CA1143455A - Edge detection apparatus - Google Patents

Abstract

ABSTRACT OF THE DISCLOSURE
A method of determining the precise location of an edge of a strip with individual elements of a radiation responsive sensor to an accuracy exceeding the pitch of the elements is disclosed. The strip is scanned at least in the region of the edge with the individual elements of the radiation responsive sensor. Each element generates a signal value associated therewith for each scan and the signal values for all the elements form a composite signal having discrete values representative of radiation received from the scanned region at respective locations corresponding to the individual elements of the sensor. A
region of interest where said edge is likely to lie is identified and the composite signal is analysed by defining a curve which when the values are plotted against the locations on a distance scale passes through the values to represent a radiation profile of the scanned region as received by the sensor. The point of maximum slope on the curve is then found and the corresponding position on the distance scale taken as representing the precise location of the edge.

Description

~3~

The present invention relates to a method of determining the precise location of an edge of a strip, particularly the edge of a strip being worked in a steel mill although it is applicable to any edge whose precise location is to be determined by means of a remote scanning device. The invention also relates to an apparatus for carrying out the method.
In the steel making industry it is necessary to know the precise width of a strip being rolled or otherwise worked. The strip is usually hot and its width may be determined by means of cameras scanning the strip. Canadian Patent No. lrlO3,018, issued June 16, 1981 to Peter Loose, describes an improved system for measuring the width of the strip where two cameras each scan the whole of the width of the strip and positional information is determined from the angular information derived from the two cameras.
A charge coupled device comprises an array of photosensitive elements, called pixels or photosites, which produce small packets of charge in dependence on the arnount of light falling on them and the length of time for which it falls. The photosltes may be arranged either in the form of a grid or in a line, but for edge detection purposes linear arrays are normally employed. In the case of a strip whose width is to be measured, the strip is focussed on the linear array of photosites by means of a lens in the charge coupled device camera. The photosite are electronically scanned in the charge coupled device which generates an output consisting of a series of pulses having heights representing a line scan across the strip.
By analysing the signal representing the line scan, it is possible to determine the location of the edges ..~. .

5~

of the strip relative to the charge coupled device.
However, commercially available charge coupled devices currently have a maximum of two thousand and fourty eight photosites in a linear array. These photosites are 13 microns square and the distance between centers, or pitch, is also approximately 13 microns, the inter photosite spacing being insignificantly small. Nonetheless, the limitation on the accuracy of measurement of the image of the strip formed on the charge coupled device by the lens is at best limited to 13 microns, the pitch of the photosites.
In a steel rolling mill, the cameras normally have to be mounted about 3 1/2 meters above the strip which is about 2 meters wide. If the lens is arranged so that the field of view of the camera in the strip just extends across the strip there is apparently an inherent limitation on the ultimate accuracy of the device in the order of millimeter, assuming there are 2,048 photosites in the charge coupled device. In practice, the actual error is greater than this because the lens does not form an absolutely sharp image of the lens on the charge coupled device and, in the case of a hot strip, the radiation profile does not obviously change abruptly from zero to maximum value at the position of the edge. Some cooling occurs at the edge of the strip and radiaton is emitted from parts of the equipment outside the strip. Practical devices in the prior art were limited to an accuracy in the order of 6 to 8 millimeters whicn, given the rate at which steel passes through a rolling mill, constitutes a considerable unknown in the quantity of steel being processed.
In accordance with the prior art thinking therefore the limiting accuracy of measurement is determined 3~

by the limiting resolution of the charge coupled device camera which in turn is dependent on the pitch of the CCD elements.
A further problem in the prior art is that a complete line of information is analysed in order to determine the location of the edge in the scan. This is time consuming and makes it difficult to produce a continuous indication of the positions of the edges of the strip in real time.
Accordingly the present invention provides a method of locating the edge of a strip, comprising forming an image of the lQ strip at least in the region of the edge on a sensor composed of a series of radiation-responsive elements, scanning said image with said elements, each said element in each scan generat-ing a discrete signal when level depends on the amount of radiation received by that element, said discrete signals for the elements in each scan constituting a line of data representing discrete points on a radiation profile of the scanned region at locations corresponding to the positions of the elements of the sensor, storing said discrete signals of each line in separate memory locations, accessing said memory locations and comparing said discrete signals with a reference signal to identify a region of interest on said line of data in which, on the basis of predetermined criteria, said edge is likely to lie, and generating with the aid of a plurality of discrete signals of said line of data in said region of interest an output signal representing the point of maximum rate of change on said line of data to an accuracy exceeding the element pitch, said output signal representing the location of the edge.
If desired, the method may be performed without identifying the region of interest.

