30 minutes can be used effectively.
The image is calibrated using a two-point gain and offset approach by which
the value of
each pixel (i.e. each of the 128 detector pixels) is corrected based on a
predetermined fully
exposed detector reading (white level) and a fully blocked detector reading
(dark level). The
detector is fully exposed by taking readings from the detectors when no
workpiece or calibration
shutter is disposed between the x-ray source and the detector (this is also
referred to as the white
level correction factor value). The detector is fully blocked by isolating the
x-ray source from the
detectors. This is also referred to as the dark correction value and can also
be considered
29
CA 02235093 2007-05-02
background signal. Calibrated pixel data for real density calibration is
generated using the
following:
v(m)=[y(m)*g(m)]+o(m) (Equation 1)
where "m' is the pixel number, "y" is the measured value, "g" is the gain
("white level"), "o" is
the offset ("dark level"), and "v" is the calibrated pixel value.
Smoothing the Data
The calibrated image is first smoothed using a low pass convolution filter
(which is a
spatial smoothing filter) which reduces signal noise due to amplification and
detector speckle. The
use of this type of filter is preferable due to its noise removal properties
and its ease of
implementation. The calibrated image data is preferably smoothed using a 3x3
spatial low pass
filter, as follows:
v(.m; n)=EEa(151)y(m-k, n-1) (Equation 2)
where u.(k, 1)=1/9 are the filter weights.
a.UOV
Additional approaches to smoothing the data such as median filters or
morphological filters
(e.g. dilations and erosions) have also been verified as equally as effective.
Candidate Object Generation and Adaptive Thresholding
Turning to Fig. 6, an exemplary single line scan of a board is shown. The line
scan shows
the intensity of the count on the vertical axis and the width across the board
on the horizontal axis.
It is seen that moving across the board from a position y of 0 to a position y
of 12.8 that the count
drops from approximately 3200 to 1000. This difference may be the result of
natural variation in
the density of the wood and not indicative of the presence of a defect. An
example of a defect is
shown at y = 7 inches where the count intensity has risen from approximately
1700 to
approximately 2500. The cross-board variation shown in Fig. 6 can also exist
in the longitudinal
CA 02235093 2007-05-02
direction of the board. The method preferably is configured to account for the
natural cross-board
density variation as well as the longitudinal density variation in the board,
absent variations due to
defects.
Natural cross-board density variation is accounted for by dividing the board
into a plurality
of zones as shown in Fig. 7 wherein the workpiece I is divided into 6 zo.nes,
401 through 406.
Although the number of zones selected is not particularly pertinent, in one
example, 10 zones were
selected. For each zone then a long-term average in the longitudinal or x
direction is determined.
The long-term average is then adaptively filtered to obtain a moving average,
as will be described
further herein. Thus, segmenting the board into transverse zones in the y
direction accounts for
natural cross-board density variations, while adaptive filtering of a long-
term average in the x
direction accounts for natural density variations in the longitudinal
direction.
The adaptive filter adapts the digital signal data to the clear wood density
of the board. It is
preferable to establish this clear wood density.in a first section of the
board, for example, in the
first meter (36 inches) of the board, to initialize the filter. As shown in
Fig. 7, a"board' is defined
as being 480 line scans long. When each line scan is 0.1 inch long, this
corresponds to 48 inches
or 4 feet. In fact, the actual length of the board may be longer, but for the
data processing purposes
described herein, it is preferable to select a single dimension as a defined
"board length". The
board is divided into frames so that defects can be generated as the board
passes through the
scanner, and to limit processing of data to a maximum size (for example, 52
inches). An
additional benefit of dividing the board into frames is that such allows
variable length material to
be processed consistently. The frame length is preferably less than the
shortest anticipated input =
material length. Thus, in scanning the board, and in processing the data, a
4801ine scan board
length is used. Preferably, 40 line scans from the next "board" are
incorporated into each board
model to allow interpolation and extrapolation of the board model at its end
points. Thus, a board
model or "frame" of 520 pixels by 128 pixels results.
31
CA 02235093 2007-05-02
In the first step of the candidate object generation, the initial or starting
clear wood density
for each cross-board zone is established. It is preferable to establish this
initial value within the
first 36 inches of the board. The assumption is that the most frequently
occurring pixel value will
represent the clear wood density of the first several feet of the board. This
can be accomplished by
assembling a histogram of the pixel counts within the first 36 inches (360
line scans) of a zone and
selecting the most frequent pixel value in the histogram. Preferably, the
program for selecting the
most frequent pixel in the histogram looks only within a certain limit of
anticipated clear wood
density. If no peak value is found within the selected range, it is indicative
of a major defect in the
wood or an equipment problem, in which case the clear wood density value from
the previous
frame is used. The initial piece of the zone which is examined to determine an
average clear wood
density may be known as the region of interest (ROI). The most frequent value
in the histogram is
established as the initial or bulk long-term average (LTAo), and is used to
seed the adaptive
thresholding filter. LTAo is calculated for each zone across the board.
