`
`PCT/AU97100198
`
`- 8 -
`
`and stores data concerning the wingspan 22 of the aircraft and other details
`
`concerning the size, shape and ICAO category (A to G) of the aircraft. The image
`
`proceSSing system 8 classifies the aircraft on the basis of the size which can be used
`
`subsequently when determining the registration markings on the port wing 20. The
`5 data obtained can also be used for evaluation of the aircraft during landing and/or
`
`take-off.
`
`Alternatively a pyroelectric sensor 27 can be used with a signal processing wing
`detection unit 29 to provide a tracking system 1 which also generates the acquisition
`
`10 signal using the trigger logic circuit 39, as shown in Figure 4 and descnbed later.
`
`Detecting moving aircraft in the field of view of the sensor 3 or 27 is based on
`
`forming a profile or signature of the aircraft, P(Y,t), that depends on a spatial
`
`coordinatey and time t. To eliminate features in the field of view that are secondary
`
`15 or slowly moving, a difference profile f).P(y.t) is formed. The profile or signature can be
`
`differenced in time or in space because these differences are equivalent for moving
`
`objects. If the intensity of the light or thermal radiation from the object is not changing
`
`then the time derivative of the profile obtained from this radiation is zero. A time
`
`derivative of a moving field can be written as a convective derivative involving partial
`
`20 derivatives, which gives the equation
`
`dP(Y,t) _ ap(Y,t)
`at
`dt
`
`+ v ap(Y,t) = 0
`ay
`
`(3)
`
`where v is the speed of the object as observed in the profile. After rearranging
`
`equation (3), gives
`
`2P(Y,t) = -v ap(Y,t)
`at
`ay
`
`(4)
`
`which shows that the difference in the profile in time is equivalent to the difference in
`the profile in space. This only holds for moving objects, when v ~ o. Equation (4) also
`25 follows from the simple fact that if the profile has a given valueP(Yo,to) at the
`
`Sony, Ex. 1002, p.1501
`
`
`
`W097137336
`
`PCT/AU97/00198
`
`- 9 -
`
`coordinate (Yo,ta), then it will have this same value along the line
`
`(5)
`
`To detect and locate a moving feature that forms an extremum in the profile,
`
`such as an aircraft wing, the profile can be differenced in space t:.1'(y,t). Then an
`
`extremum in the profile P(y, t) will correspond to a point where the difference profile t:.1'(Y, t)
`
`5 crosses zero.
`
`In one method for detecting a feature on the aircraft, a profile P(Y,t) is formed
`
`and a difference profile t:.?(y,t) is obtained by differencing in time, as described
`below. According to equation (4) this is equivalent to a profile of a moving object that
`
`10 is differenced in space. Therefore the position yp of the zero crossing point of t:.,P(y,t)
`at time t is also the position of the zero crossing point of t:.?(y,t) which locates an
`extremum in P(y,t).
`
`In another method for detecting a feature on the aircraft, the difference between
`
`15 the radiation received by a sensor 27 from two points in space is obtained as a
`
`function of time, t:.ys(t) , as described below. If there are no moving features in the field
`
`of view, then the difference is constant. If any object in the field of view is moving, then
`
`the position of a point on the object is related to time using equation (5). This allows
`
`a profile or signature differenced in space to be constructed
`
`(6)
`
`20 and, as described above, allows an extremum correspondIng to an aircraft wing to be
`
`located in the profile from the zero crossing point in the differential signature.
`
`The image acquisition system 10 includes at least one high resolution camera
`
`7 to obtain images of the aircraft when triggered. The images are of sufficient
`
`25 resolution to enable automatic character recognition of the registration code on the
`
`port wing 20 or elsewhere. The illumination unit 16 is also triggered simultaneously to
`
`provide illumination of the aircraft during adverse lighting conditions, such as at night
`
`Sony, Ex. 1002, p.1502
`
`
`
`W097137336
`
`PCT/AU97100198
`
`-10-
`
`or during inclement weather.
`
`The acquired images are passed to the analysis system 12 which performs
`
`Optical Character Recognition (OCR) on the images to obtain the registration code.
