GB2258780A - Target recognition and location - Google Patents

Abstract

An optical correlator responsive to targets independent of scale receives radiation from a field of view on the write face of a spatial light modulator 12 which is illuminated in its read face by a polychromatic light source 13, e.g. a multi-line Krypton-Argon gas laser or a white light source. A coherent Fourier transform of the field of view is then formed in a correlation plane on a single matched spatial filter 16 on which is recorded a Fourier transform image representative of the target at a preselected wavelength. Light transmitted by the filter means then provides an indication of the presence of the target. Information in the correlation plane is preferably sampled within different spectral intervals to detect the spectral interval corresponding to a peak of correlation from which the target size and hence range can be obtained. The multi- spectral processing may be done by scanning the optical spectrum at the source plane by means of an acousto-optic cell which scans the spectrum. <IMAGE>

Description

A Broadband Optical Target Detector The invention relates to optical detectors and in particular to detectors employing correlators for automatic target detection.

For many automatic detection tasks based on electro-optical imaging sensors both the size and orientation of the target image are unknown.

Since the usual matched spatial filters represent a single target size and orientation, adequate performance is only achieved for a limited range of these target parameters.

The object of the invention is to provide a detector which is insensitive to the scale of a target for a given target orientation.

A secondary object of the invention is to provide a detector capable of determining the range of a target of known size.

The invention provides an optical target detection system comprising: a spatial light modulator arranged to receive incoherent radiation at wavelengths within in a broad wavelength band from a field of view on a write face thereof; means to illuminate a read face of the spatial light modulator with polychromatic light and to form a coherent Fourier transform of the field of view in a correlation plane; a matched spatial filter located in the correlation plane to receive the Fourier transform of the field of view and having recorded thereon a Fourier transform image representative of the target at a preselected wavelength; and means to view the coherent Fourier transform of the field of view after modulation by the matched spatial filter and to detect therefrom the presence of the target.

The invention thus differs from the prior art arrangements in that it uses information from a target image illuminated by polychromatic radiation rather than quasi-monochromatic illumination.

A TV camera may be used to view the modulated image and a signal processor may be connected to receive an output from the camera for automatic target detection.

Preferably information in the correlation plane is sampled within different spectral intervals within the broad wavelength band to thereby detect a spectral interval corresponding to a peak of correlation in the correlation plane.

Once the colour (or wavelenth) of light corresponding to a correlation peak is determined then the presence of the target is verified and the angular subtense of the target can be found. From this the range can be determined if the absolute size of the target is known Preferably the coherence length of the read light source is greater than lO3m. It is preferable that the source covers a wide spectral range.

This latter requirement can be met by ensuring that there are spectral components distruted over the spectral band of interest. In one arrangement a multi-line laser, for example with a Krypton-Argon ion gas mix may be used. Such lasers can produce spectral components covering the range 336.6nm to 793.lnm.

In alternative arrangements true white light sources (polychromatic) can be used provided that appropriate signal processing is done in the optical domain of the correlation plane. Suitable sources might be halogen and arc discharge lamps. Other possible wide band light sources can include pico or preferably femto sec pulse lasers, wide band solid state devices (lets and lasers), the product of certain non-linear processes (eg Raman shifting of laser light in optical fibres or gas cells) or dye lasers. Preferably these should be small and this may be achieved by use of a demagnifying lens. Multi-spectral processing can be done by scanning the optical spectrum either at the correlation plane or at the input (coherent source) plane. Conveniently this is done at the source plane and advantageously use is made of an acousto-optic cell.

The invention will now be described with reference to the accompanying Drawings of which: Figure 1 is a schematic drawing of the basic target correlator image processing system; Figure 2 illustrates apparatus for producing the matched spatial filter used in the Figure 1 arrangement; Figure 3 illustrates the diffraction geometry at the surface of a liquid crystal light valve as used in Figure 1; Figure 4a shows diagrammatically the use of a read light source of finite size and Figure 4b shows the Figure 4a geometry unfolded; Figure 5 illustrates the effect of off-axis illumination; Figure 6 shows a scanning spectral source for use in the Figure 1 arrangement; and Figures 7a-d show graphically the performance of the detector.