3~5S
The invention makes use of the fact that the image of the str.~p on the sens:or, usually a charge coupled device, is not perfectly sharp and in ~act covers several photo site s .~ ~ 3a~

S

or elements of the device. By analysing the signal covering several photosites it is possible to determine the position of the actual edge of the strip in the image to an accuracy exceeding the pitch of the elements of the charge coupled device. Furthermore, since a region of interest is first identified where the edge is likely to lie, it is not necessary to analyse a complete line of information in order to determine the position of the edge, resultin in a considerable saving in timeO The whole operation can take place in real time and the invention is capable of providing a continous read-out giving the width of a strip with a standard deviation in the order of o.n7 millimeters. This is considerably better than was thought to be possible in the prior art, given the inherent limitations resulting from the limiting resolution of charge coupled devices.
In a preferred embodiment a threshold value is defined on the composite signal adjacent the edge region and such that the actual edge is unlikely to lie at a photosite giving a signal below the threshold value. The threshold value may conveniently be set at 25% of the peak signal value. A processing unit then processes the composite signal moving along the photosites from the one at the threshold value differentiating as it goes to find the first point on the curve to the nearest photosite where the slope passes through a maximum. In the case of a rising edge the processing unit works successively along the photosites in ascending order. In the case of a trailing edge the processing unit works in ~escending order. Having foung the point of maximum slope the processing unit then defines upper and lower curve limit positions at which the slope has fallen to a predetermined fraction of its maximum value on 3~S

either side of the position o maximum slope. The mean of the signal values a~ the upper and lower limit positions is calculated to give a signal value on said curve which is taken to be the point on the curve corresponding to the actual position of the edge. This position of the edge is found to an accuracy exceeding the photosite pitch by noting the ~ mmediately enclosing the mean signal value and then effecting a linear interpolation between the enclosing photosites.
In an alternative embodiment the upper and lower curve limit values are deEined as above and the actual position of the edge is found by determining the centre of gravity of the curve between the upper and lower curve limit positions. The slope dy/dx and the moment of slope (dy/dx)DX are determined at each photosite between the upper and lower curve limit positions and respectively summed.
The sum of the moments of slope is then divided by the sum of the slopes to find the centre of gravity of the curve and this gives the actual postion of the edge directly as a fraction of the photosite pitch.
In a further embodiment the processing unit finds the best fit function y=f(x~ in the region of interest, sets the second order differential to zero to find the point of inflection and solves for x to find the actual position of the edge again directly as a function of the photosite pitch.
The invention will now be described in more detai1, by way of example only, with reference to the accompanying drawingst in which:-Fig. 1 illustrates an edge detection apparatusmounted over a strip being rolled in a steel mill;

Fig. 2a represents a typical radiation profile from a hot strip being rolled;
Fig. 2b represents a typical radiation profile from a hot strip with edge cracking;
Fig. 2c illustrates how an abrupt intensity change is symmetrically blurred by a lens;
Fig. 3 is an enlarged view of the encircled portion of the curve shown in Fig. 2a representing a region of interest in which an edge of a strip is likely to l:ie;
Fig. 4a is an idealized representation of this edge region;
Fig. 4b is a diagram illustrating the linear interpolation method employed in one embodiment of the invention;
Fig. 4c is a diagram illustrating how -the linear interpolation method gives a good approximation of the position of the actual edge;
Fig. S is a diagram representing the edge region of the strip illustrating an alternative method of determining the actual position of the edge;
Fig. 6 is a block diagram of an input stage of the edge detection apparatus;
Fig. 7a is a block diagram illustrating the operation of one embodiment of edge detection apparatus according to the invention;
Fig. 7b is a block diagram illustrating the operation of apparatus employing the linear interpolation method; and Fig. 7c is a block diagram illustrating the operation of the apparatus employing the center of gravity interpolation methodO

( As shown in Fig. 1, a pair of spaced charge-coupled-device (CCDj cameras 1,2 are mounted at a height H, which is approximately 3 1/2 meters, above a two-meter wide hot steel strip 7 emerging from a rolling mill. The cameras 1,2 have lenses 3 and 4 which form an image of the strip on respective charge coupled devices 5 and 6. The cameras 1,2 are housed in a water-cooled jacket to protect ~hem from the heat radiating from the strip 7.
Each charge coupled device ~omprises a linear array of 2,048 photoelements, also known as photosites or pixels. A
suitable charge-coupled-device is Fairchild's CCD device number 143. As described in the above-referenced Canadian Patent No. 1,103,018, the cameras 1 and 2 each scan the whole width of strip 7. Although the strip is shown as lying flat on its supporting bed~buckling may occur and the use of two spaced-apart cameras in this manner enables the precise position in space of the edge of the strip to be determined even when the strip is not lying flat on the supporting bedO
In this embodiment, the cameras 1 and 2 are responsive to the radiation emitted by the hot strip. This generally has a profile as shown in Fig. 2a. As each camera scans across the strip, the radiation level is initially low outside the strip, rises quite sharply in the first edge region, levels of in the middle of the strip near a peak value, falls again quite sharply in the second edge region and finally falls to the background level outside the strip.
The regions of interest lie in the cirlces referenced 8 and 9. In these regions the radiation level changes rapidly and it is here that the actual edges of the strip lieO The profile shown in Fig. 2a represents the radiation profile of --7~