`
`5 The registration code corresponds to aircraft type and therefore the aircraft
`
`classification determined by the image processing system 8 can be used to assist to
`
`the recognition process, particularly when characters of the code are obscured in an
`
`acquired image. The registration code extracted and any other information concerning
`
`the aircraft can be then passed to other systems via a network connection 24.
`
`10
`
`Once signals received from the pyroelectric sensors 21 indicate the aircraft 28
`
`is within the field of view of the sensors 3 of the tracking sensor system 6, the tracking
`
`system 1 is activated by the proximity detector 4. The proximity detector 4 is usually
`
`the first stage detection system to determine when the aircraft is in the proximity of the
`
`15 more precise tracking system 1. The tracking system 1 includes the tracking sensor
`
`system 6 and the image processing system 8 and according to one embodiment the
`
`images from the detection cameras 3 of the sensor system 6 are used by the image
`
`processing system 8 to provide a trigger for the image acquisition system when some
`
`point in the image of the aircraft reaches a predetermined pixel position. One or more
`
`20 detection cameras 3 are placed in appropriate locations near the airport runway such
`
`that the aircraft passes within the field of view of the cameras 3. A tracking camera 3
`
`provides a sequence of images, {/ n}' The image processing system 8 subtracts a
`
`background image from each image 'nof the sequence. The background image
`
`represents an average of a number of preceding images. This yields an image b.1n that
`
`25 contains only those objects that have moved during the time interval between images.
`The imagellJn is thresholded at appropriate values to yield a binary image, i.e. one
`that contains only two levels of brightness, such that the pixels comprising the edges
`
`of the aircraft are clearly distinguishable. The pixels at the extremes of the aircraft in
`
`the direction perpendicular to the motion of the aircraft will correspond to the edges
`
`30 18 of the wings of the aircraft. After further processing, described below, when it is
`
`determined the pixels comprising the port edge pass a certain position in the image
`
`Sony, Ex. 1002, p.1503
`
`
`
`W097J37336
`
`PCT/AU97/00198
`
`- 11 -
`
`corresponding to the acquisition point, the acquisition system 10 is triggered, thereby
`
`obtaining an image of the registration code beneath the wing 20 of the aircraft.
`
`Imaging the aircraft using thermal infrared wavelengths and detecting the
`
`5 aircraft by its thermal radiation renders the aircraft self-luminous so that it can be
`
`imaged both during the day and night primarily without supplementary illumination.
`
`Infrared (IR) detectors are classified as either photon detectors (termed cooled
`
`sensors herein), or thermal detectors (termed uncooled sensors herein). Photon
`
`detectors (photoconductors or photodiodes) produce an electrical response directly
`
`10 as the result of absorbing IR radiation. These detectors are very sensitive, but are
`
`subject to noise due to ambient operating temperatures. It is usually necessary to
`
`cryogenically cool (80 0 K) these detectors to maintain high sensitivity. Thermal
`
`detectors experience a temperature change when they absorb IR radiation, and an
`
`electrical response results from temperature dependence of the material property.
`
`15 Thermal detectors are not generally as sensitive as photon detectors, but perform well
`
`at room temperature.
`
`Typically, the cooled sensing devices are formed from Mercury Cadmium
`
`Telluride offer far greater sensitivity than uncooled devices, which may be formed from
`
`20 Barium Strontium Titanate. Their Net Equivalent Temperature Difference (NETD) is
`
`also superior. However, with the uncooled sensor a chopper can be used to provide
`
`temporal modulation of the scene. This permits AC coupling of the output of each pixel
`
`to remove the average background. This minimises the dynamic range requirements
`
`for the processing electronics and amplifies only the temperature differences. This is
`
`25 an advantage for resolving differences between cloud, the sun, the aircraft and the
`
`background. The advantage of differentiation between objects is that it reduced the
`
`load on subsequent image processing tasks for segmenting the aircraft from the
`
`background and other moving objects such as the clouds.
`
`30
`
`Both a cooled and uncooled thermal infrared imaging system 6 has been used
`
`during day, night and foggy conditions. The system 6 produced consistent images of
`
`Sony, Ex. 1002, p.1504
`
`
`
`WO 97137336
`
`PCT/AU97100198
`
`- 12 -
`
`the aircraft in all these conditions, as shown in Figures 8 and 9. In particular, the sun
`
`in the field of view produced no saturation artefacts or flaring in the lens. At night, the
`
`entire aircraft was observable, not just the lights.
