GB2607054A - Stand-off screening system and method of the same - Google Patents

Abstract

A stand-off screening system 100 for determining an indication that a person is carrying a concealed knife. The system comprises: an array of radar transmitters and receivers 114, each configured to transmit radar waves and receive reflected radar waves; and a processor, connected to the array of radar transmitters and receivers, and configured to utilise one or more properties of a reflected signal derived from one or more received reflected radar waves to determine the indication as to whether the person has a concealed knife. The properties preferably including one of more of a shape, a width of a diffraction angle and an amplitude of the reflected signal. The determination that a person has a concealed knife comprising

Description

STAND-OFF SCREENING SYSTEM AND METHOD OF THE SAME

Field of the Invention

The present invention relates to a stand-off screening system, and a method of screening people using the stand-off screening system.

Background

Knife crime us problem in many communities worldwide. The detection of knives being carried covertly has an important role to play in reducing the scope and harm of this problem. However, there is a strong desire to avoid manually searching large numbers of people. This is because it is both impractical due to the numbers involved, and also difficult from privacy and consent viewpoints (including the need to avoid any perceived victimisation).

Walk-through and hand-held metal detectors have been used previously. They are effective in situations where a benign individual is expected to carry few small or no metal items, and to divest them for inspection when challenged. There is also a requirement that people are directed through or past the metal detector in a well-controlled way, for example at an airport. However such detectors are not suited for application outside of these strictly controlled settings, where people are carrying everyday metallic items such as keys, phones, or wallets.

Stand-off screening systems i.e. systems which does not require physical interaction with a person being screened, partially address this by allowing the remote screening of individuals. For example as they walk past the screening system. An issue arises with conventional stand-off scanners (for example as disclosed in US 10,948,587 81) in that knives are easily confused with similar sized non-threat items such as cellular phones, keys, coins, etc. There is a need then for a stand-off screening system which can reliably detect knives, and is able to discern between knives and benign objects.

Accordingly, in a hrst aspect, embodiments of the invention provide a stand-off screening system for determining an indication hat a person is carrying a concealed knife, the system comprising: an array of radar transmitters and receivers, configured to transmit radar waves and receive reflected radar waves; and a processor, connected to the array of radar transmitted and receivers, arid configured to utilise one or more properties of a reflected signal derived from one or more received reflected radar waves to determine the indication as to whether the person has a concealed knife.

Advantageously, such a system demonstrates an enhanced capability of scriminating between knives and benign objects.

Optional features of the invention will now be set out. These are applicable singly or in any combination with any aspect of the invention.

The properties of the reflected signal may include data indicative of a diffraction pattern within the reflected signal, for example data indicative of spatial broadening of the reflected signal due to diffraction. The properties of the reflected signal may include one or more of: a shape of the reflected signal; a width of a diffraction angle of the reflected signal; and an amplitude or an intensity of the reflected signal.

The radar transmitters and receivers may be provided as transceiver units, i.e. a radar transceiver configured to transmit radar waves and receive reflected radar waves. The transceiver units may be operable to both transmit and receive radar waves. In some examples, only a subset of the transceiver units transmit radar waves.

The indication may be a discrete value provided to an operator indicating whether the person is predicted to be carrying a concealed knife, e.g. "YES" or "NO". The indication may be a continuous value provided to the operator providing the determined likelihood, e.g. "80%". The indication can also include an indicator of where the concealed knife is predicted to be on the person (for example, a graphical indication of a person with an area highlighted) The processor may be located centrally, e.g. physically connected to the array of radar transmitters and receivers. The processor may comprise a plurality of processing units which operate in parallel or in sequence. The processor may be located in a remote computer, and connected via a computer network.

The stand-off screening system may comprise a feedback unit, configured to provide the determined indication to an operator. For example, the feedback unit may be a screen configured to display the indication, a light source configured to illuminate when the determination is that the person is carrying a concealed knife, and/or an audio source configured to emit an audio signal when the determined is that the person is carrying a concealed knife. The stand-off screening system may further comprise a camera, configured to take a picture of the person being scanned. This picture can be provided to the operator.

