GB2463703A

GB2463703A - Estimating the direction in which a camera is pointing as a photograph is taken

Info

Publication number: GB2463703A
Application number: GB0817391A
Authority: GB
Inventors: Chris Marshall
Original assignee: Geotate BV
Current assignee: Geotate BV
Priority date: 2008-09-22
Filing date: 2008-09-22
Publication date: 2010-03-24
Also published as: GB0817391D0; TW201126189A; WO2010032057A1

Abstract

An apparatus is provided for estimating the azimuthal orientation of a camera 100, the camera comprising: an antenna 210 having a predetermined directional radiation pattern 450 which is anisotropic in the azimuthal plane when the camera is held in a normal orientation for photography; a receiver receiving signals from each of a plurality of transmitters; and a processor measuring the strength of signals received at the antenna from each of the plurality of transmitters, obtaining the position of each of a plurality of transmitters, obtaining the position of the camera, predicting the expected strength of each of the received signals, and estimating the azimuthal orientation of the camera based on the obtained positions of the camera and the plurality of transmitters, the measured signal strengths, the antenna radiation pattern, and the predicted signal strengths. The transmitters may be satellites 401,402. The signals for estimating the position of the camera may be received at a further antenna having a radiation pattern which is isotropic in the azimuthal plane when the camera is held in a normal orientation for photography.

Description

DESCRIPTION

DIRECTION ESTIMATION

This invention relates to estimating the direction in which a camera is pointing as a photograph is taken. It is particularly relevant for cameras which include, or are connected to, a satellite positioning receiver.

"Geo-tagging" refers to recording the location of a place or event of interest. Geo-tagging is an increasingly popular way to annotate and organize io recorded information like images and videos. By adding location metadata to the media, users can search and browse an archive in interesting and intuitive ways -for example by plotting the set of locations on a map.

To avoid the labour-intensive task of manually typing the location * ** names or geographic coordinates necessary to create such annotations, satellite positioning is increasingly being used to provide precise location metadata.

Satellite positioning systems, such as the Global Positioning System (GPS), enable a receiver to calculate its position based on the measured time of arrival of signals from a set of orbiting satellites (known as space vehicles -S. ** : * ** 20 SVs). *. . * S * * *.

According to an aspect of the current invention, there is provided a method of estimating the azimuthal orientation of a camera having an attached antenna, the antenna having a predetermined directional radiation pattern which is anisotropic in the azimuthai plane when the camera is held in a normal orientation for photography, the method comprising: measuring the strength of signals received at the antenna from each of a plurality of transmitters; obtaining the position of each of the plurality of transmitters; obtaining the position of the camera; predicting the expected strength of each of the received signals; and estimating the azimuthal orientation of the camera based on the obtained positions of the camera and the plurality of transmitters, the measured signal strengths, the antenna radiation pattern, and the predicted signal strengths.

The method enables the pointing-direction of the camera to be estimated using a single, directional antenna. The antenna should be attached to the camera such that it has directional characteristics in the horizontal (azimuthal) direction when the camera is held normally to take a photograph.

Such normal orientation would usually include one or both of portrait or landscape orientations. The method uses measurements of signal strength from transmitters or beacons in known locations to deduce the orientation (also known as attitude) of the directional antenna. The need to provide an electronic compass or other hardware is avoided; therefore, for a camera already provided with a positioning system, the method can allow azimuthal direction estimation at little extra cost. The antenna used can be a simple * ** directional type and there is no requirement for multiple elements or a phased-is array. Since direction estimation is based on signal strength, there is no need, for example, to measure phase or to explicitly determine angle-of-arrival in the :. usual sense. For example, there is no need to rotate the antenna to determine S. * an angle-of-arrival or direction of strongest reception, as is usual is some other radio direction finding methods. By using the position of the camera to *. S. : * ** 20 determine the azimuthal orientation, the method makes use of information **, already produced by conventional geo-tagging solutions. The additional computational complexity is therefore minimal. The azimuthal-or compass-direction of the camera can be used to differentiate between photographs taken from the same vantage point, but facing in different directions.

Accordingly, it can help to identify the subjects of the different photos. This represents a significant enhancement to simply measuring the location of the camera when the photo was taken.

The transmitters may be satellites, in which case the received signals are satellite signals.

Satellite positioning is commonly used in conventional geo-tagging applications. It is generally reliable, accurate and provides global coverage.

The position of the satellites is a necessary input for the position-estimation process. Therefore the only additional measurements needed to carry out the current method in a satellite positioning receiver are measurements of the received signal strengths corresponding to each satellite.

