Viewing and Tracking of Target Objects
The present invention relates to methods and apparatuses for the illumination and viewing of objects that may be hidden from an observer.
The methods and apparatuses have particular applicability to optical sensing and imaging, especially where direct line of sight of a target object or objects of interest to an observer is not possible.
Optical imaging and image analysis are very important within any number of situations including but not limited to photography, astronomy, security, surveying, military operations, aerial and non-invasive inspection, livestock and crop management etc.
When the human eye looks at an object, it requires a direct "line of sight" of the object within its field of view so that any light incident on the object, and then reflected from it, is captured by the eye and processed by the brain as a recognisable image of the object. If the same object is hidden from the observer field of view, for example behind a wall, light incident on it will not reach the eye and the object is not visible to the observer.
To aid human observers, there exists imaging devices such as the periscope, where mirrors are used to manipulate the path of reflected light from an object to the eye of the observer. These can be used to observe an object behind a wall. Periscopes are quite common at sporting events or concerts to help people see over the crowd, and also find use when an observer wishes to remain hidden from sight, for example in submarines or warfare where a soldier may wish to remain hidden behind a fortification giving protection while watching out for an enemy or target.
The human eye has evolved to operate at its peak efficiency during daylight hours and is most sensitive to wavelengths that lie within the visible part of the
electromagnetic spectrum, typically 400 to 700 nanometres (nm). Outwith this spectral range, the human eye is less sensitive (e.g., to longer wavelengths) or can even be physically damaged (e.g., by shorter wavelengths) and a number of solid state devices and optical systems have been developed to allow imaging at these longer or shorter wavelengths within the electromagnetic spectrum.
Depending on the wavelength, these devices can include Charge Coupled Devices (CCD), photo-multiplier tubes, infra-red detectors, terahertz detectors, microwave or radio antenna, gamma-ray or x-ray detectors or photographic film sensitised to the wavelength of interest etc.
Although these detectors offer a broad range of wavelengths of detection, the mode of operation is generally limited to "line of sight" operation; exception being where the energy or wavelength is such that it can travel through solid objects to the detector.
It would therefore be desirable to obtain a wavelength independent method for viewing or tracking objects that are not within the direct line of sight of an observer.
SUMMARY OF INVENTION
In a first aspect of the invention there is provided an apparatus for obtaining positional information relating to a target object, the apparatus comprising:
an illumination source operable to illuminate a scattering surface, said scattering surface being within the line of sight of the target object, such that scattered radiation is scattered by said scattering surface;
a detection device operable to detect reflected radiation, said reflected radiation being said scattered radiation which has reflected off said target object to within the field of view of said detection device; and
a processor operable to calculate said positional information from the detected reflected radiation.
In a second aspect of the invention, there is provided a method for obtaining positional information relating to a target object, the method comprising:
illuminating a scattering surface, said scattering surface being within the line of sight of the target object, such that scattered radiation is scattered by said scattering surface;
detecting reflected radiation within an imaged area, said reflected radiation being said scattered radiation which has reflected off said target object to said imaged area; and
calculating said positional information from the detected reflected radiation. Other optional features are as described in the dependent claims.
BRIEF DESCRIPTION OF THE DRAWINGS
Embodiments of the invention will now be described, by way of example only, by reference to the accompanying drawings, in which:
Figure 1 is a side view diagram showing schematically an arrangement for imaging and tracking a hidden from view object according to a first embodiment of the invention; Figure 2 is a top view diagram of the arrangement of Figure 1 showing (a) the radiation from the illumination source scattering from the floor and (b) the scattered radiation reflecting from the target object;
Figure 3 is a top view diagram of the arrangement of Figures 1 and 2 illustrating the ensemble of possible locations for the object; and
Figure 4 is a top view diagram of the arrangement of Figures 1 and 2 illustrating the step of combining the probability distributions to estimate target object location.