The stand-off screening system may include a plurality of arrays of radar transmitters and receivers, the individual radar transmitters and receivers within an array extending in a first direction and the arrays extending in a second direction transversal to the first direction. in some examples; the arrays are arranged in an arc and so extend in a third direction transversal to the first and second. One or more array of the plurality of arrays may be angled relative to adjacent arrays. For example, an array furthest from the around may be angled down and an array closest to the ground may be angled up so tne interrogation areas of the plurality of arrays overlap. Each plurality of arrays may provide a bank of radar transmitters and receivers, and the system may comprise a plurality of banks of radar transmitters and receivers, the banks being separated in the first direction. Such a system increases the spatial area covered. Each bank may be angled relative to one or more adjacent banks. For example, one or more peripheral banks may be angled relative to a central bank such that the interrogation areas of the banks overlaps.

Each radar transmitter and receiver may include a respective controller, or the array of radar transmitters and receivers may be divided into a plurality of subsets of radar transmitters and receivers, each subset connected to a respective microcontroller; and each microcontroller may be configured to pre-process radar data before sending it to the processor. This can reduce the computational load on the processor, and so increase the framerate at which it analyses radar data The radar transmitters may be configured to transmit radar waves at a frequency of at least 3 GHz, or at least 10 GHz, or at least 24 GHz, and no more than 300 GHz, or no more than 100 GHz or no more than 77 GHz, or no more than 64 GHz, or combinations thereof. In some examples, the frequency is at least 24.0 GHz and no more than 24.25 GHz or at least 57 GHz and no more than 64 GHz, or combinations thereof. The radar transmitters may be configured to sweep in frequency.

The system may further comprise a central oscillator, preferably running at a frequency lower than a frequency at which the radar waves are transmitted by the radar transmitters; and each radar transmitter and/or receiver may be configured to synchronize to the central oscillator. This results in a frequency lock between the central oscillator and each radar transmitter and receiver. This can further ensure that the signals received by the processor are synchronized in frequency. Accordingly, full phase coherence between all of the receivers is not required. A significant cost saving can therefore be realised, as compared to systems in which full phase coherence would be required. The central oscillator may be configured to run at a frequency of no more than 24 GHz, or at least 1 MHz and no more than 900 MHz.

The radar transmitters may be configured to utilise multiple transmission channels; the multiple transmission channels may be separated in frequency and/or time. Sets of radar transmitters in respectively different frequency-separated transmission channels may be configured to transmit simultaneously. In some examples the multiple transmissions channels are separated in frequency and then swept in time and/or frequency. In some examples, the radar transmitters and receivers may be configured to multiplex in time (e.g. one set of transmitters performs a sweep, then another set of transmitters does a sweep). This multiplexing can allow he person being screened to be illuminated from different angles simultaneously, whilst not incurring interference between sets of radar transceivers.

The system may further comprise a camera, for example a 20 camera, 3D camera; or: set of cameras configured to determine the position of the person relative to the array of radar transceivers. The processor may be configured to derive expected properties of a diffraction signal derived from one or more received reflected radar waves utilising the determined position, and to utilise this in the determination of the indication as to whether the person has a concealed knife. For example, the processor may be configured to derive one or more of: an expected shape of the reflected signal; an expected width of a diffraction angle of the diffraction signal; and an expected amplitude of the diffraction signal. These expected properties may be derived on the assumption that the person is not carrying a knife, and then the expected properties may be compared to the properties determined from the received radar reflections.

The system may further comprise baffles and/or radar absorbing film configured to reduce cross talk between the radar transceivers. The baffles and/or radar absorbing film may be located between respective radar transceivers of the array. This can improve the signal to noise ratio, and so further enhance the ability with which the system discriminates between objects.

The processor may be configured to utilise a machine classifier to determine the indication as to whether the person has a concealed knife. The machine classifier may be one or a combination of: a random forest classifier, a convolutional neural network, and an inverse ray-tracing algorithm.

The processor may be configured to determine the indication as to whether the person has a concealed knife utilising one or more of: a fraction of a plurality of frames of received radar waves where an amplitude is above a predefined threshold; for example twice a background noise amplitude; an average peak width; a variance in the peak width; and data indicative of how a peak of the received radar wave varies in time (for example; how smoothly the position of the peak on the array of radar receivers moves in time). The processor may be configured to determine the indication when an amplitude of the reflected signal exceeds a screening threshold, the screening threshold may be determined based on a representative threshold indicative of an expected reflectance of a person at a given range. The average peak width may be approximately double the reflection received from a person's body.