`
`5
`
`The image processing system 8 uses a background subtraction method in an
`
`attempt to eliminate slowly moving or stationary objects from the image, leaving only
`
`the fast moving Objects. This is achieved by maintaining a background image that is
`
`updated after a certain time interval elapses. The update is an incremental one based
`
`on the difference between the current image and the background. The incremental
`
`10 change is such that the background image can adapt to small intensity variations in
`
`the scene but takes some time to respond to large variations. The background image
`
`is subtracted from the current image, a modulus is taken and a threshold applied. The
`
`result is a binary image containing only those differences from the background that
`
`exceed the threshold.
`
`15
`
`One problem with this method is that some slow moving features, such as
`
`clouds, still appear in the binary image. The reason for this is that the method does not
`
`select on velocity but on a combination of velocity and intensity gradients. If the
`
`intensity in the image is represented by l(x,Y,t). where x and Y represent the position
`20 in rows and columns, respectively, and t represents the image frame number (time)
`and if the variation in the intensity due to ambient conditions is very small then it can
`
`be shown that the time variation of the intensity in the image due to a feature moving
`
`with velocity v is given by
`
`al(x,Y,t)
`at
`
`- v.V'/(x,Y. t)
`
`(7)
`
`In practice, the time derivative in equation (7) is performed by taking the
`
`25 difference between the intensity at (x,Y) at different times. Equation (7) shows that the
`
`value of this difference depends on the velocity v of the feature at (x,Y) and the
`
`intensity gradient. Thus a fast moving feature with low contrast relative to the
`
`background is identical to a slow moving feature with a large contrast. This is the
`
`Sony, Ex. 1002, p.1505
`
`
`
`WO 97/37336
`
`PCT/AU97/00198
`
`- 13 -
`
`situation with slowly moving clouds that often have very bright edges and therefore
`
`large intensity gradients there, and are not eliminated by this method Since features
`
`in a binary image have the same intensity gradients, better velocity selection is
`
`obtained using the same method but applied to the binary image. In this sense, the
`
`5 background-subtraction method is applied twice, once to the original grey-scale image
`
`to produce a binary image as described above, and again to the subsequent binary
`
`image, as described below.
`
`The output from the initial image processing hardware is a binary image B(x,Y,t)
`10 where B(x,Y,t) = 1 If a feature is located at (x,Y) at time t, and B(x,Y,t) = Orepresents
`the background. Within this image the fast moving features belong to the aircraft. To
`
`deduce the aircraft wing position the two-dimensional binary image can be
`
`compressed into one dimension by summing along each pixel row of the binary image,
`
`P(Y,t) = f B(x,Y,t) dx
`
`(8)
`
`where the aircraft image moves in the direction of the image columns. This row-sum
`
`15 profile is easily analysed in real time to determine the location of the aircraft. An
`
`example of a profile is shown in Figure 10 where the two peaks 30 and 31 of the
`
`aircraft profile correspond to the main wings (large peak 30) and the tail wings (smaller
`
`peak 31).
`
`20
`
`in general, there are other features present, such as clouds, that must be
`
`identified or filtered from the profile. To do this, differences between profiles from
`
`successive frames are taken, which is equivalent to a time derivative of the profile.
`
`Letting A(x,Y,t) be the aircraft where A(x,Y,f} = 1 if (x,Y) lies within the aircraft and 0
`
`otherwise and letting C(x,Y,t) represent clouds or other slowly moving objects, then
`
`25 it can be shown that the time derivative of the profile is given by
`aP(y,t) = faA(X,y,t) dx + faC(x,Y,f) dx - I.E.. [A(x y f)C(X,y,t)jdx
`at
`at
`at
`a t ' ,
`= faA(~r,t) [1 - C(x,y,t)]dx + €(C)
`
`(9)
`
`Sony, Ex. 1002, p.1506
`
`
`
`W097137336
`
`PCT/AU97/00198
`
`- 14 -
`
`where e( C) '" 0 is a small error term due to the small velocity of the clouds. Equation
`
`(9) demonstrates an obvious fact that the time derivative of a profile gives information
`
`on the changes (such as motion) of feature A only when the changes in A do not
`
`overlap features C. In order to obtain the best measure of the location of a feature, the
`5 overlap between features must be minimised. This means that C(x,Y,t) must cover as
`
`small an area as possible. If the clouds are present but do not overlap the aircraft,
`
`then apart from a small error term, the time difference between profiles gives the
`
`motion of the aircraft. The difference profile corresponding to Figure 10 is shown in
`Figure 11 where the slow moving clouds have been eliminated. The wing positions
`
`10 occur at the zero-crossing paints 33 and 34. Note that the clouds have been removed,
`
`apart from small error terms.