Figure 1 shows a conventional optical image processing system according to the invention. Incoherent radiation 10 from an image in a field of view is incident on the 'write' face 11 of a spatial light modulator (SLM) 12. Light from a quasi-monochromatic laser 13 is used to form a coherent image from the 'Read' face 14 of the SLM 12. A parallel beam of light from the laser 13 is transmitted via lens L3 and then reflected from a polarising beam splitter 15 to the 'Read' face 14 of the SLM.

The coherent image from the SLM 12, after transmission through the beam splitter 15 is focused by lens L1 on to a matched spatial filter (MSF) in the image plane 16. The MSF has a stored representation of a target, representing a specific target size and orientation. A TV camera 17 views light transmitted through the MSF via lens L2. The image planel6 is where the coherent image of the field of view is correlated with the stored target representation on the MSF using relatively simple processing in an electronic processor 18, enabling automatic identification of any correlation point images.

The MSF is conventionally made using the arrangement shown in Figure 2.

An image of a target is incident on the 'write' face of the SLM 12. The laser 13 produces a reference beam 20, transmitted through the beam splitter 15, and a 'Read' beam 21 reflected from the beam splitter 15 to the SLM. Light reflected from the SLM 12 is combined with the reference beam 20 after reflection from mirror 22 in recording plane 23.

According to the present invention the MSF is made in the conventional way using a monochromatic light source as shown in Figure 2. The Figure 1 detector system, however, uses a polychromatic source in place of laser 13. As will be described below this renders the system insensitive to the range of the target.

Consider the case where the MSF is made using light of wavelength X1, and the whole optical correlator is subsequently used with light of wavelength 2.

In 'Introduction to Fourier Optics McGraw-Hill 1968' (Goodman J W) the amplitude transmittance of the correctly develped MSF is given by:

where k1 = constant rO = amplitude of recording reference beam #1 = recording light wavelength f = focal length of lens L1 N1 # = angle between the reference beam axis and the main optical axis of the system.

and H(x2 , y2 ) = Fourier transform of the target complex amplitude (ss1f ,f) distribution on the "read" face of the SLM when the MSF is being made.

Consider now the case when a complex scene image g(xyl) is formed on the "read" face of the SLM using a source of wavelength /\2 with the arrangement of Figure 1. The field amplitude transmitted by the MSF is:

where K2 amd K3 are constants.

It is the fourth term in the brackets

which leads to the correlation function.

Lens L2 in Figure 1 Fourier transforms U2(x21y2 to give the correlation function.

From equation (4) it can be seen that the result of using the optical correlator with light of wavelength #2 A2 is that the 'stored image' on the MSF of the sought-after target is scaled by the ratio "1/ 2.

Hence, if g(x,y) contains an image of a scaled target, there is a value of /\1/ AZ which suitably scales the MSF 'image' to give good correlation.

Thus, if the optical correlator is used with polychromatic illumination which does not violate the coherence requirements of the system (as described below), targets of all sizes lead to a correlation peak at the output plane, however the 'colour' of the correlation peak depends on the required scaling ratio A1/ #2- Note that from equation (4), regardless of the colour of the correlation peak, it always appears at a position corresponding to the target position in g(x,y).

To make best use of the information at the correlation plane, each spectral interval needs to be sampled separately. This can be done in a number of ways as will be described below.

Having identified the colour ( k2) of a correlation peak, then the scaling ratio %1/ ss 2 is easily calculated and hence the angular subtense of the detected target can be found. If the target absolute size is known, the range of the detected target can then be calculated.

For this technique to be of value, there are four main requirements to be satisfied. They are: suitable coherence characteristics of the correlator 'read' light, convenient sources for use with the optical correlator (eg, they must not be too big or too low in output irradiance), relatively simple wavelength sensitive processing of the multi-spectral correlation peaks and finally, the various optical components must perform adequately over the spectral band of use.

The arrangement shown in Figure 1 can only be used if the light used to 'read' the SLM satisfies certain coherence requirements. To simplify the analysis, the two effects (temporal coherence and source size) are treated separately.