In a second aspect; embodiments of the invention provide a method of determining an indication that a person is carrying a concealed knife using the stand-off screening system of the first aspect; the method comprising: (a) illuminating the individual with radar waves transmitted from one or more of the radar transmitters; (b) receiving reflected radar waves at one or more of the radar receivers; (c) deriving one or more properties of a diffraction signal from one or more of the received radar waves; and (d) determining, by the processor, the indication as to whether the person has a concealed knife based on the derived one or more properties of the diffraction signal.

The invention includes the combination of the aspects and preferred features described except where such a combination is clearly impermissible or expressly avoided. For example the method of the second aspect may include any one, or any combination insofar as they are compatible, of the optional features set out with reference to the first aspect.

Further aspects of the present invention provide: a computer program comprising code which, when run on a computer, causes the computer to perform the method of the second aspect; a computer readable medium storing a computer program comprising code which, when run on a computer, causes the computer to perform the method of the second aspect; and a computer system programmed to perform the method of the second aspect.

Brief Description of the Drawings

Embodiments of the invention will now be described by way of example with reference to the accompanying drawings in vvhich: Figure 1 shows a schematic of a stand-off screening system; Figures 2A -20 show various arrangements of the stand-off screening system of Figure 1; Figure 3 shows a variation in the arrangement of the stand-off screening system of Figure 1; Figure 4 shows recorded data ndicative of the strength of signals received at a plurality of transceivers; Figures 5A -5D show plots of returned signal against position for, respectively, a knife, a smartphone, a metal bottle, and a folded umbrella; Figures 6A and 6B show plots of diffraction width events for knifes and phones in a simulated system respectively; Figures 7A and 7B show plots of diffraction amplitude events for knives and phones in a simulated system respectively; Figure 8 is a ROC curve of a machine classifier trained on real-life data; Figure 9 is an architecture used for a convolutional neural network; and Figure 10 illustrates B kernels learnt by the CNN model.

Detailed Description and Further Optional Features

Aspects and embodiments of the present invention will now be discussed with reference to the accompanying figures. Further aspects and embodiments will be apparent to those skilled in the art.

Figure 1 shows a schematic of a stand-off screening system 100. The system includes a user interface 102, connected to an array controller 104. The array controller is connected to both a 3D camera 106, and each of a plurality of multi-transceiver module (MTM) controllers 108. Each MTM controller is connected to a plurality of single-transceiver modules (STM) 110.

The STMs each include a transceiver controller 112 and a transceiver 114. Each STM is a fully self-contained radar and includes all of the peripheral electronics and antennae required to provide a radar transceiver, as well as a rnicrocontroller all on a single board. For example, each STM may be based on an lnfineon BGT24MTR lx radar sensor, along with a suitable microcontroller, phase lock loop, and set of antennae, A plurality of STMs 110 and a respective MTM controller 108 are combined into an MTM (e.g. MTM*I, MTM2, MTMn). For example, the STMs of a given MTM may be mounted onto a single MTM backplane containing the MTM controller 108 In one example, an MTM includes eight STMs. The MTM controller contains the power, control, and interfacing electronics to operate all of the STMs of that MTM simultaneously. The MTMs also distribute synchronisation signals so that the STMs of a given MTM are operatable as a multi-static array (i.e. a spatially diverse array of multiple components with a shared area of coverage). Each MTM in the system can be slaved to another MTM, so that all transceivers across the system are synchronized and therefore included in the multi-static array. Each rviTm is also operable as a standalone module. The STMs in this example were connected to their respective MTM controller using a multi-drop universal asynchronous receiver-transmitter (UART) scheme. This connection scheme is more robust than, e.g. USB and other connection schemes, and further allows the timing of signals to be more deterministic.

The array controller 104 sits above the MTMs, and is connected to all of the MTMs and the 3D camera 106. The array controller configures and triggers the MTMs to transmit radar waves, and processes data received from the MTM controllers. In one example, each STM performs a fast Fourier transform (FFT) of the raw radar data received and sends an envelope of the FFT via the MTM to the array controller. The raw data may include, for example, voltage-vs-time data received by a given transceiver and may be converted into power-vs-frequency data by the EFT. The processing performed by the array controller includes the determination of the indication as to whether the person has a concealed knife.