~3~5 the image of the strip projected onto the charge-coupled device The actual radiation from the strip exhibits a fairly sharp step at the edge position, even though cooling of the strip takes place at the edges, but this i5 distorted substantially symmetrically about the edge position as the result of the defocussing effect of the lenses 3 and 4.
This is shown more clearly in Fig. 2c where line 10 shows the step which would occur at an idealized edge and line 11 shows how the step is blurred in the image of the strip owing to the defocusslng effect of the lens which will have a characteristic transfer function. In some cases cracking occurs at the edge of the strip and gives rise to a local peak at the base of the edge-region curve as shown in Fiq.
2b. The edge detection apparatus can be adapted to take this into account as will be discussed below.
In practice each camera generates a series of sequential ~ulses for each scan, each representing the amplitude level at the respective CCD photosites numbered from 0 to 2047. The first signal pulse represents the amplitude of light falling on photosite number 0, the second on photosite number 1 sequentially through to photosite number 2047. The pulse amplitudes are digitized and represented by a number from 0 to 255. Thus, in Fig. 2a, the y axis represents discrete amplitude levels having values from 0 to 25S and the x axis represents discrete photoelement locations having numbers from 0 to 2047. As shown in Fig.6 ~which shows the circuitry for the camera 2), the output of each camera is fed to an analog-to-digital converter 50, which is actually mounted in the camera housing above the hot strip, to form an eight-bit digital word which in turn is fed to a noise filter 12 where 3~5 characteristic random noise associated with each camera isremoved, When the equipment is initially set up a neutral back~round i9 scanned to determine the noise signature of the cameras and then this i5 used to calibrate the filter 12.
The output of the noise filter 12 is connected alternately to 2k*8 random-access memories 13 and 14 through an electronic switch 15. The RAMs 13 and 14 are arranged to store alternate scans across the strip from the camera and each holds a complete line of information. An electronic switch 16 operates in synchronism with the electronic switch 15 and is so arranged that while a line of information from the camera is being written into ~he RAM 13, the information in the RAM 14 representing the previous line is analysed by means of circuîtry to be described. ~urthermore, in practice, since the gain from one element to the next may vary, but the alternate elements of the charge-coupled are matched, signals from alternate elemen~s are analysed separately to avoid the superimposition of a signal having a wavelength related to the pitch of the photoelements.
The RA~s 13 and 14 are connected alternately to ecti~r 5/
-r ~ the threshold th~ee~'by the electronic switch 16. The ,~, ~
digital words representing the amplitude levels at the successive photosites are sequentially accessed and fed to the threshold detector 51 which determines whether the signal is above a predetermined threshold 17 as shown in Fig. 2a. The threshold may be conveniently set to 25~ of the peak signal value 18 in Fig. 2a. The exact threshold value can be determined by experiment, but the closer it can be to the actual position of the edge, the shorter the processing time for analysing the CCD signals. The region of interest stack stores the photosite positions where the ~3~5 amplitude level crosses the predetermined threshold, i.e.
the photosite positions corresponding to the points A and B
in Fig. 2a. Further transitions through the threshold may occur owing the the presence of water droplets and the like on the steel. rrhese are also stored by the region of interest stack and discarded during subsequent processing The mode of operation of the apparatus to locate the precise position of a rising edge will now be described with reference to Figs. 7a through 7c. As shown in Fig~ 7a, unit 52 first examines the region of interest stack 108 to check for the presence of steel. If steel is present, unit 53 controlling the accessing of the region of interest stack 108 passes the entry to unit 54 where it is examined by unit 55 to check that the entry is on a rising edge. The photosite elements are then examined sequentially in ascending order in unit 57 by comparing the signal values at adjacent photosites until the position of maximum slope is found. If unit 58 confirms that the amplitude is acceptable the value of the peak slope and its position are stored in the unit 59. A predetermined ~raction of this slope is then calculated in unit 60 and the photosites where the values of the slope fall to this predetermined fraction on either side of the photosite where the slope has its peak value are found by units 61 and 63 and stored in units 62 and 64 respectively. Units 65, 66 and 67 check whether the distance from the positions calculated in units 61 and 63, and combined in some suitable manner, such as by taking their average value, is at least ten per cent of the scan size and the amplitude in this region exceeds twenty-five per cent of the scan peak. The selection of the mode of interpolation is then effected by unit 68 3,.~ S~