`
`The method is implemented using a programmable logic circuit of the image
`
`processing system 8 which is programmed to perform the row sums on the binary
`
`15 image and to output these as a set of integers after each video field. When taking the'
`
`difference between successive profiles the best results were obtained using
`differences between like fields of the video image, i. e. even-even and odd-odd fields.
`
`The difference profile is analysed to locate valid zero crossing points
`
`20 corresponding to the aircraft wing pOSitions. A valid zero crossing is one in which the
`difference profile initially rises above a threshold IT for a minimum distance YT and
`falls through zero to below -IT for a minimum distance Yr' The magnitude of the
`threshold f T is chosen to be greater than the error term e( C) which is done to discount
`
`the affect produced by slow moving features, such as clouds.
`
`25
`
`30
`
`In addition, the peak value of the profile, corresponding to the aircraft wing, can
`
`be obtained by summing the difference values when they are valid up to the zero
`
`crossing point. This method removes the contributions to the peak from the non(cid:173)
`overlapping clouds. It can be used as a guide to the wing span of the aircraft.
`
`The changes in position of the aircraft in the row-sum profile are used to
`
`Sony, Ex. 1002, p.1507
`
`
`
`WO 97137336
`
`PCT/AU97/00198
`
`- 15 -
`
`determine a velocity for the aircraft that can be used for determining the image
`
`acquisition or trigger time, even if the aircraft is not in view. This situation may occur
`
`if the aircraft image moves into a region on the sensor that is saturated, or if the trigger
`
`pOint is not in the field of view of the camera 3. However, to obtain a reliable estimate
`5 of the velocity, geometric corrections to the aircraft position are required to account
`for the distortions in the image introduced by the camera lens. These are described
`
`below using the coordinate systems (x,Y,z) for the image and (X, Y,Z) for the aircraft
`
`as shown in Figures 12 and 13, respectively_
`
`10
`
`For an aircraft at distance Z and at a constant altitude Yo' the angle from the
`horizontal to the aircraft in the vertical plane is
`
`Yo
`tanS = (cid:173)
`y Z
`
`(10)
`
`Since Yo is approximately constant, a normalised variableZN = ZIYo can be used. If Yo
`is the coordinate of the centre of the images, f is the focal length of the lens and Se
`
`is the angle of the camera from the horizontal in the vertical plane, then
`
`)
`Yo - y
`tanSy - tanSe
`- - = tanf8 - 8 = --.!.---
`f
`\ Y
`1 + tane
`tane
`e
`Y
`c
`
`(11 )
`
`15 where the tangent has been expanded using a standard trigonometric identity. Using
`
`(10) and (11) an expression for the normalised distance ZN is obtained
`
`Z (y) = _1 _+_~,.,..(Y_-_~_o~,-a_n8_c
`[3(;' - Yo)
`tanSe -
`N
`
`(12)
`
`where [3 = 1 If. This equation allows a point in the image at y to be mapped onto a
`true distance scale, Zw Since the aircraft altitude is unknown, the actual distance
`
`cannot be determined. Instead, all points in the image profile are scaled to be
`
`20 equivalent to a specific point, Y"
`line or image acquisition line. The change in the normalised distance ZN(Yl) at Y1 due
`to an increment in pixel value 6.Y1 is 6.ZJY1) = ZN(Y1 + 6.Y,) - ZN(Y1 )' The number
`
`in the profile. This point is chosen to be the trigger
`
`Sony, Ex. 1002, p.1508
`
`
`
`W097/37336
`
`PCT/AU97/00198
`
`- 16 -
`
`of such increments over a distance ZN(Y2) - ZN(Y1) is M = (ZN(Y2) - ZN(Y1 ))/b.ZN(y1).