Consider first the purely temporal coherence effects. If the source is such that the light transmited by lens L3 in Figure 1 is a plane wave (as is the case for a gas laser) even for short coherence-length light the illumination on the read face of the SLM is perfectly spatially coherent. This however is only true if the SLM face is flat compared with the coherence length of the light. The optical thickness of the liquid crystal layer in the Hughes LCLV, for example, is of the order of 10 micron (used in reflection mode), and non-uniformities in optical thickness over the read face are less than 1 micron. The modulus of the complex degree of coherence | t2(L)' is 0.82 for light of Gaussian spectral profile and coherence length lc equal to 5 micron when the path difference L is 1 micron.Thus in this case, light of coherence length greater than 5 micron can be used.

A more significant source of path difference is that introduced by light diffracted away from the normal of the LCLV by high spatial frequency components of - the image. For the system to act "coherently" the path difference L, for an image of size D producing a diffraction spot x in the Fourier plane as shown in Figure 3, must be less than the light coherence length Lc. In a practical arrangement for a maximum path difference of 2 x 10 -4m, it can be shown that light of coherence length, Lc > 10~3m is required. This restriction can be relaxed if the SLM read surface is suitably curved.

Assuming these temporal coherence requirements are satisfied, then the optical correlator (Figure 1) can be successfully operated using non-monochromatic light derived from a 'point' source.

However, the situation is more complicated when the 'read' light source 40 cannot be treated as a point source. Figure 4a illustrates a source of finite extent w and Figure 4b shows the geometry unfolded for ease of interpretation. There are two main types of extended source which are relevant to this general application. The first is one where the emissions from the entire surface retain spatial coherence (a laser diode for example). The second is one where the emissions are not spatially coherent across the surface (hot filament, gas discharge etc.) The former class tends to be spectrally narrow-band and is therefore not relevant to the discussion here; it is however relevant to compact systems which may use laser diodes. The latter class is more relevant as many spectrally broad-band sources fall into this category. The following relates to both classes.

It is appropriate to think of the extended source as a combination of infinitesimal elements, each emitting light independently of neighbouring elements. The manner in which the outputs from all the elements combine is dealt with later in this section. Due to its angular position with respect to the optical axis and the centre of lens L3, the light emitted by each element makes an angle 0 with the optical axis on passing through L3.Thus, as shown in Figure 4b, a path difference L is introduced between light arriving at the extremities of the target image of size D where # # L thus Lmax :t D 43nay (5) From Figure 4b w max ~ 2f thus eqn (5) becomes L'max = DW 2f (6) From the previous discussions about path differences introduced by diffraction effects, a value for Lmax = 2 x10-4m was obtained. If it is decided that the system should be designed such that the maximum path difference in the system is Lmax, then the largest acceptable source is found from eqn (6) by setting L'max = Lmax.This gives DW $ 2 x 10 4m 2f (7) If the light-source angular subtense relative to lens L1 is defined as # = W f (8) then Lmax < 2 x 10-4m, provided (using eqn 7): D < 4 x 10~4m (9) The Hughes LCLV has a usable diameter of about 4cm, and a target subtends about a tenth of this under normal operating conditions, hence D = 4-x 10-3m which from eqn (9) gives 0 # 10-1 Radians (10) Using eqn (8) with the lens focal length f=1O1m gives W 4 lOmm (11) Thus, if the source spatial extent criteria which led to eqn (11) are satisfied, the light from each infinitesimal element of the source maps 'coherently' to the optical correlation plane.

Performance of the system for off-axis illumination when the source 'elements' are not all on the system optical axis needs to be considered. The off-axis source elements yield displaced Fourier transforms of the scene, leading to a complex exponential (phase) term.

This reduces the intensity of the correlation peak at target detection.

Generally, it can be shown that the source must be small.