The array controller 104 also controls the 3D camera; which is used to detect when a person is in the field of view of the array and what distance they are from the array. The array controller is also connected to the user interface 102 which provides the determined indication to the user e.g. via a visual, haptic, and/or audio notification The 3D camera in this example is an Intel RealSense D435 camera. The 3D camera is configured to capture 3D depth video of a person as they walk past the MTMs. The 3D camera determines whether there is a person close to the MTMs and measures the approximate position of the person.

In some examples, the system includes an additional circuit board used to generate a 40 MHz clock signal and a trigger signal which is distributed to each STM in the array. The 40 MHz clock is used to synchronize the rnicrocontrollers on each STM, and to establish a common frequency reference for the radar local oscillator frequency control circuitry on each STM. Phase coherence of the radar need not be maintained across multiple STMs, but the 40 MHz signal is easier to distribute than a higher frequency at which the STMs would otherwise operate (e.g. 24 GHz). Notably, unlike a conventional imaging radar focal plane array, the system does not require phase synchronization. When a trigger request is made by the array controller 104, the trigger signal is generated and distributed to each of the STMs simultaneously. The STMs all stream data through the system to the array controller 104 and are synchronized so that they simultaneously operate with each of the transceivers detecting and differentiating radar signals from multiple transmitting channels. In one example, there is a total of 128 STMs located across 16 MTMs.

In one example, the transceivers were swept with a consistent frequency offset. This is accomplishing by using the same frequency ramp, but setting the start frequencies differently. This pushes the signal received at the receiving STM up in frequency; and therefore into an optimal working range for a baseband amplifier included in the SIM. This can result in the received radar signal appearing at an artificial distance offset, which is corrected by the MTM controller or array controller. By adopting this approach; the received signals are significantly stronger. A further benefit of this is that the frequency offset can be used to operate multiple transceivers at the same time. Typically; only one or two closely spaced peaks are detected as reflections from a person. Therefore two or more transceivers can be used to transmitted with different frequency offsets, and received by other STMs simultaneously. The signals from each transceiver are decouplable in frequency space, and therefore illumination can be performed from multiple angles simultaneously. This further increases the detection performance, as the likelihood of detecting a diffraction pattern is increased.

Figures 2A -2D show various arrangements of the stand-off screening system of Figure 1. In Figure 2A, four banks of MTMs 204a -204d (each bank containing a plurality of MTMs) are located along a walking path 202. Each in turn scans a person walking along path 202.

In Figure 2B, the four banks are located in two pairs: 204a and 204b, and 204c and 204d.

The first pair are located on a first side of the walking path 202, and the second pair are located on a second side of the walking path 202. This allows two of the banks, e.g. 204a and 204c, to simultaneousiy scan a person from both sides.

In Figure 20 there are again tvvo pairs of banks as in Figure 2D. However, in Figure 20, the banks are each aligned so as to all share a common interrogation region 206. An individual walking along path 202 can thereby be scanned from four different sides simultaneously. Figure 2D is a variation of this concept, where the walking path 202 is varied so that the person walks past all four banks, and so can be scanned from four different sides sequentially.

Figure 3 shows a variation in the arrangement of the stand-off screening system of Figure 1.

It comprises four banks of MTMs 204a -204d located within a single unit projecting from the ground 302. The two central banks 204b and 204c are neutraily angled, so that they transmit radio waves at an angle of around 90' to the ground (i.e. along the plane of the ground). Whereas the upper bank 2042 and lower bank 204d are angled relative to the two central banks, The upper bank 204a is angled down, so that its interrogation area (indicated by the dashed line) substantially overlaps with the interrogation areas of the two central banks 204b and 2040. The lower bank 204d is angled up, so that its interrogation area (indicated by the dashed line) substantially overlaps with the interrogation areas of the two central banks 204b and 204c. As knives are typically not carried very close to the head or feet, disposing the bank of MTMs as shown in Figure 3 further enhances the degree to which the system can discriminate between knives and benign objects.