DETAILED DESCRIPTION OF THE EMBODIMENTS
The ability to detect motion and track a still or moving object hidden around a corner or behind a wall has many uses and may, for example, provide a crucial advantage when physically going around the obstacle is impossible or dangerous. Previous methods have demonstrated that is possible to reconstruct the shape of an object hidden from view. However, these methods do not enable the tracking of movement in real-time. A compact non-line-of-sight laser ranging technology is disclosed which relies upon the ability to send light around an obstacle using a scattering surface (e.g., floor or wall) and to detect the return signal from a target object with only a few seconds acquisition time. By detecting this signal with a single-photon avalanche diode (SPAD) camera, it is possible to follow the movement of an object located a distance away from the camera with centimetre precision. The disclosed imaging technique in combination with analysis of the data collected can reveal both the presence of a target object (or objects) in a scene and indicate size, shape and direction of movement of that object (or objects), where the object(s) is/are outside the normal field of view of the observer. This is beneficial over current imaging systems such as those using Light Detection and Ranging (or LIDAR) approaches where there is a need for direct "line of sight" of an object within the field of view. The object being tracked may be entirely hidden from the field of view of the observer. The imaging system is such that it does not require a direct line of sight between the object of interest and the detector of the imaging system. The target object(s) of interest may be stationary or may be moving in space and time. Where the object(s) of interest is non-stationary, there is disclosed a method of tracking the movement of the object (or objects). The object(s) may be tracked in real time. Within a scene containing multiple objects of interest to an observer a portion of the objects of interest may be stationary and a further portion may be non-stationary.
The imaging system may comprise an illumination source, a detection device, a means of processing the data from the detection device and an output device. The wavelength of operation of the illumination source may be such that it is within the electromagnetic spectrum of radiation and can be detected by the detection device. The illumination source of the invention provides illumination to an area within a scene; the scene containing an object, or objects, of interest to the observer.
The illumination source may provide a continuous or discontinuous mode of illumination. In discontinuous mode, the illumination source may be operated at a certain, known frequency. The frequency of operation or number of cycles per second (Hertz, Hz) of the source may be optimised to provide a balance between the need to provide illumination and the ability to detect the illumination at a detection device. The frequency of operation of the illumination source can be slow (few Hz) or can be fast (many thousands of Hz or greater). The frequency of operation of the illumination source can be between 0 and 1000Hz (lKhz), between lKhz and 1 million Hz (lMHz), between lMhz and lOOOMhz (lGhz) or in some cases beyond 1GHz.
Any illumination source could be used provided that a detector can be matched to it. One example of an illumination source of the invention is a laser source. Laser sources cover a broad and useful part of the electromagnetic spectrum of radiation (~200nm to well beyond 1 micron in wavelength) and a number of commercial detectors are available to match the illumination wavelength of the laser. Laser sources provide a high power of illumination and depending on their design parameters can operate in a continuous mode of operation or a discontinuous mode of operation making them suitable for applications disclosed herein.
Depending on their design, Laser sources can provide different laser pulse widths. These pulses are typically very short in duration and can be in the millisecond, nanosecond, picosecond or femtosecond range.
The detection device may have a wavelength of detection being within the output spectrum of the illumination source and ideally having a maximum sensitivity at the same output region of the electromagnetic spectrum as the illumination source. The detector of the imaging system may be chosen such that it is optimised to the operation of the illumination source. The detection device is capable of capturing the illumination data under conditions of either continuous illumination or discontinuous illumination.
The detector may collect illumination data over a short period of time or over a long period of time. Illumination data collected by the detector may be transferred from the detection device to an image processing system. The illumination data collected by the detector is processed in such a manner that the relative movement of the object(s) hidden from view can be determined. When data from the detector is processed with respect to time, additional information can be gathered on the speed and, or, the direction of travel of the object(s).
The illumination data from the detector can be processed instantaneously, in "real time", or it can be recorded and stored for processing at a future time. The output device may be a presentation means or any other media to visually represent the data collected by the imaging system to the observer. The illumination source and the detector are located at a distance, X, from the object of interest to the observer. The distance, X , can be small or can be large. The illumination source and the detector may be located at an angle, Y, to the object of interest to the observer. The angle can be small or can be large depending on the size of the object of interest and the distance, X.
Figure 1 shows schematically a system for imaging and tracking a hidden from view object (target object), from a side view. Figure 2 shows the same arrangement in top down view (a) illustrating the radiation from the illumination source scattering from the floor and (b) illustrating the scattered radiation reflecting from the target object.