Figure 5 shows the source element 51 distance Wx/2 from the optical axis causes the Fourier transform of the scene to be displaced by an amount x'. For simplicity, the scene is shown as a transparency 52. This arrangement is equivalent to that based on the LCLV, but involves a less complicated geometry. From Figure 5 it can be seen that Wx = x' 2f3 f1 (12) Since the source has spatial extent along the y''-axis also, the most extreme source element at (x " ,y " ) = (Wx/2, Wy/2) yields a displacement of the Fourier transform of the scene such that the zero frequency component is at (x',y').Thus the complex amplitude function at the Fourier plane due to this element is given by eqn (3) as

which can be shown to give the corresponding correlation plane function as

where f1 = f2 = f Comparing eqns (4) and (13) it can be seen that the displacement of the Fourier transform of the scene introduces a complex exponential (phase) term. Hence U3(x3,y3) cannot be the true correlation function between the scene and the 'stored'(via the MSF) target image. The effect of the phase term is to reduce the intensity of the correlation peak at target detection.

The effect of the phase term can be ignored if for all x' and y' 2 # [(x-x3)x + (y-y3)y'] < 1 (14) Since 1y-y Since | x-x| max # |y-y3|max corresponds to the radius of the target image (there being no significant contribution to the correlation integral beyond the physical extent of the target image centred at (x3,y3)) then using the following values:: |x-x3|max = |y-y3|max = 2x10-3m (target diameter = 4 mm) (#2)min = 4 x 10-7m fmin = 1O1m gives from (14) x' and y' < 1.6 x 10~6m (15) From eqns (12) and (15), the size of the source behind lens L3 in Figure 5 must be given by Wx < 2 x' t (16) f1 (16) Thus unless f1 f3 or 'W ' is a de-magnified image of the actual source, it can be seen that the source must be small; much smaller than implied by the temporal coherence considerations which led to eqn (11).

A further consideration concerns the way in which individual outputs from each source element combine. Assuming the temporal coherence criteria described earlier on in this section are satisfied, the answer to the question depends on whether the source is spatially coherent across its surface. If it is, then the individual elements combine as a super-position of complex amplitude, and the intensity profile at the

correlation plane forweach spectral component is given by I1(x31y3) = I j{ S(x",y") JJ S(x ,y ) U3(x :x",y3:y")dx"dy" 12 (17) where S(x",y") is the complex amplitude distribution at the source plane.Note, from eqn (12), the source plane co-ordinates (x",y") are related to the shifted Fourier transform plane co-ordinates (x',y') are related to the shifted Fourier transform plane co-ordinates (x',y') by x" = x' f1 = f3 y' f1 (18) If the light emitted from the source surface is spatially incoherent, then the intensity profile at the correlation plane for each spectral

component is given by I (x3,y3) = I3(x",y") 1u,(x3 12 12 12 dx" dy" (19) 1U3(x3:x",y3:y") where I (x",y") is the intensity distribution at the source plane.

For a polychromatic source, be it spatially coherent or not, the individual spectral components combine to give the overall intensity profile at the correlation plane.

where I > ( X,) is the spectral profile of the source.

Although eqn (20) describes the final intensity profile at the correlation plane, the technique of the present invention requires the intensity at the individual spectral component to be analysed. The manner in which this can be done is escribed later.

SUITABLE SOURCES As described above, various requirements for the 'read' light source were established which must be satisfied for the present technique to be viable. The requirements are: i. source coherence length > 10-3m ii. the maximum size of the source prior to the collimating lens (see Figure 5) is given by Wx and Wy < 3.2 f + 10-6m iii. the source does not have to be spatially coherent across its emitting surface.

These requirements were based on the following system parameters: a. Minimum source wavelength = O.4pm b. Maximum source wavelength = 1.Opm c. LCLV working-area diameter = 4 cm d. Minimum value for fl and f2 (lenses L1 and L2 focal lengths - see Figure 4) = 10 cm.

e. The product of target-image size and maximum spatial frequency at the LCLV < 200.

Two other requirements not mentioned above are: iv. The source must cover a wide spectral range to allow the necessary scaling effect which forms the basis of the technique.

v. The source must be sufficiently intense to ensure there is enough light to be detected at the correlation plane.

It appears that requirements (i) and (iv) are mutually exclusive.