`Thus the geometrically corrected pixel position at Y2 is
`
`(13)
`
`F or an aircraft at distance Z and at altitude Yo' a length X on it in the X direction
`subtends an angle in the horizontal plane of
`
`(14)
`
`5 where normalised values have been used. If Xo is the location of the centre of the
`image and f is the focal length of the lens, then
`
`x - Xo = tan8x
`f
`
`(15)
`
`Using (12), (14) and (15), the normalised distance XN can be obtained in terms of x
`and Y
`
`(16)
`
`As with the Y coordinate, the x coordinate is corrected to a value at Y1 0 Since XN
`10 should be independent of position, then a length x2 - Xo at Y2 has a geometrically
`corrected length of
`
`(17)
`
`The parameter ~ = 11f is chosen so that x and yare measured in terms of pixel
`numbers. If Yo is the centre of the camera centre and it is equal to half the total
`number of pixels, and if 9FOV is the vertical field of view of the camera, then
`
`Sony, Ex. 1002, p.1509
`
`
`
`W097/37336
`
`PCT/AU971OO198
`
`- 17 -
`
`~ = tan(6 FO.}2)
`Yo
`
`(18)
`
`This relation allows p to be calculated without knowing the lens focal length and the
`dimensions of the sensor pixels.
`
`The velocity of a feature is expressed in terms of the number of pixels moved
`5 between image fields (or frames). Then if the position of the feature in frame n is Yn ,
`the velocity is given by vn :: Yn - Yn-1' Over N frames, the average velocity is then
`
`1 N
`1 N
`1
`(v) = - L v n = - L (Y n - Y n -1) = -(Y N
`Nn:l
`Nn:l
`N
`
`- Yo)
`
`(19)
`
`which depends only on the start and finish points of the data. This is sensitive to errors
`
`in the first and last values and takes no account of the positions in between. The error
`in the velocity due to an error oYN in the value YN is
`
`e:(v») = °YN
`N
`
`(20)
`
`10 A better method of velocity estimation uses all the position data obtained between
`these values. A time t is maintained which represents the current frame number. Then
`the current position is given by
`
`Y ::; Yo - vt
`
`(21)
`
`where Yo is the unknown starting point and v is the unknown velocity. The number n
`of valid positions Yn measured from the feature are each measured at time tn'
`15 Minimising the mean square error
`
`(22)
`
`with respect to v and Yo gives two equations for the unknown quantitiesyo and v.
`Solving for v yields
`
`Sony, Ex. 1002, p.1510
`
`
`
`W097/37336
`
`PCT/AU97/00198
`
`-18 -
`
`N
`
`N
`
`N
`
`L Yn L tn - N L yin
`n=1
`n=1
`v = n=1
`N
`N
`N
`N L t; - L tn L tn
`
`n=l
`
`n=l
`
`n=1
`
`(23)
`
`This solution is more robust in the sense that it takes account of all the motions of the
`
`feature, rather than the positions at the beginning and at the end of the observations.
`If the time is sequential, so that tn = n6t where tn = 1 is the time interval between
`image frames, then the error in the velocity due to an error 0Yn in the value Yn is
`e{(v») = eYn{6(N + 1 - 2n)}
`N
`(N + 1 )(N - 1)
`
`(24)
`
`5 which, for the same error eYN in (19), gives a smaller error than (21) for N > 5. In
`
`general, the error in (24) varies as 1/N2 which is less sensitive to uncertainties in
`
`position than (19).
`
`If the aircraft is not in view, then the measurement of the velocity v can be used
`10 to estimate the trigger time. If y, is the position of a feature on the aircraft that was last
`then the position at any time t is estimated from
`seen at a time t"
`
`(25)
`
`Based on this estimate of position, the aircraft will cross the trigger point located
`at Y T at a time t T estimated by
`
`Yr - y,
`tT = t, - - - (cid:173)
`v
`
`(26)
`
`An alternative method of processing the images obtained by the camera 3 to
`
`15 determine the aircraft position, which also automatically accounts for geometric
`
`corrections, is described below. The method is able to predict the time for triggering
`
`the acquisition system 10 based on observations of the position of the aircraft 28.