In Figure 1, a person behind a wall 100 represents the target object 110, but it could be anything and could encompass very different scenarios, e.g. a car behind a corner, a stranded person in a room that is on fire, an object in an underwater wreckage with limited access from outside etc.
The system comprises an illumination source 120 and detection device 130. Control of these may be performed using processor 145. The illumination source may comprise, for example, a high repetition rate, pulsed laser. Specifically, The laser may comprise an 800 nm wavelength femtosecond oscillator which emits pulses of 10 nj energy and 10 fs duration at a 67 MHz repetition rate (0.67 W average power).
The system may be arranged to generate a synchronisation signal to synchronise the acquisition to the propagation of the radiation pulses from the illumination source 120. In an embodiment, a small portion of the illumination source 120 output may be sent to an optical constant fraction discriminator (OCF) which then generates the synchronization signal (e.g., TTL signal) which is sent to the detection device 130 and/or the processor.145 for synchronisation. The laser pulses 135 are directed towards a scattering surface 140 that lies beyond the obstacle that obscures or limits the direct line of sight to the object (e.g. the wall 100 in the Figure 1). This scattering surface 140 can be any surface within line of sight of the target object 110, and may be another wall, a ceiling or roof„ an open door, the surface of another object or, as shown in this example, the floor. When a scattered laser pulse 150 hits the scattering surface 140 (e.g., floor), it will scatter into a spherical wave 160 (Figure 2(a)).
This spherical wave 160 will then propagate outwards in all directions, including behind the wall 100 and therefore reach the target object 110 to be detected and tracked. The radiation in the spherical wave will then in turn be reflected from the target object 110, with some of the reflected radiation 170 (Figure 2(b)) being
reflected towards the imaged area 180, which is within the direct line of sight of the detection device 130.
In Figures 1 and 2, the detection device is shown to be imaging an imaged area 180 on the floor, just beyond the edge of the obscuring wall. It should be understood that each scattering event results in the emission of a spherical wave. In the case of a complicated object, there will probably be many spherical waves of reflected radiation 170 originating from different parts of the object. These will enable determination of the actual shape of the target object 110. In other cases, e.g. a car imaged from a distance or with a lower resolution, it may be that only a single spherical wave emitted from the object is imaged. This spherical wave will appear as a section of circle on the imaged area 180 as the spherical wave of reflected radiation 170 intersects the imaged area 180. In Figure 2(b), an example of real data 190 is shown, illustrating the image of the spherical wave of reflected radiation 170 within the imaged area 180 as captured by the detection device 130. This data shows the spherical wave of reflected radiation 170 at a precise time instant: the picosecond temporal resolution of exemplary detection devices allows the capture of the propagation of the spherical wave of reflected radiation 170 as it traverses the imaged area 180. The total data cube acquired by the detection device 130 will provide a video in which (part of) the spherical wave of reflected radiation 170 is seen to propagate from left to right in the image. It is the combination of the temporal information, i.e. how the spherical wave of reflected radiation 170 moves over time, together with the spatial information, i.e. the actual shape of the spherical wave of reflected radiation 170 that allows the exact location of the target object 110 to be determined. This determination may be performed by processor 145. The location of the target object 110 is retrieved by utilising the fact that: (i) the time it takes for the radiation to propagate from the illumination source 120 to the target object 110 and back, similarly to a LIDAR system, gives information about the target object's distance and (if) the curvature