Figures 4A --40 show plots of returned signal against position for, respectively, a knife, a smartphone, a metal bottle, and a folded umbrella. The plots show diffraction patterns for ultrasound pulses (used as a proxy for radar, as the wavelength is more easily adjustable) which demonstrates that properties of the diffraction pattern can be used to discriminate between objects. Note that the srnartphone produces narrower peaks than the knife, and that objects without a flat face (e.g. the metal bottle and folded umbrella) do not produce any clear peak. The physical width of the pattern is also governed by the distance between the object and the receiver. A larger distance between the radar and object leads to a wider pattern.

Figure 5 shows the strength of signals received at a plurality of transceivers. The first step of processing the data, which is performed frame-by-frame, is to clean or pre-process the raw received signals from each STM. For example, by subjecting any static background and correcting for any inter-STM variation in gain. The result of this pre-processing is a single amplitude for each STM representing the amount of power that is reflected from objects within the field of view of the system. This is shown in Figure 5, with the size of a circle indicating the strength of the signal received at the given STM.

Figures 6A and 6B show plots of diffraction width events for knifes and phones respectively from data collected using the system. As can be seen from the plots, the peaks of the diffraction patterns associated with a knife are on average significantly broader than the peaks of the diffraction patterns associated with a smattphone.

Figures 7A and 7B show plots of diffraction amplitude events for knives and phones respectively. As can be seen from the plots, knives give significantly higher amplitude (e.g. brighter) diffraction patterns than phones. The plots demonstrate that knives reflect more strongly than phones, but phones have a broader range of radar reflectively (in frequency).

This further demonstrates that these parameters are usable to differentiate between knives and benign objects.

The determination of an indication is now described. If the power received by a given STM is above a pre-determined threshold, for example twice the measured background signal level, then the STM is flagged for further investigation. This results in a list of STMs who's signal is to be further investigated, which is typically significantly smaller than the entire array. STMs which are directly adjacent to flagged STMs are also included for investigation, as it gives a useful indication of the spatial extent of the diffraction pattern.

Next, it is determined whether there are one or more spatially separated diffraction patterns to be investigated. It may be that separate reflections are detected from separate objects (for example, a closely located knife and phone on a person). A spatial clustering approach is used, for example DBSCAN (Density-based spatial clustering of applications with noise as disclosed by Ester et al in "A density-based algorithm for discovering clusters in large spatial databases with noise", AAAI Press. 1996, pp 226 -231). This specific approach does not require the pre-definition of the number of clusters to be extracted and is relatively robust to outliers. In one example, conditions were set on the clustering which included: STWIs needed to be no more than 25 cm apart to be within the same cluster; and a minimum of 3 STMis are in any given cluster.

This process results in one or more clusters of STIVa. The position of each SIM in the cluster, together with its amplitude, is used to fit a set of parameters to the cluster. In one example, the parameters were the horizontal and vertical width and amplitude of the signal. A 2D Gaussian function was used. This results in several numbers produced by the fit: the number of STIvls used for the fit, the peak amplitude of the diffraction pattern, the horizontal and vertical width, and the horizontal and vertical position of the centre of the diffraction pattern on the radar array. Information indicating where the subject is (from the 3D camera) is used to correct for distance., and so turns the spatial widths into angular widths, and to compensate for a decrease in amplitude with increased range.

These features are then passed to a classification algorithm to determine the indication that this specific combination of features belongs to a knife reflection pattern or a benign reflection pattern. The classifier is pre-trained on a representative labelled data set in order for it to determine the optimal weightings or coefficients to put the training data into the correct category. The data was acquired through use of the system in Figure 1, where various persons carrying knives or benign items (or no items) were scanned by the system and supplemented with simulation data. In some examples, the simulation was based on an analytical model of the reflected power from a knife or other objects. The object under illumination from a simulated radar transmitter was broken down into small elements (smaller than the radar wavelength). A wavelet Huygens-style approach was used, moving from an emitted to the reflecting element and then to a receiver. The phase and amplitude were tracked. This allows for the computation (For example through summation and squaring) of a result proportional to the total amount of radar power reflected by the knife from the emitter to the receiver. The simulation of the system could be performed using computational electrodynamics or electromagnetic modelling of types known per se in the art.