In Fig. 3 the position of the peak slope is referenced 20 and the upper limit of the working curve is referenced 21 and this corresponds to the photosite where, in this example, the slope of the curve has fallen to 50% of the slope at point 20 which corresponds to the photosite where the slope is a maximum. Point 22 indicates the lower limit of the working curve on the other side of the point 20 where the slope has also fallen to 50~ of its peak value~
For detection of the edge in the region of interest 8, a similar procedure is followed except of course the photosites are examined in descending order from the threshold point B in Fig. 2 until the position of maximum slope is found. The upper and lower curve working limits are then defined by finding the photosites where the slope has fallen to a predetermined fraction of this maximum value on either side of the central photosite.
The precise location of the edge is then determined by analyzing the signal within the defined working region. As can be seen in Fig. 4a, the position of maximum slope generally corresponds to the precise position of the edge of the strip. The curve shown in Fig. 4a, although shown as a continuous curve, is in fact of course composed of a series of discrete values representing the signal levels at the photosites of the charge-coupled device. The discrete values are spaced by a distance equal to the pitch, i.e., the distance between the centers, of the photosites. The position of peak slope of the curve passing through the discrete signal values may be determined by linear interpolation to an accuracy exceeding the photosite pitch. When this method is selected unit 81 of Fig. 7c calulates the amplitude differene be~ween the signal values ~3~.~5 at the upper and lower curve limit positions of the curve in the defined working region. Unit 82 th~m multiplies this difference by a system constant rho, normally 1/2. The result i5 then added to the amplitude of the signal at the lower curve limit in unit 83. Thix defines a signal value which, assuming rho is 1/2, represents the mean value theta of the amplitudes of the signal values 'at the upper and ,,lower curve limits. It is unlikely that this signal value will correspond exactly to the location of a photosite in the working region. The curve is examined in unit 84 to locate the photosites have associated signal values immediately surrounding the calculated mean value theta. As indicated in Fig. 4b, these photosites are referred to as the lower and upper enclosing photosites respectively. If the amplitude and photosite count'of the lower and upper enclosing photosites are respectively ~ , pf and ~ , p~ and the calculated mean signal value is ~ , the exact position of the edge e may be calculated by linear interpolation as a fraction of the photosite count from the following relation:
(Q -a~ /a~ -a~ ) = (e-p~ p~ ).
The result of this calculation is rounded to 1/16 of a photosite in unit 70 (Fig. 7a) and stored in a dynamic averaging stack 71 which continuously averages the result of this analysis. The value stored in the dynamic averaging stack 71 is displayed in the display unit 72. It may also be continuously recorded on a, chart recorder or used as a feedback signal to control the rolling operation., Fig. 4c shows how the slope of the curve is calculated. For any photosite position P the cord between equidistant positions on either side of the position P will have approximately the same slope as the tangent to the curve at position P. The slope of the curve at position P
may therefore be calculated approximately with reasonable accuracy by dividing the amplitude difference between the signal values at the enclosing photosites by the difference in count numbers between these photosites, i.e 2 in this example. It is possible to effect linear interpolation between photosites which are not immediately adjacent the peak-value position and this can be advantageous to avoid false readings caused by local defects or edge cracking~
An alternative method of interpolating to ind the accurate position of the edge is shown in Fig. 5. The upper and lower curve working limits are foun~ in exactly the same manner as in the linear interpolation method by the arrangement shown in Fig. 7b. However, once these positions have been found, the slope of the curve is calculated at each photosite position. This may be calculated rom the curve values at enclosing photosites as described with reference to Figs. 4c and 7h. Unit 73 sets the pointer of the register controlling the accessing of a stack storing the photosite levels in the working region to the curve lower limit position and an end-stop to the upper limit position. Unit 75 calculates the slope dy/dx and the moment of the slope, i.e. (dy/dx)DX at successive photosite positions. The results of these calculations are stored in respective summing registers 76. Unit 77 ensures that the calculation is repeated for the each photosite position until the end-stop is reached which is determined by the unit 78. When this occurs the summing registers contain values equal to the sums of the slopes and of the moments of the slope over the whole working- region. The sums of the moments of slopes are then divided by the sums of the slopes ss over the curve working region in unit 79 and the result added to ~he curve lower limit position in unit 8~. The result of this calculation gives the value x of the exact position of the edge as a fraction of the photosite count numbers. This is then fed to the unit 70 of Fig~ 7a where it is rounded to one sixteenth of the photosite pitch before being fed to the dynamic averaging stack 71 and the dispaly unit 72~ This method has the advantage over the linear interpolation method that it compensates for spurious signals which may arise at some of the photosites. If such a signal happens to arise at a photosite in the linear interpolation method which lies at the predetermined fraction slope value of the peak slope, it will give rise to an inaccuracy in the calculated position o the edge.
In the edge region cracking may occur which gives rise to a spurious peak 9' in the radiation profile as shown in Fig. 2b. The edge detection apparatus is preferably arranged to disregard this spurious peak. This can be achieved by moving outwards alon the curve from the threshold value and determinîng the slope dy/dx for five pairs o photosites. The slopes between these adjacent photosites are summed and if the resulting sum is greater than or equal to a predetermined number, the memory accessing pointer is assumed to be in the noise region, i.e., on the curve base and it then moves into the steel region differentiating as previously described to find the photosite position correspondin to maximum slope.
In an alternative method, the best fit function y = f(x) can be calculated for the curve between ~he curve upper and lower limits. The point of inflection~ or position of maximum slope, may then be calculated by setting .