`
`Sony, Ex. 1002, p.1511
`
`
`
`W097137336
`
`PCT/AU97{OO198
`
`- 19 -
`
`To descnbe the location of an aircraft 28 and its position, a set of coordinates
`are defined such that the x axis points vertically upwards, the i axis points
`horizontally along the runway towards the approaching aircraft, and y is horizontal
`and perpendicular to the runway. The image 66 of the aircraft is located in the digitised
`
`5 image by pixel values (xp,Yp) ' where xp is defined to be the vertical pixel value and yp
`
`the horizontal value. The lens on the camera inverts the image so that a light ray from
`the aircraft strikes the sensor at position (-xp, -yp,O) , where the sensor is located at
`the coordinate origin. Figure 14 shows a ray 68 from an object, such as a point on the
`
`aircraft, passing through a lens of a focal length f, and striking the imaging sensor at
`10 a point (-xp' -yp) , wherexp andyp are the pixel vail..!es. The equation locating a point
`on the ray is given by
`
`(27)
`
`where z is the horizontal distance along the ray, and the subscript c refers to the
`
`camera coordinates. The camera axis ic is collinear with the lens optical axis. It will
`be assumed that zit» 1 , which is usually the case.
`
`15
`
`Assuming the camera is aligned so that Yc = y is aligned with the runway
`coordinate, but the camera is tilted from the hOrizontal by angle S. Then
`
`Xc = X cos8 - i sin8
`ic = x sinS + i cosS
`
`and a point on the ray from the aircraft to its image is given by
`r = 4(xp cosa J f + sina) x + (Yp J 1 y + (cose - Xp sina I 1 z]
`
`Letting the aircraft trajectory be given by
`t{t) = (z(t) tan8 GS + xo) x + YoY + z(t)z
`
`(28)
`
`(29)
`
`(30)
`
`20 where z(t} is the horizontal position of the aircraft at time t, 8 GS is the glide-slope
`angle, and the aircraft is at altitude Xo and has a lateral displacement Yo at z(to) = O.
`Here, t = to is the time at which the image acquisition system 10 is triggered, i. e. when
`
`Sony, Ex. 1002, p.1512
`
`
`
`WO 97137336
`
`PCT/AU97/00198
`
`- 20-
`
`the aircraft is overhead with respect to the cameras 7.
`
`Comparing equations (29) and (30) allows z to be written in terms of z(t) and
`
`gives the pixel positions as
`
`x (t) = f
`p
`
`(
`
`z(t)[cosS tanS GS - sinSl + Xo COSS]
`z(t)[sinS tanSGS + cosS) + Xo sinS
`
`5 and
`
`fyo
`Yp(t) = -~-------::.----
`Z(t)[stnS tanS GS + cosS] + xa sinS
`
`(31)
`
`(32)
`
`Since zp(t) is the vertical coordinate and its value controls the acquisition
`
`trigger, the following discussion will be centred on equation (31) The aircraft position
`
`is given by
`
`z(t) = v(to - t)
`
`(33)
`
`where v is the speed of the aircraft along the taxis.
`
`10
`
`The aim is to determine to from a series of values of zp(t) at t determined from
`
`the image of the aircraft. For this purpose, it is useful to rearrange (31) into the
`
`following form
`
`where
`
`c - t + ax
`p
`
`- btx = 0
`p
`
`a = vfo(tanS GS + cotS) + Xo
`fv(1 - tanS GS cotS)
`
`tan8GS + cotS
`b = - , - - - - - - -
`~1 - tan8 GS cote)
`
`15 and
`
`(34)
`
`(35)
`
`(36)
`
`Sony, Ex. 1002, p.1513
`
`
`
`WO 97137336
`
`PCT/AU97/00198
`
`- 21 -
`
`X.,xo
`C = ~ - ---------------
`fv( 1 - tane GS cote)
`
`(37)
`
`The pixel value corresponding to the trigger point vertically upwards is xr = f cote.