A number of classifier methods can be used, from logistic regression through to neural networks. A Random Forest classifier was trained, which is an ensemble method where many decision tree classifiers are created at random from random subsets of features and a random subset of training data. To test the performance of the classifier, the available data was split into a training and testing set (in one example, 80:20), and performance was evaluated on unseen data to avoid bias.

It was observed from the simulated data that in frames where a notable diffraction pattern was detected, the classifier was able to distinguish between knives (the dataset comprising four different knives) and the smartphones (the dataset comprising three different smartphones) with an accuracy of over 90%. This represented a potential true positive rate (TFR) of over 90% and a false positive rate (FFR) of around 5%. This performance was arrived at only on a single frame-by-frame basis, and performance is improved if averages are taken over a plurality of frames of radar data.

Figure 8 is a ROC curie generated from the random forest classifier, using 729 real-life data runs of which 362 comprised knives and 367 comprise benign objects. Of note is that in the low sensitivity region of the ROC curie, the curve itself is very steep. This is indicative that the false alarm rate of the system is very low whilst still operating an appropriate sensitivity.

For example, per the ROC curve shown in Figure 8, the system detected 60 -80% of knives with a very low false positive rate.

Figure 9 is an architecture used for a convolutional neural network or CNN, The CNN is an alternative classifier which was trained on the data available. As is known per se in the art; a CNN is a deep learning architecture in which convolutional filters, or kernels, are applied to the input data. The kernel combines with the input data typically through use of a dot-product. The output of each frame of data from the system, in some examples, is a 32x4 pixel grayscale image. The size of the convolutional kernel in the initial layer was chosen to be 8x2 pixels, which reflects the fact that the width of the diffraction pattern was found to be significantly more important than the height (i.e. amplitude) when distinguishing between knives and phones.

The architecture shown in Figure 9 achieved good results from the simulated data. The initial layer used an 8x2 convolutional filter to produce 16 hidden feature maps in the first hidden layer. This was followed by a 2x2 max-pooling layer with stride 2, to reduce dimensionality and complexity. A second convolutional layer with another 8x2 kernel was applied before final fully connected layer condensed the network into the binary output space. Cross-entropy loss was used with the Adam optimizer.

Figure 10 illustrates the 8 kernels learnt by the CNN model of Figure 9, with one convolutional layer feeding into 8 hidden feature maps. The kernels are shown in the left-hand column with the sum of the weights assigned to that kernel shown in the right-hand column (blue bars, e.g. 902, show positive values whilst red bars, e.g. 904, show negative values). Large blue bars in the right-hand column therefore snow kernels which are highly indicative of a knife being present. In kernels 3 and 4, which were found to be by far the most important for positive identification, wide horizonal glint can be observed.

Table 1 1 below shows the results for three different CNNs with increasing numbers of convolutional layers: Accuracy Speed N layers Training set Validation set Test set Training time Run speed (fps) 1 87.20% 86.20% 83.70% 20 minutes 71500 2 97.20% 94.40% 94.30% 1 hour 63500 97.30% 95.10% 94.50% 2 hours 56000

Table 11

In the three models shown above, it can be seen that performance increases rapidly when the number of convolutional layers is increased from 1 to 2. The two models with higher numbers of hidden layers showed excellent levels of discrimination with accuracy scores on the test set greater than 94%. Further the run speed shows that they are suitable for implementation in real-time detection.

The features disclosed in the description, or in the following claims, or in the accompanying drawings, expressed in their specific forms or in terms of a means for performing the disclosed function; or a method or process for obtaining the disclosed results, as appropriate; may, separately, or in any combination of such features, be utilised for realising the invention in diverse forms thereof.

While the invention has been described in conjunction with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when given this disclosure. Accordingly, the exemplary embodiments of the invention set forth above are considered to be illustrative and not limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the invention.

For the avoidance of any doubt, any theoretical explanations provided herein are provided for the purposes of improving the understanding of a reader. The inventors do not wish to be bound by any of these theoretical explanations.

Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

Throughout this specification, including the claims which follow, unless the context require otherwise, the word "comprise' and "include", and variations such as "comprises", "comprising", and "including" will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps It must be noted that, as used in the specification and the appended claims; the singular forms "a." "an," and "the" include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" one particular value, and/or to "about" another particular value. When such a range is expressed; another embodiment includes from the one particular value and/or to the other particular value. Similarly, when values are expressed as approximations; by the use of the antecedent "about," it will be understood that the particular value forms another embodiment. The term "about" in relation to a numerical value is optional and means for example +/-10%.