3~5~

the second order of differential d2y/dx2 = O and solving forx. This method also compensates for spurious signals from individual photosites and gives the exact position as a fraction of the photosite count number.
In the embodiments described the cameras are responsive to the radiation emitted from a hot steel strip being worked, usually rolled, in a steel mill~ The cameras can however operate on radiation reflected from a strip which is not self luminous. The radiation is normally light although it can of course be electromagnetic radiation lying outslde the visible region, or any other form of radiation such as ultrasonic radiation.
As will be evident the invention can be implemented with off-the-shelf discrete logic components or by employing microprocessor technology.
In this specification the term strip is intended to cover any generally elonyate object having a width dimension which can be continuously monitored as it moves under the cameras and, in particular, includes bars, slabs and the like.

Claims (56)

THE EMBODIMENTS OF THE INVENTION IN WHICH AN
EXCLUSIVE PRIVILEGE OR PROPERTY IS CLAIMED ARE DEFINED AS
FOLLOWS:

1. A method of locating the edge of a strip, comprising forming an image of the strip at least in the region of the edge on a sensor composed of a series of radiation-responsive elements, scanning said image with said elements, each said element in each scan generating a discrete signal when level depends on the amount of radiation received by that element, said discrete signals for the elements in each scan constituting a line of data representing discrete points on a radiation profile of the scanned region at locations corresponding to the positions of the elements of the sensor, storing said discrete signals of each line in separate memory locations, accessing said memory locations and comparing said discrete signals with a reference signal to identify a region of interest on said line of data in which, on the basis of predetermined criteria, said edge is likely to lie, and generating with the aid of a plurality of discrete signals of said line of data in said region of interest an output signal representing the point of maximum rate of change on said line of data to an accuracy exceeding the element pitch, said output signal representing the location of the edge.

2. A method according to claim 1 wherein the sensor forms part of a charge-coupled device camera.

3. A method according to claim 2 wherein said discrete signals generated by the sensor elements are passed through an analog-to-digital converter.

4. A method according to claim 1 wherein said reference signal is a predetermined threshold selected such that said edge is not likely to be located on an element giving a signal less than said threshold.

5. A method according to claim 2 wherein said threshold is set at about 25% of the peak value of a complete scan across the field of view of said sensor.

6. A method according to claim 4, comprising progressively accessing the memory locations from the location storing the signal at threshold to the location storing signals from elements located towards the centre of the image of the strip, operating on the signals from said memory locations to generate a signal marking the element were the rate of change of signal is a maximum, operating on said signals to generate signals marking the elements where the rate of change of signal on the line of data falls to a predetermined fraction of the maximum to define upper and lower working limit positions, and generating with the aid of the signals from the elements at said upper and lower working limit positions said output signal.

7. A method according to claim 6 wherein the upper and lower working limit positions are defined as being the elements positions where the rate of change of signal on the line of data falls to 50% of its maximum value in the region of interest.

8. A method according to claim 6 wherein, prior to generating the signal marking the element of maximum rate of change of signal, rate-of-change signals are generated for elements outside the region of interest, said rate-of-change signals are summed over a predetermined number of pairs of elements to generate a sum signal, and the sum signal is compared with a predetermined reference signal to confirm that these elements do lie in the noise region and therefore the base region of the radiation profile.

9. A method according to claim 6 or 7 comprising forming a difference signal for said signals at the limit positions, multiplying the difference signal by a predetermined constant value, adding the resulting signal to the signal at said lower limit position to form a sum signal, comparing said sum signal to the signals between the upper and lower limit positions to identify the enclosing elements producing signals enclosing the value of said sum signal, and generating the output signal on the basis of linear interpolation from the positions of the enclosing elements.