`
`The trigger time, to' can be expressed in terms of the parameters a, band c
`
`t = c + aXr
`1
`o
`b
`+ xr
`
`(38)
`
`The parameters a, band c are unknown since the aircraft glide slope, speed,
`altitude and the time at which the trigger is to occur are unknown. However, it IS
`
`5 possible to estimate these using equation (34) by minimising the chi-square statistic
`
`Essentially. equation (34) is a prediction of the relationship between the measured
`values xp and t , based on a simple model of the optical system of the detection
`camera 3 and the trajectory of the aircraft 28. The parameters a, band c are to be
`
`chosen 50 as to minimise the error of the model fit to the data, I. e. make equation (34)
`
`10 be as close to zero as possible.
`
`Let xn be the location of the aircraft in the image. i.e. pixel value, obtained at
`time tn. Then the chi-square statistic is
`
`N
`X2 = L
`
`n~l
`
`(c - tn + aXn - bt~nf
`
`(39)
`
`for N pairs of data points. The optimum values of the parameters are those that
`
`15 minimise the chi-square statistic, i.e. those that satisfy equation (34).
`
`F or convenience, the following symbols are defined
`
`N
`N
`N
`N
`X= Lxn.
`T = L tn. P = L Xntn. Y = LX;,
`n,l
`n:l
`n=1
`n=l
`N
`N
`N
`= Lx;tn. R LXnt;. S = Lx;t;.
`n:l
`n,l
`n:l
`
`(40)
`
`Sony, Ex. 1002, p.1514
`
`
`
`W097137336
`
`PCT/AU97/00198
`
`- 22-
`
`Then the values of a, band c that minimise equation (39) are given by
`
`a = (NP - Xn(p 2 - NS) - (NR - PD(PX - NQ)
`(NY - X2XP2 - NS) + (PX - NQ)2
`
`b = (NP - XTXpX - NQ) + (NR - PT'/..NY - X2)
`(NY - X 2XP2 - NS) + (PX - NQ)2
`
`and
`
`T + bP - aX
`c= - - - - -
`N
`
`(41)
`
`(42)
`
`(43)
`
`On obtaining a, band c from equations (41) to (43), then to can be obtained from
`equation (38).
`
`5
`
`Using data obtained from video images of an aircraft landing at Melbourne
`
`airport, a graph of aircraft image position as a function of image frame number is
`
`shown in Figure 15. The data was processed using equations (41) to (43) and (38) to
`yield the predicted value for the trigger frame number to = 66 corresponding to trigger
`10 point 70. The predicted point 70 is shown in Figure 16 as a function of frame number.
`The predicted value is to = 66 ± 0.5 after 34 frames. In this example, the aircraft can
`be out of the view of the camera 3 for up to 1.4 seconds and the system 2 can still
`
`trigger the acquisition camera 7 to within 40 milliseconds of the correct time. For an
`aircraft travelling at 62.5 mis, the system 2 captures the aircraft to within 2.5 metres
`
`15 of the required position.
`
`The tracking system 6, 8 may also use an Area-Parameter Accelerator (APA)
`
`digital processing unit, as discussed in International Publication No. WO 93/19441,
`
`to extract additional information, such as the aspect ratio of the wing span to the
`
`20 fuselage length of the aircraft and the location of the centre of the aircraft.
`
`The tracking system 1 can also be implemented using one or more pyroelectric
`
`sensors 27 with a signal proceSSing wing detection unit 29. Each sensor 27 has two
`
`Sony, Ex. 1002, p.1515
`
`
`
`W097137336
`
`PCT/AU971OO198
`
`- 23-
`
`adjacent pyroelectric sensing elements 40 and 42, as shown in Figure 17, which are
`
`electrically connected so as to cancel identical signals generated by each element. A
`
`plate 44 with a slit 46 is placed above the sensing elements 40 and 42 so as to
`
`provide the elements 40 and 42 with different fields of view 48 and 50. The fields of
`
`5 view 48 and 50 are significantly narrower than the field of view of a detection camera
`discussed previously. If aircraft move above the runway in the direction indicated by
`the arrow 48, the first element 40 has a front field of view 48 and the second element
`
`42 has a rear field of view 50. As an aircraft 28 passes over the sensor 27 the first
`
`element 40 detects the thermal radiation of the aircraft before the second element 42,
`
`10 the aircraft 28 will then be momentarily in both fields of view 48 and 50, and then only
`
`detectable by the second element 42. An example of the difference signals generated
`
`by two sensors 27 is illustrated in Figure 18 where the graph 52 is for a sensor 27
`
`which has a field of view that is directed at 90° to the horizontal and a sensor 27 which
`
`is directed at 75° to the horizontal. Graph 54 is an expanded view of the centre of
`
`15 graph 52. The zero crossing points of peaks 56 in the graphs 52 and 54 correspond
`
`to the point at which the aircraft 28 passes the sensor 27. Using the known position
`
`of the sensor 27 the time at which the aircraft passes, and the speed of the aircraft 28,
`
`a time can be determined for generating an acquisition signal to trigger the high
`
`resolution acquisition cameras 7. The speed can be determined from movement of the
`
`20 zero crossing points over time, in a similar manner to that described previously.