and direction with which the spherical wavefront of reflected radiation 170 propagates across the imaged area 180 provides information on the target object's position. The detection device may comprise single photon counting technology using a single-photon avalanche diode (SPAD) camera. A high temporal resolution of the detection device can be obtained by operating it in a Time-Correlated-Single- Photon-Counting (TCSPC) mode. The arrival times of single photons may be measured with a resolution in terms of picoseconds, e.g., with a time bin between 10-lOOps, or between 30 and 80ps. As such, the system may use (millions) of laser pulses in order to properly reconstruct the fully animated video. The acquisition time depends on the repetition rate of the laser: the higher the repetition rate, the faster the acquisition time. SPAD detectors, originally developed as single pixel elements, are gradually becoming widely available as focal plane arrays. The single photon sensitivity and picosecond temporal resolution make them good candidates for real-time non-line-of-sight ranging of a moving target. Specifically, in this example, the detection device 130 may comprise a 32x32-pixel array of Si CMOS SPADs. A SPAD is based on a p-n junction device biased beyond its breakdown region. The high reverse bias voltage generates a sufficient magnitude of electric field such that a single charge carrier introduced into the depletion layer of the device can cause a self-sustaining avalanche via impact ionisation. The avalanche is quenched, either actively or passively to allow the device to be "reset" to detect further photons. The initiating charge carrier can be photo-electrically generated by means of a single incident photon striking the high field region. It is this feature which gives rise to the name 'Single Photon Avalanche Diode'. This single photon detection mode of operation is often referred to as 'Geiger Mode'. The high sensitivity of the detection device 130 allows extremely short acquisition times, which in turn allows one to locate target objects on timescales sufficiently short to be able to track their movement. Locating the position of an object hidden
behind a wall with centimetre precision is possible, without the need for pre- acquiring a background in the absence of the object. It can also be shown that realtime acquisition is possible for an object moving at a few centimetres per second. The detection device 130 may comprise an array of SPADs individually operable in a time-correlated single-photon counting (TCSPC) mode: every time a photon is detected by a pixel, the time difference between its arrival and the arrival of the synchronisation signal (e.g., TTL trigger from the OCF) is measured and stored in a time histogram. Each histogram comprises a number of time pixels and a time bin of certain duration. Specifically, histogram may comprise 1024 time pixels with a time- bin of 45.5 ps. The time resolution is limited by the electronic jitter of the system, which may be approximately 110 ps (measured at full-width-half-maximum), for example. This impulse response corresponds to a spatial (depth) resolution of a 1.65 cm, allowing the approximation of the back scattering as a single spherical wave originating from the target.
The target object-position retrieval algorithm may rely on both the temporal and spatial information recorded by the detection device (SPAD camera). Every pixel i of the (e.g., 32x32-pixel) camera, corresponding to a position r- = (χι, Υι) in the imaged area, records a histogram of photon arrival times. These histograms give a probability distribution of arrival times for photons scattered by a hidden target. This time distribution can be mapped into a probability density in space for the target position. A spatial probability density is calculated for each pixel, and the product of all probabilities constitutes the joint probability density of finding the target at a specific position in space.
First, the signal of interest coming from the target object alone is isolated from the signal coming from unwanted sources in the environment such as the walls and the ceiling. This can be achieved by simply acquiring a background signal in the absence of the target object; however, this may not be a practical solution if we are interested in tracking non-cooperative moving target objects. Instead, by acquiring data with the target object at different positions, it is possible to distinguish the
signal that is not changing at each acquisition (generated by the static sources) and the signal that is changing (generated by the target object). An average of the temporal histograms for each pixel proves to be a very a good approximation of the background signal and allows to effectively isolate the signal generated from the target object alone. Background subtraction is discussed in more detail later.