10. A method according to claim 6 or 7 comprising forming a signal representing the mean of the signals at said limit positions, identifying the elements producing signals enclosing said means signals and generating the output signal on the basis of linear interpolation from the positions of said enclosing elements.

11. A method according to claim 6 or 7 wherein the signals are processed to form rate of change signals and moment of rate-of-change signals at the elements between the upper and lower limit positions, the rate of change signals are summed, and moment of rate-of-change signals are divided by a signal representing the sum of the rate-of-change signals to derive the position of the edge directly as a fraction of the element pitch.

12. A method according to claim 1 comprising process-ing said discrete signals in said region of interest to find the best fit curve passing therethrough y=f(x), where y represents the value of the function and x represents the distance along said distance scale, setting the second order differential d2y/dx2 equal to zero, and solving for x to determine the precise location of the edge.

13. A method according to claim 1 wherein consecutive lines of data are stored in alternate memories.

14. A method of locating the edge of a strip, comprising forming an image of the strip at least in the region of the edge on a sensor composed of a series of radiation-responsive elements, scanning said image with said elements, each said element in each scan generating a discrete signal whose level depends on the amount of radiation received by that element, said discrete signals for the elements in each scan constituting a line of data representing discrete points on a radiation profile of the scanned region at locations corresponding to the positions of the elements of the sensor, storing said discrete signals of each line in speparate memory locations, accessing said memory locations and comparing said discrete signals with a reference signal to identify a region of interest on said line of data in which, on the basis of predetermined criteria, said edge is likely to lie, processing the discrete signals in the region of interest to identify a first discrete signal where the rate of change of signals is a maximum and second discrete signals where the rate of change of signals is a predetermined fraction of the maximum, forming the mean of said second discrete signals, and, if said mean is exactly equal to any one of said discrete signals, generating an output signal indicating the centre of the corresponding element as being the position of the edge, and, if said mean is not exactly equal to any one of said discrete signals, generating an output signal indicating the position of the edge to an accuracy exceeding the element pitch on the basis of linear interpolation.

15. A method according to claim 14 wherein the discrete signals are passed through an analog-to-digital connection prior to processing.

16. A method according to claim 15 wherein the sensor forms part of a charge-coupled device camera.

17. A method according to claim 14 wherein the interpolation is carried out between elements giving corresponding discrete signals closest to said mean.

18. A method according to claim 14 or 17 wherein the region of interest is found by comparing the discrete signals with a threshold equal to about 25% of the peak signal.

19. A method according to claim 11 wherein, prior to locating the element where the rate of change of signal is a maximum, the elements outside the region of interest are analysed in a direction away from the centre of the strip by differentiating the signals between adjacent elements in pairs over a predetermined number of pairs, summing the differentiated signals and establishing whether the resulting sum is greater than or equal to a predetermined number to ensure that these elements lie in the noise region and hence the base region of the radiator profile.

20. A method of locating the edge of a strip, comprising forming an image of the strip at least in the region of the edge on a sensor composed of a series of radiation-responsive elements, scanning said image with said elements, each said element in each scan generating a discrete signal whose level depends on the amount of radiation received by that element, said discrete signals for the elements in each scan constituting a line of data representing discrete points on a radiation profile of the scanned region at locations corresponding to the positions of the elements of the sensor, storing said discrete signals of each line in separate memory locations, accessing said memory locations and comparing said discrete signals with a reference signal to identify a region of interest on said line of data in which, on the basis of predetermined criteria, said edge is likely to lie, processing the discrete signals in said region of interest to generate rate-of-change signs dy/dx at each element and moment-of-rate-of-change signals (dy/dx)dx at each element, respectively summing the rate-of-change signals and the moment-of-rate-of-change signals, and dividing the sum of the moment signals by the sum of the rate-of-change signals to generate an output signal representing the position of the edge directly as a fraction of the element pitch.

21. A method according to claim 20 wherein the region of interest is identified by taking a threshold equal to about 25% of the peak signal value in each scan.

22. A method according to claim 20 werein the discrete signals are passed through an analog-to-digital converter prior to processing.

23. A method according to claim 22, wherein the sensor forms part of a charged-coupled-device camera.

24. A method according to claim 20, wherein the discrete signals for consecutive lines of data are stored in alternate memorories.

25. An apparatus for locating the edge of a strip, comprising a sensor composed of a series of radiator-responsive elements, means for forming an image of the strip at least in the region of the edge on the sensor, means for scanning the image with the elements of the sensor such that each element generates a discrete signal whose level depends on the amount of radiation received by that element, the discrete signals from the elements in each scan thereby constituting a line of data representing discrete points on a radiation profile of the scanned region, a memory, means for storing said discrete signals of each line in separate locations of said memory, means for accessing said memory locations and comparing said discrete signals with a reference to identify a region of interest on said line of data, in which, on the basis of predetermined criteria, said edge is likely to lie, and means for processing a plurality of discrete signals of said line of data falling in said region of interest to generate an output signal representing the point of maximum rate of change on said line of data to an accuracy exceeding the element pitch, said output signal representing the location of the edge.