`
`The image acquisition system 10, as mentioned previously, acquires an image
`
`of the aircraft with sufficient resolution for the aircraft registration characters to be
`
`obtained using optical character recognition. According to one embodiment of the
`
`25 acquisition system 10. the system 10 includes two high resolution cameras 7 each
`
`comprising a lens and a CCO detector array. Respective images obtained by the two
`
`cameras 7 are shown in Figures 19 and 20.
`
`The minimum pixel dimension and the focal length of the lens determme the
`30 spatial resolution in the image. If the dimension of a pixel is Lp ' the focal length f and
`the altitude of the aircraft is h, then the dimenSIon of a feature W
`on the aircraft that
`mm
`
`Sony, Ex. 1002, p.1516
`
`
`
`W097137336
`
`PCT/AU97/00198
`
`- 24-
`
`is mapped onto a pixel is
`
`or
`
`(44)
`
`The character recognition process used requires each character stroke to be
`
`mapped onto at least four pixels with contrast levels having at least 10% difference
`
`from the background. The width of a character stroke in the aircraft registration is
`
`5 regulated by the ICAO. According to the ICAO Report, Annex 7, sections 4.2.1 and
`
`5.3, the character height beneath the port wing must be at least 50 centimetres and
`
`the character stroke must be 1/6th the character height. Therefore, to satisfy the
`
`character recognition criterion, the dimension of the feature on the aircraft that is
`mapped onto a pixel should be Wmin = 2 centimetres, or less. Once the CCO detector
`10 is chosen, Lp is fixed and the focal length of the system 10 is determined by the
`maximum altitude of the aircraft at which the spatial resolution W min = 2 centimetres
`is required.
`
`The field of view of the system 10 at altitude h is determined by the spatial
`15 resolution W min chosen at altitude hmax and the number of pixels N pi along the length
`of the CCO,
`
`(45)
`
`For h = hmax and N pi = 1552 the field of view is WFOV = 31.04 metres.
`
`To avoid blurring due to motion of the aircraft, the image moves a distance less
`
`20 than the size of a pixel during the exposure. If the aircraft velocity is v, then the time
`
`to move a distance equal to the required spatial resolution W min is
`
`(46)
`
`The maximum aircraft velocity that is likely to be encountered on landing or
`
`Sony, Ex. 1002, p.1517
`
`
`
`W097/37336
`
`PCT/AU97/00198
`
`- 25-
`
`take-off is v = 160 knots = 82 ms -1. With Wmin = 0.02 m, the exposure time to
`avoid excessive blurring is t < 240 ~s.
`
`The focal length of the lens in the system 10 can be chosen to obtain the
`
`5 required spatial resolution at the maximum altitude. This fixes the field of view.
`
`Alternatively, the field of view may be varied by altering the focal length according to
`
`the altitude of the aircraft. The range of focal lengths required can be calculated from
`
`equation (44)
`
`10
`
`The aircraft registration, during daylight conditions, is illuminated by sunlight
`
`or scattered light reflected from the ground. The aircraft scatters the light that is
`
`incident, some of which is captured by the lens of the imaging system. The
`
`considerable amount of light reflected from aluminIum fuselages of an aircraft can
`
`affect the image obtained, and is taken into account. The light power falling