This is not the case however, for two reasons. Firstly, the requirement that the source emissions cover a wide spectral range need not mean emitting light at all wavelengths within that range. All that is necessary is that there are spectral components (each emission line can be very narrow and hence have long temporal coherence) distributed over the waveband of interest, the restrictions on spectral separation of the individual lines is discussed later. Secondly, if truly polychromatic light is spectrally band-limited, the temporal coherence of the band-limited light is different to that of the polychromatic light.

This phenomenon can be used to advantage.

One way of satisfying all the requirements is to use a multi-line laser, for example with a Krypton-Argon ion gas mix. These lasers can be run using multi-line mirrors to give spectral components from 333.6 mn to 793.1 nm.

It has been shown that a single MSF produces adequate correlation for target size changes of up to 20%. Thus referring to eqn (4), scaling wavelength ( A1/ A2) are only required for ( < 1) = [200 + D](i-l) ( > 2)i [200 - p] i = 1 - > i,ax (21) where P = percentage change in scale that can be accommodated by the MSF for monochromatic illumination. Also, it is assumed that the MSF is made at > = maximum wavelength such that all: ( #1) # 1 ( /\2)i ie the recorded MSF corresponds to the largest expected size of the target image.

Thus: ( #1) = ( #1 ) = (#1) (#2)i max (#min) (#2)max For good performance, setting P = 10; (not 20% which represents the limit of adequate performance), using ( A,/ > 2) max = (1.0 m/0.4 pm) = 2.5 and solving eqn (21) for i = imax gives imax#10.2 (22) Thus approximately 10 spectral lines equally spaced from 0.4 pm to 1.0 pm allows changes of 150%.

Although a Krypton-Argon laser can meet all the performance requirements, all known sources of such lasers are quite large, the smallest known device is an all-lines Argon laser which is approximately 30 cm long by 15 cm wide and 15 cm high; the controller which can be remote from the laser head is about the same size and requires single phase 240 V mains (no cooling water is needed) at a maximum of 10 amps.

The all-lines output of the laser is between 50 and 100 milliwatts; a Krypton Argon mix in this laser would probably have a lower total output power.

True white light (polychromatic) sources can be used provided the appropriate signal processing is used in the optical domain of the correlation plane. The requirement is for an intense but small source.

There are a number of sources which might be suitable, eg halogen lamps, arc discharges. The main problem with these sources is that they are all relatively large. Size restrictions can be reduced to some extent by using a de-magnified image of the source at the source plane in Figure 5. It can however be shown that as the image size is reduced, the divergence angle of light leaving the image increases leading to a reduction in usable source light due to vignetting: The extent to which source (image) size and intensity can be traded-off depends on the responsivity and noise characteristics of the correlation plane detector.

SIGNAL PROCESSING Signal processing can be used for processing the multi-spectral information at the correlation plan.

Many schemes for processing the correlation plane are possible which may make better use of a-priori information about the correlation to maximise discrimination and signal to noise ratio. For example, known techniques based upon schemes that make use of both spectral and spatial information in the optical domain may be useful.

However, to investigate multi-spectral processing relatively simple arrangements are suitable. The simplest techniques for the purpose of verifying the general principles require the optical spectrum to be scanned. This can be performed at either the correlation plane or at the input (coherent source) plane. Since the diameter of the light beam is smallest at the source plane, techniques for scanning the wavelength of the input illumination are likely to be the most convenient. In Figure 6 an arrangement for scanning wavelength is shown which is based upon an acousto-optic (A-O) cell 60. Light from a source 60 is transmitted via a first lens 63 to illuminate the A-O cell 60 driven by an oscillator 64.Modulated light after transmission through the A-O cell 60 is focussed by a second lens 65 to a slit 61 and then expanded by a third lens 66 to produce an output beam 67.

If the slit 61 shown in Figure 6 is placed distance x from the zero order position of the acousto-optic cell Fourier transform then wavelength of the light passing through the slit is given by x = fm (23) Vm = = xVm dfm where Vm = the speed of sound in the acousto-optic cell.

d = distance between the slit and the Fourier transforming lens next to the cell.

fm = the acousto-optic cell electrical signal frequency.