26. A method according to claim 25 wherein said processing means processes said discrete signals to generate a first signal representing a first discrete signal from a first said element where the rate of change of signals is a maximum, and a second signal representing second said discrete signals from second element on either side of said first discrete element where the rate of change of signal is a predetermined fraction of said maximum to define upper and lower working limits, said processing means further forming the mean of said second discrete signals, and, if said mean is exactly equal to any one of said discrete signals, said processing means generating an output signal indicating the centre of the corresponding element as representing the precise location of said edge, and, if said mean does not exactly equal any one of said discrete signals, said processing means generating an output signal representing the position of the edge as a fraction of the element point by linear interpolation.

27. An apparatus according to claim 26 wherein said processing means interpolates between the elements having corresponding discrete signals closest to said mean.

28. An apparatus according to claim 25 comprising a comparator for comparing said discrete signals with a predetermined threshold to identify said region of interest.

29. An apparatus according to claim 25 wherein said predetermined threshold is set at about 25% of the peak value of each scan.

30. An apparatus according to any one of claims 25 to 27 wherein the radiation responsive sensor is a charge-coupled device camera.

31. An apparatus according to claim 25 further comprising means for checking that said threshold lies at the boundary of the edge region comprising means for differentiating the discrete signals from pairs of elements moving in a direction away from the centre of the steel from the threshold, means for summing the differentiated signals, and means for establishing whether the resulting sum is greater than or equal to a predetermined value to ensure that these elements lie in the noise region.

32. An apparatus according to claim 25, comprising a second memory, said storing means being arranged to store consecutive lines of data alternately in said memories.

33. An apparatus for locating the edge of a strip, comprising a sensor composed of a series of radiation-responsive elements, means for forming an image of the strip at least in the region of the edge on the sensor, means for scanning the image with the elements of the sensor such that each element generates a discrete signal whose level depends on the amount of radiation received by that element, the discrete signals from the elements in each scan thereby constituting a line of data representing discrete points on a radiation profile of the scanned region, a memory, means for storing said discrete signals of each line in separate locations of said memory, means for accessing said memory locations and comparing said discrete signals with a reference to identify a region of interest on said line of data, in which, on the basis of predetermined criteria, said edge is likely to lie, means for processing said discrete signals in the region of interest to generate a first signal representing the element where the rate of change of signal is a maximum, means for generating a second signal representing said elements on either side of said first discrete signal where the rate of change of signal is a predetermined fraction of said maximum to define upper and lower working limits, means generating signals representing the rates of change of signal dy/dx and the moments of the rates of change (dy/dx)dx at the elements between the upper and lower limit values, summing registers for storing the respective sums of said rates of change and said moments of rates of change, and divider means for dividing the sum of said moments by the sums of said rates of change to generate an output signal representing the position of the edge directly as a fraction of the element pitch.

34. An apparatus according to claim 33 further comprising a comparator for comparing said discrete signals with a predetermined threshold to identify said region of interest.

35. An apparatus according to claim 34 wherein said predetermined threshold is set at about 25% of the peak value of the signal in each scan.

36. An apparatus according to claim 25 comprising means for generating a signal representing a best fit curve y=f(x) through said discrete signals were y represents the value of the function and x represents the distance along said scale, means for generating signals representing the second order differential d2y/dx2, and means for setting said second order differential equal to zero and solving for x to generate the output signal representing the position of the edge.

37. A method of locating the edge of a strip, comprising forming an image of the strip at least in the region of the edge on a sensor composed of a series of radiation-responsive elements, scanning said image with said elements, each said element in each scan generating a discrete signal when level depends on the amount of radiation received by that element, said discrete signals for the elements in each scan constituting a line of data representing discrete points on a radiation profile of the scanned region at locations corresponding to the positions of the elements of the sensor, storing said discrete signals of each line in separate memory locations, accessing said memory locations and generating with the aid of a plurality of discrete signals of said line of data an output signal representing the point of maximum rate of change on said line of data to an accuracy exceeding the element pitch, said output signal representing the location of the edge.

38. A method according to claim 37 wherein the sensor forms part of a charge-coupled device camera.

39. A method according to claim 38 wherein said discrete signals generated by the sensor elements are passed through an analog-to-digital converter.