Once the target object signal is isolated, the processing proceeds to time-of-flight measurements and fitting of a Gaussian function to the temporal histograms. For each pixel i, the peak position of the Gaussian fit (t)j is a measure of the average total photon flight time, with an uncertainty that is taken to be the Gaussian standard deviation ot.. The arrival time tt is a measure of the light travel-time from the moment the laser hits the ground, scatters to a target object at a point = (½, y0) and scatters back to the specific point rt in the field of view (imaged area) of the camera. This is illustrated in Figure 3. There is an ensemble of locations that satisfy this condition, thus forming a three-dimensional ellipsoid which collapses to a two-dimensional ellipse 300 on a plane parallel to the floor (scattering surface), defined by the target object's height. This ellipse 300 is defined by:
\r^- \ + |r¾ - | = tt X c
where |r¾— f \ and |r¾— rt\ are the distances from the laser point Ϋι (labelled 150) on the floor to the target object and from the target object to the point rt (labelled 310) respectively, c is the speed of light (e.g., relevant to the medium). Solving this equation for the object position r¾ gives infinitely many solutions lying on the surface of an ellipsoid defined by foci Ϋι and r-: the possible positions in space that can generate a signal at pixel i at time tt lie on an ellipsoidal surface with evenly distributed probability. By way of example, one possible photon path 320, not corresponding to the actual location of the object 110, is shown. In the absence of any uncertainties in the measured signals, the resulting pixel probability density of the object's location ^l pse ( ^) calculated from the data collected by pixel i is therefore given by:
peiupse ^ χ fl if |r¾ - I + |r¾ - rt\ = tt X c 1 0 I 0 otherwise
The function above can be represented in a simpler form by using ellipsoidal coordinates with foci Ϋι and r-:
Pt eI"pM( S) oc ff(e - ctt)
where ε = |r¾— f \ + |r¾— r-| In any real implementation, the histogram hi(t) recorded by any given pixel i will contain an uncertainty on the arrival time of the signal. As a result, the probability density P^l pse ( ^) will no longer be a uniform ellipsoid, and the uncertainty in the time histogram can be mapped onto the spatial probability density as:
/ iipse (r¾) oc S e - ct)fi t)dt = fi (-)
J -co
The signal recorded in histogram hi(t) has a Gaussian form with a standard deviation of ot... The uncertainty is originating from different sources, for example the jitter on the system and the finite size of the target. As a result of the Gaussian form of the recorded signal, the general expression of the pixel probability density
P ^iellipse r (r0) becomes:
This ellipse 300 represents a probability distribution for the position of the target object with uncertainty ot.. The uncertainty ot.. is represented here by the line thickness of ellipse 300. Here {t)t is the mean arrival time registered by the pixel i, and fft..its standard deviation.
In the described arrangement, it is aimed to locate the object in x and y, on the same plane as the scattering surface (i.e., the plane of the floor in this example), making the assumption that such an object is not moving considerably in the vertical direction. A two dimensional search space at a given height close to ground level is therefore defined, and r¾ takes the form r¾= (xo; yo). In a real-life implementation, the height can be appropriately estimated based on the type of target being tracked or located. An error Δζ in estimating this height will result in the worst case in an error Ar0 of the same order in the determination of the target's r¾ coordinates,
although it will be typically much smaller and decreases rapidly for objects that are further away: Ar0 « (z0/ |r¾ |)Az.
As mentioned above, the calculated pixel probability densities p.ellipse 0f each pixel are multiplied to obtain the joint probability density (r¾). However, there is a risk that a given pixel i will lead to an unsignificant fit of the signal and that the values of (t)j and ot.. will be unreliable. To avoid that these unreliable p.ellipse ' (ψ^) affect the joint probability density by multiplying relevant densities by zero, the probability Pj (r¾) associated with the pixel i can be taken to be a linear combination of the pixel probability density p.ellipse ' (ψ^) and a uniform probability density punif°rm that will prevent any point in space to be multiplied by zero:
P ) =
+ (1 - at)PuiaTorn
Here ajis a coefficient between 0 and 1, related to the reliability of the probability density Pj (r¾). The choice of at for each pixel depends on how well the probability density Pi ( o overlaps with the space in which the object is being sought. More precisely at may be set as Aj/A where A, is the area contained in the search space where the probability density Pj (r¾) is over a certain threshold (half its maximum) and A is the area of the search space.
Pixel probability distributions p.ellipse ' (ψ^) are calculated for every pixel i in the imaged area. In order to retrieve the target object's position, the joint probability density is calculated by multiplying the probability distributions from all x (e.g., x=1024) camera pixels:
X
P(r¾) = N ]^[ pi (r¾)
i=i
P(r¾) determines the overall probability distribution of the location of the target, and N is a normalisation constant.
Figure 4 illustrates this step graphically. Ellipses 300 calculated from different pixels 310 give slightly displaced probability distribution that intercepts at a given point. Here four ellipses 300, are shown, each corresponding to one of four pixels
310 highlighted. The area where the ellipses overlap indicates the region of highest probability for the target location. In a real example, there will be many more pixels and therefore the same number of ellipses (of the order of magnitude of 1000, for example). Multiplying these probability distributions provides an estimate of the location of target ob j ect 110.