Claims (18)

1. CLAIMS1. A stand-off screening system for determining an indication that a person is carrying concealed knife, the system comprising: an array of radar transmitters and receivers configured to transmit radar waves and receive reflected radar waves; and a processor, connected to the array of radar transmitters and receivers, and configured to utilise one or more properties of a reflected signal derived from one or more received reflected radar waves to determine the indication as to whether the person has a concealed knife.
2. 2. The stand-off screening system of claim 1, wherein the properties of the reflected signal include one or more of: a shape of the reflected signal; a width of a diffraction angle of the reflected signal; and an amplitude of the reflected signal.
3. 3. The stand-off screening system of claim 1 or claim z, comprising a plurality of arrays of radar transmitters and receivers, the individual radar transmitters and receivers within an array extending in a first direction and the arrays extending in a second direction transversal to the first direction.
4. 4. The stand-off screening system of claim 3, wherein each plurality of arrays provides a bank of radar transmitters and receivers, and the system comprises a plurality of banks of radar transmitters and receivers, the banks separated in the first direction.
5. 5. The stand-off screening system of claim 4, wherein each bank is angled relative to an adjacent one or more banks.
6. 6. The stand-off screening system of any preceding claim, wherein either: each radar transmitter and receiver includes a respective microcontroller, OF the array of radar transmitters and receivers is divided into 3 plurality of subsets of radar transmitters and receivers, each subset connected to a respective microcontroller; and wherein each microcontroller is configured to pre-process radar data before sending it to the processor.
7. 7. The stand-off screening system of any preceding claim, wherein the radar transmitters are configured to transmit radar waves at a frequency of at least 3 GHz and n more than 300 GHz.
8. 8. The stand-off screening system of any preceding claim, further comprising a central oscillator, and wherein each radar transmitter and/or receiver is configured to synchronize to the central oscillator.
9. 9. The stand-off screening system of claim 8, wherein a frequency of the central oscillator is lower than a frequency at which the radar waves are transmitted by the radar transmitters.
10. 10. The stand-off screening system of any preceding claim, wherein the radar transmitters are configured to utilise multiple transmission channels; the multiple transmission channels being separated in frequency and/or time.
11. 11. The stand-off screening system of claim 10, wherein sets of radar transmitters in respectively different frequency-separated transmission channels are configured to transmit simultaneously.
12. 12. The stand-off screening system of any preceding claim, further comprising a camera configured to determine the position of the person relative to the array of radar transmitters and receivers.
13. 13. The stand-off screening system of claim 11; wherein the processor is configured to derive expected properties of a reflected signal derived from one or more received reflected radar waves utilising the determined position, and to utilise this in the determination of the indication as to whether the person has a concealed knife.
14. 14. The stand-off screening system of any preceding claim, further comprising baffles and/or radar absorbing film configured to reduce cross talk between the radar transmitters and receivers.
15. 15. The stand-off screening system of any preceding claim; wherein the processor is configured to utilise a machine classifier to determine the indication as to whether the person has a concealed knife.
16. 16. The stand-off screening system of claim 15, 'wherein machine classifier is one or a combination of a random forest classifier, a convolutional neural network, and an inverse ray-tracing algorithm.
17. 17. The stand-off screening system of any preceding claim, wherein the processor is configured to determine the indication as to whether the person has a concealed knife utilising one or more of a fraction of a plurality of frames of received radar waves where an amplitude is above a predefined threshold; an average peak width; a variance in the peak width; and data indicative of a smoothness of how a peak of the received radar wave varies in time.
18. 18. A method of determining an indication that a person is carrying a concealed knife; using the stand-off screening system of any preceding claim, the method comprising: (a) illuminatino the individual with radar waves transmitted from one or more of the of radar transmitters; (b) receiving reflected radar waves at one or more of the radar receivers; (c) deriving one or more properties of a reflected signal from one or more of the received radar waves; and (d) determining, by the processor, the indication as to whether the person has a concealed knife based on the derived one or more properties of the diffraction signal.