40. A method according to claim 37, comprising progressively accessing the memory locations from a predetermined location to the locations storing signals from elements located towards the centre of the image of the strip, operating on the signals from said memory locations to generate a signal marking the element where the rate of change of signal is a maximum, operating on said signals to generate signals marking the elements where the rate of change of signal on the line of data falls to a predetermined fraction of the maximum to define upper and lower working limit positions, and generating with the aid of the signals from the elements at said upper and lower working limit positions said output signal.

41. A method according to claim 40 wherein the upper and lower working limit positions are defined as being the elements positions where the rate of change of signal on the line of data falls to 50% of its maximum value on either side of the element where the rate of change of signal is a maximum.

42. A method according to claim 40 wherein, prior to generating the signal marking the element of maximum rate of change of signal, rate-of-change signals are generated for elements outside the region of interest, said rate-of-change signals are summed over a predetermined number of pairs of elements to generate a sum signal, and the sum signal is compared with a predetermined reference signal to confirm that these elements do lie in the noise region and therefore the base region of the radiation profile.

43. A method according to claim 40 comprising forming a difference signal for said signals at the limit positions, multiplying the difference signal by a predetermined constant value, adding the resulting signal to the signal at said lower limit position to form a sum signal, comparing said sum signal to the signals between the upper and lower limit positions to identify the enclosing elements producing signals enclosing the value of said sum signal, and generating the output signal on the basis of interpolation from the positions of the enclosing elements.

44. A method according to claim 43 wherein said interpolation is liner.

45. A method according to claim 40 comprising forming a signal representing the mean of the signals at said limit positions, identifying the elements producing signals enclosing said mean signals and generating the output signal on the basis of interpolation between said enclosing elements.

46. A method according to claim 45 wherein the interpolation is linear.

47. A method according to claim 40 wherein the signals are processed to form rate of change signals and moment of rate-of-change signals at the elements between the upper and lower limit positions, the rate of change signals are summed, and moment of rate-of-change signals are divided by a signal reprepresenting the sum of the rate-of-change signals to derive the position of the edge directly as a fraction of the element pitch.

48. An apparatus for locating the edge of a strip, comprising a sensor composed of a series of radiator-responsive elements, means for forming an image of the strip at least in the region of the edge on the sensor, means for scanning the image with the elements of the sensor such that each element generates a discrete signal whose level depends on the amount of radiation received by that element, the discrete signals from the elements in each scan thereby constituting a line of data representing discrete points on a radiation profile of the scanned region, a memory, means for storing said discrete signals of each line in separate locations of said memory, means for accessing said memory locations, and means for processing a plurality of discrete signals of said line of data to generate an output signal representing the point of maximum rate of change on said line of data to an accuracy exceeding the element pitch, said output signal representing the location of the edge.

49. An apparatus according to claim 48 wherein said processing means processes said discrete signals to generate a first signal representing a first discrete signal from a first said element where the rate of change of signals is a maximum, and a second signal representing second said discrete signals from second element on either side of said first discrete element where the rate of change of signal is a predetermined fraction of said maximum to define upper and lower working limits, said processing means further forming the mean of said second discrete signals, and, if said mean is exactly equal to any one of said discrete signals, said processing means generating an output signal indicating the centre of the corresponding element as representing the precise location of said edge, and, if said mean does not exactly equal any one of said discrete signals, said processing means generating an output signal representing the position of the edge as a fraction of the element point by interpolation.

An apparatus according to claim 49 wherein said interpolation is linear.

51. An apparatus according to claim 48 wherein said processing means interpolates between the elements having corresponding discrete signals closest to said mean.

52. An apparatus according to any one of claims 48 to 50 wherein the radiation responsive sensor is a charge-coupled device camera.

53. An apparatus according to claim 48 further comprising means for checking that said threshold lies at the boundary of the edge region comprising means for differentiating the discrete signals from pairs of elements moving in a direction away from the centre of the steel from the threshold, means for summing the differentiated signals, and means for establishing whether the resulting sum is greater than or equal to a predetermined value to ensure that these elements lie in the noise region.

54. An apparatus according to claim 48, comprising a second memory, said storing means being arranged to store consecutive lines of data alternately in said memories.

55. A method according to claim 6 or 7 comprising forming a difference signal for said signals at the limit positions, multiplying the difference signal by a predetermined constant value, adding the resulting signal to the signal at said lower limit position to form a sum signal, comparing said sum signal to the signals between the upper and lower limit positions to identify the enclosing elements producing signals enclosing the value of said sum signal, and generating the output signal on the basis of interpolation from the positions of the enclosing elements.

56. A method according to claim 6 or 7 comprising forming a signal representing the mean of the signals at said limit positions, identifying the elements producing signals enclosing said mean signals, and generating the output signal on the basis of interpolation from the positions of said enclosing elements.