As already mentioned, in a case where an object-free background is impractical or impossible to acquire, the average of histograms recorded at different times can be used to estimate the background. In an embodiment, the average may comprise a median of the histograms recorded at different times. The moving target results in signal differences between the recorded histograms. It can be shown that the average of the histograms is very similar to the background signal. In a real-life implementation, the object-present background can be calculated with the first few acquisitions. If a target object is present, but is not moving, , this algorithm will fail to detect the target; but as soon as the target starts moving, it will take around 15 seconds (about 5 acquisitions) to record a background and begin to accurately locate the target. The best approximation of a background will come from a target that is moving in both the x and y directions. However, it should be appreciated that during this initialisation period of 15 seconds, the camera is acquiring data which does provide information regarding the movement of the target: this data may be less accurate with respect to the data acquired at later times, but nonetheless indicates target movement. For stationary targets, other methods of adjusting for background noise can be performed, as known in the art. To retrieve the position of the target object, the position rt of pixel i on the floor is also required. The camera is looking down at the floor, but still records a (e.g., 32x32 pixel) square image. This image actually corresponds to a field of view that is trapezoidal and stretched both spatially and temporally, with respect to a squared field of view perpendicular to the line of sight of the camera. To correctly retrieve the positions of a target object, the actual position r- = (x^ Yi) of each pixel of the camera is determined. To do so, the dimensions of the imaged area and its distance to the camera are measured, to reconstruct the actual shape of the imaged area on
the scattering surface (e.g., floor). In a real-life application, knowing the height and angle of the camera with respect to the scattering surface will be sufficient to determine the geometry of its field of view and therefore know the positions (x,; y,) of each pixel in the field of view. Temporal distortion of the recorded data may also be corrected for: because the imaged area is not perpendicular to the camera's line of sight, photons recorded at the top of the field of view take longer to reach the camera than the ones coming from the bottom. Again, knowing the geometry of our imaging system, the measured (t)j can be corrected accordingly. The values of , (t)j and ot can then be used to retrieve the position of a target object, as explained.
When trying to track target objects that are far away from the detector, the error in the position determination tends to increase due to the fact that the curvature of the scattered waves decreases (spherical waves at a large distance from the source look like plane waves) meaning that their areas of overlap become less well defined. This problem can be offset by repeating the measurement with the illumination source pointing to a slightly different position and/or using multiple detection devices, each looking at slightly different positions (different imaged areas). In general, this generalisation to multiple illumination points and/or detection devices will increase the tracking resolution of the system.
The methods described herein can be extended to tracking multiple target objects, provided that the signal originating from the objects do not significantly overlap. As a proof of principle, the signal from two targets separated by 45 cm were recorded. Once the background is retrieved from the recorded signal, it is possible to distinguish the two signals coming from the two distinct targets, as they produce backscattered spherical waves that can be distinguished both in time and in space. For this proof-of-principle experiment, the multiple peaks in each histogram were located and the retrieval algorithm individually applied to each of the two separate signals. The retrieved probability densities are in good agreement with the positions retrieved from single-target measurements. More precise tracking of multiple targets may be enhanced by some relatively straightforward solutions such as increasing the field of view of the system by using large-area arrayed detectors or
decreasing the temporal response of the system. Large-format SPAD array cameras with these properties are in development.
The motion tracking will be more precise if the object moves by less than its physical dimension during the sub-second acquisition time. For a person, this amounts to maximum speeds of the order of a few meters/second, i.e. to a person walking at a relatively fast pace (4 km/h). For a car, it would be of the order of 20- 30 km/h. If the object is moving faster than this, then we would still be able to track its motion correctly but the object would appear to be larger than in reality due to blurring effects. This can be offset by adopting detectors which have faster acquisition times and/or achieve all of the data processing described above directly on-board. In this case, the time limitations mentioned here (that are due mainly to data download times) can be reduced by factors lOx or even lOOx and thus allowing tracking of very fast moving objects.
Nevertheless, for many applications, localisation of an object or simply information regarding if it is moving or not is sufficient. In these cases, blurring effects are of no consequence. Moreover, we should underline that this is just the first step to tracking target objects using lasers and cameras. New faster lasers and cameras are already being developed that will allow us to increase even further the specifications of this first demonstration.
It should be appreciated that the above description is for illustration only and other embodiments and variations may be envisaged without departing from the spirit and scope of the invention.