The next step is to process each line scan in the frame (520 pixels x 128
pixels in the
example) by first calculating a short-term average (STA) filter value for each
zone across the
board. The short-term average filter value for a zone is calculated by
determining the mean of
those pixels in the current line scan of the zone which fall within
predetermined upper and lower
bounds (upper and lower percentages) of the current long-term average value.
Once the short-term
average filter value has been determined for a line scan within a zone, the
long-term average is
adjusted based on a fraction of the short-term average. This fraction is
predetermined by defining
the length of the long-term tracking window. The resulting effect is that the
short-term filter value
for each line scan within a zone biases the long-term filter value either up
or down based on
whether the short-term filter value is above or below the long-term filter
value, respectively.
Various schemes can be employed in defining the relationship between the short-
term and long-
term filters. In essence, this can be viewed as a control system in which the
long-term filter is
attempting to track the slowly varying density map topology while the short-
term filter is
providing the feedback as to the direction and rate at which the long-term
filter should be moving.
The approach employed can be considered a linear feedback loop with hard
rejection points. Non-
32
CA 02235093 2007-05-02
linear or other approaches can be employed based on particular image
characteristics or desired
result.
For the technique employed,
LTA;+1= LTA; (f(sta;)),
where i is the current line scan, and f is a function. The short-term average
can be Iimited such
that it is not greater than nor less than 1.15LTA;, for example. This prevents
high and low readings
from defects artificially driving up or down the clear wood density average.
LTA;+t can be
properly described as the "long-term moving signal average" for a zone.
At this point, the long-term average value has been determined for each zone
for the
current line scan being processed. Next, high and low density candidate object
threshold levels are
established for each zone based on predetermined high and low density
percentages above and
below the clear wood density long term average (filter) value, respectively.
For instance, high
density objects are thresholded (separated) out whenever a pixel is determined
to have a value
greater than twenty percent (or other predetermined percentage) of the moving
clear wood density
long term average (filter) value (pixel) value. That is,
T;=LTA;(1 P),
where T is the candidate object threshold count, and P is <1 (preferably
0.5>P>0.1). For example,
if a 20% band is established as the threshold, P = 0.2, so T; = 0.8 LTA; to
1.2 LTA;.
The adaptive thresholding is used to separate high and low density candidate
objects from
the density map. Here the term "candidate object" refers to the fact the
object is a candidate (i.e. it
has potential to become part of a defect). Typically several "candidate
objects" are j oined to form
a candidate defect object. Candidate objects are separated using adaptive
thresholding and are
then joined to form candidate defect objects which are then classified as
belonging to a specific
defect type, as described below.
33
CA 02235093 2007-05-02
Linear interpolation is preferably performed between adjacent zone moving long-
term
average (filter) values in order to remove discontinuities between zones and
to better approximate
the varying cross-board density. Linear extrapolation is also preferably
performed beyond the mid-
point of the first and last zones to better handle the effects near the edge
of the board. A piecewise
linear interpolation/extrapolation approach was determined to be satisfactory
for determining high
and low density objects. A polynomial or similar curve fitting approach can
also be employed.
Each pixel in the line scan is then determined to either belong to a high
density object, a low
density object, or neither, based on the previously determined high and low
density threshold maps
(hereinafter referred to as. object(s)).
Referring to Fig. 8, a graphical representation of the moving long-tenn
average is shown.
Line 180 tracks the workpiece signal density (the actual instantaneous density
measured by the
sensors). The centerline 160 tracks the moving long-term average LTAi. Line
162 marks the lower
threshold limit TAL. Pixel values below or less than line 162 are candidates
for low density
defects. Line 164 tracks the upper threshold limit TAH. Pixel value`s above
line 164 are candidates
for high density defects. An example of a low density defect is shown at 166,
while an example of
a high density defect is shown at 168. Line 170 is the short-term average
(STA) of any given line
scan. It can be seen that as the short-term average line 170 moves up and
down, the long-term
average line 160 generally tracks the motion of the short-term average line
170. Also, during an
excursion (166 or 168) the short term average is not generated, atid the last
LTA is used until the
signal returns to within the threshold limits.
The resulting data model may properly be referred to as a density map, a
defect map, or a
candidate map.
The adaptive thresholding algorithm is summarized as follows:
- Initialize long-term pixel average for each cross board zone if this is the
first image
frame of the board.