From eqn (23) it can be seen that the wavelength of light passed by the slit 61 in Figure 6 can be selected by varying fm. Thus if fm is varied at a rate corresponding to a full wavelength scan in the optical correlator dwell time (typically the spatial light modulator rise-time).

Thus, for any intense point of light at the optical correlation plane which crosses the set threshold a plot of normalised peak (for wavelength dependent source intensity) intensity above threshold versus wavelength (derived from acousto-optic cell electrical frequency fm) can be made. Figure 7 shows the general profiles of various correlations as functions of wavelength and position. Figure 7a shows the intensity profiles 70,71 for a poorly correlated feature at two wavelengths Aa and Best correlation ('sharpest' correlation feature) occurs at whilst very poor correlation is at Aa. . The corresponding intensity versus wavelength plot is given in Figure 7b. Figures 7c and 7d show graphs similar to 7a and 7b corresponding to a well correlated feature.

Comparing Figures 7b and 7d shows that the intensity difference and gradient can be used to identify well correlated features. The wavelength at which best correlation occurs defines target size from the wavelength scale ratio.

Since the target shape is known (it is stored via the MSF) and so is the scale ratio, then the target range can be calculated. Hence, the wavelength for peak correlation corresponds directly to target range.

The main components of the system are the spatial light modulator (SLM) the lenses and the matched spatial filter (MSF) and these should be designed such that performance as a function of light wavelength is good at all select wavelengths.

The arrangement described is one of a number of possible arrangements.

Another arrangement could employ spectral dispersion just after the beam splitter in Figure 1 to produce a wavelength dependent spatial displacement of image elements. In this manner, any one point in the SLM polychromatic image plane is transformed by the spectral dispersion such that the point appears simultaneously at all points in the dispersal image plane. It should therefore be possible to use techniques such as polar and Mellin transformations (which require the image to be on-axis) since at least one of the coloured 'images' is on axis; the colour of a resulting "correlation spot" corresponding to the location of the target in the image plane.

Furthermore, positional encoding may also be possible by using polarisation as well as spectrum. In alternative arrangements spectral dispersion could be employed.

Claims (13)

Claims

1. An optical target detection system comprising: a spatial light modulator arranged to receive incoherent radiation at wavelengths within in a broad wavelength band from a field of view on a write face thereof; polychromatic light source means to illuminate a read face of the spatial light modulator with light and to form a coherent Fourier transform of the field of view in a correlation plane; a matched spatial filter located in the correlation plane to receive the Fourier transform of the field of view and having recorded thereon a Fourier transform image representative of the target at a preselected wavelength; and means to view the coherent Fourier transform of the field of view after modulation by the matched spatial filter and to detect therefrom the presence of the target.

2. An optical target detection system as claimed in claim 1 wherein a TV camera is provided to view the modulated image and a signal processor is connected to receive an output from the camera for automatic target detection.

3. An optical target detection system as claimed in claim 1 or 2 wherein information in the correlation plane is sampled within different spectral intervals within the broad wavelength band to thereby detect a spectral interval corresponding to a peak of correlation in the correlation plane.

4. An optical target detection system as claimed in claim 3 wherein the signal processor determines from the wavelenth of light corresponding to a correlation peak the angular subtense of the target.

5. An optical target detection system as claimed in any one preceding claim wherein the coherence length of the read light source is greater than 10-3m.

6. An optical target detection system as claimed in any one preceding claim wherein the source covers a wide spectral range.

7. An optical target detection system as claimed in claim 6 wherein the source has spectral components distruted over the spectral band of interest.

8. An optical target detection system as claimed in claim 7 wherein the source is a multi-line Krypton-Argon gas laser.

9. An optical target detection system as claimed in claim 6 wherein a white light source be used.

10. An optical target detection system as claimed in claim 9 wherein the source is a halogen or arc discharge lamp.

11. An optical target detection system as claimed in any one preceding claim wherein a demagnifying lens is used to reduce the size of the source small.

12. An optical target detection system as claimed in any one preceding claim wherein multi-spectral processing is done by scanning the optical spectrum at the source plane.

13. An optical target detection system as claimed in claim 12 wherein an acousto-optic cell is used to scan the spectrum.