34
CA 02235093 2007-05-02
rk = n,t(nm),
where rk is the kth gray level, nk is the number of pixels in the image with
that gray
level, nk(m,,,,) is the maximum value of nk exclusive of background pixel
values, and
k = 0,1,2,..., L - 1.
- Step line by line through the image frame and perform the following for each
cross
board zone:
- Calculate the line pixel average within the cross-board zone.
- Modify the zone long-term pixel average based on the line pixel average.
- Calculate the zone low and high density thresholds based on a percentage
of the long-term pixel average.
- Threshold line for low and high density objects (linear interpolation is
performed between zones).
The next step is to perform localized joining of similar candidate object
types. This is
performed using a region growing technique that merges like-candidate objects
(high density or
low density) that are within a predetermined neighborhood or distance of one
another. This is
performed in both the width and length directions. Typically several candidate
obj ects will form a
high or low density object grouping.
Like candidate (high or low density) objects are joined using local object
merging if the
distance between adjacent objects is less than a preselected distance. The
following is performed:
R; is merged with Rj if I R;(x,y) - Rj(x,y)15d
where R denotes a region.
CA 02235093 2007-05-02
Referring to Fig. 5, an example of local object merging is shown. A board
model 150 is
shown with identified defects. The board model may be considered as an image
map or a defect
map. Although the board model may be displayed graphically to a user, it can
also be stored as a
digital representation in a computer readable medium for further processing by
the method as
described herein. In the example shown, the candidate object generation has
identified potential
defects 420, 421, 422, 423, 427, 428, and 430. Prior methods of merging defect
objects might
attempt to merge defect objects 420, 427, 428, 421, 422 and 423 into a single
defect bounded by
boundary 426. In the preferred method of candidate object merging defects 420,
427 and 428 are
merged and surrounded by boundary 424. Defect object 421 may be added to the
defect bounded
by boundary 424 by increasing the boundary with boundary line 425.
Alternately, defect 421 may
be isolated to its own defect boundary. It is apparent tha.t bounding defects
by boundaries local to
the defects decreases the quantity of wood which may need to be removed as
waste product.
Likewise, prior methods of bounding defect 430 would produce a defect boundary
433 which
would result in loss of area 434. Preferably, defect 430 is bounded by
rectangles 431 and 432 thus
allowing area 434 to be salvaged as good wood.
The next step is to extract a feature vector from the candidate defect object
which will be
used to classify or validate the high or low density objects. Some of the
features extracted are the
object width, height, pixel area, and aspect ratio of the defect object.
Additional features such as
the object's average pixel value, elongation, orientation, second and third
moments, perimeter
length, etc. can also be utilized.
The following features are generated for each candidate object:
bounding box locations (X,,;,,, X,,., Yiõ Y.)
- pixel area (P,,,õ,)
- aspect ratio
36
CA 02235093 2007-05-02
The features from the feature vector are then used to either reject the
candidate defect
object as invalid, in which it is discarded, or to determine the type of high
or low density defect
detected. The classification approach taken is one of a rule base in which
several rules are
compared to the generated feature vector. For example, the following criteria
is preferably be
satisfied:
- minimum and maximum width and height criteria
Width: (yu- Ymin) ? W,nin and (ymax - ,Ymin) :S Wmx,
Height: (x. - xmil) _ Hmm and (x. - xmin) _< H.
- minimum and maximum area criteria:
Area: Parea ?Amin and Pw< Amax
- minimum and maximum aspect ratio criteria:
Aspect Ratio: (JCmcuc - zmin) / (Ymax - Ymtn) ~! R, and
(Ymax - ,Ymid / (xmax - xmin) ? R
More sophisticated approaches such as using a fuzzy logic and neural networks
can also be
used. Any candidate objects not rejected are probable defect types, such as
rot or knot, that have
been detected and classified from the candidate high and low density objects.
Defect objects are then converted from internal pixel unit space to external
measurement
unit space preferably in thousandths of inches (0.000"). In addition, length
(x) and width (y)
correction scale factors and offset are applied to correctly reference the
coordinate space to some
given datum reference. The following conversions are performed:
xExternal - xlnternal * xSecaleFactor
YExternal = (ylnternal * YSscaleFactor) + YO,Qset
In compliance with the statute, the invention has been described in language
more or less
specific as to structural and methodical features. It is to be understood,
however, that the invention
37
CA 02235093 2007-05-02
is not limited to the specific features shown and described, since the means
herein disclosed
comprise preferred forms of putting the invention into effect. The invention
is, therefore, claimed
in any of its forms or modifications within the proper scope of the appended
claims appropriately
interpreted in accordance with the doctrine of equivalents.
38