The step of obtaining the position of each of the plurality of satellites preferably comprises at least one of: decoding a data message of a received satellite signal; and obtaining satellite-position or satellite-trajectory information via a communications network.

In a real-time satellite navigation application, satellite positions can be obtained by decoding ephemeris and/or almanac information contained in the satellite transmissions. In this scenario, the satellite positions used by the current method are a by-product of the processing already carried out at the conventional satellite positioning receiver. In other scenarios, it will be preferable to obtain the satellite ephemeris and/or almanac data from an * .* s.,.. independent source, such as a data server. In this case, the obtained positions 55.

is may be more reliable, since a centralised database can maintain accurate and complete records of satellite health and correct for any errors or deviations.

The step of obtaining the position of the camera may comprise **.

* estimating the position of the camera based on signals received from each of :.**, the plurality of transmitters.

This is efficient, because the same set of signals used to obtain the camera position are also used to estimate the direction. This means that many of the hardware components can be shared between the tasks. The transmitters may be satellites, as discussed above, or may be terrestrial beacons, such as cellular telephony base-stations or Wireless Local Area Network (WLAN) Access Points (APs). In either case the receiver hardware used to receive the signals from the transmitters can be adapted to measure signal strength via the directional antenna.

Optionally, the signals for estimating the position of the camera are received at a further antenna, the further antenna having a predetermined radiation pattern which is isotropic in the azimuthal plane when the camera is held in a normal orientation for photography.

A directional antenna, with a radiation/gain-pattern varying in the horizontal, azimuthal plane, is necessary for the direction estimation process.

However, the receipt of signals for successfully estimating camera position usually demands maximising the overall received signal power, across all transmitters. In other words, for position estimation it is in general better not to have direction selectivity in the azimuthal plane. Providing a further antenna therefore prevents direction estimation from reducing the robustness of position estimation. The directional antenna is used for direction-estimation; the further antenna, which may be isotropic (omni-directional) or weakly io directional but pointing skyward, is used for position estimation. Note that the signals received from the transmitter and used in each process are preferably the same, to allow the same receiver hardware to be reused. Indeed, it is possible that the further antenna comprises the same antenna as that used for * *.

direction estimation, but with a changed configuration. *...

When a further, isotropic antenna is provided, the method may further comprise measuring the strength of the signals received at the further antenna, and estimating the azimuthal orientation of the camera based on the measured S..

0 signal strengths for the further antenna.

Signal strength measurements at the further antenna can be useful for * S direction estimation. Although they are received omni-directionally (that is, * 04 without discrimination as to azimuthal angle), they can, for example, provide useful reference signal levels for the direction estimation process, when compared with the signal strengths measured for the directional antenna. This can provide greater accuracy than is possible with theoretically predicted signal strengths alone.

The camera can have a plurality of attached antennae, each of the plurality having a different predetermined directional radiation pattern which is anisotropic in the azimuthal plane when the camera is held in a normal orientation for photography, wherein the method comprises measuring the strength of signals received at each of the plurality of antennae from each of the plurality of transmitters, and wherein the step of estimating the azimuthal orientation of the camera is based on the measured signal strengths for each of the plurality of antennae, the antenna radiation patterns, and the predicted signal strengths.

Although the method works for just one antenna, multiple antennae can be applied for improved robustness and/or accuracy. The direction estimation algorithm can be applied to each antenna independently, or measurements of signal strength can be combined across antennae and the algorithm applied on the aggregate data. In either case, it is necessary to known the relative mutual orientations of the multiple antennae. This is information is provided by the directional radiation patterns.

The method may further comprise: sensing an angle of elevation or an angle of roll of the camera and/or the antenna; and estimating the azimuthal orientation of the camera based on the sensed angle.

The sensing of the angle of elevation, or tilt, and the angle of roll (for * ** example, rotation about the optical axis of the camera) can be used to improve **** is the method. The information can be combined with the radiation/gain-pattern of the antennae to more precisely reconcile the predicted and measured signal * * * * strengths.

* The method may also comprise estimating the azimuthal orientation of the camera based on a predetermined polarization of the antenna. b

*. : 20 Knowledge of the polarization of the directional antenna can help to * *i discount reflected signals when comparing the actual received signal strength and expected signal strength. Thus, this variation can help to improve the robustness of the method. It is assumed that the original (transmitted) polarisation of the signals is also known.

The step of estimating the azimuthal orientation of the camera may be performed after an intentional non-zero delay after the signals are received.

It is possible to perform direction-estimation off-line -that is, at some time after the signals were actually received. In this case, the signals or the signal strength measurements can simply be recorded for later analysis. This minimises the amount of processing necessary at the time of capture of a photograph, for example. It is also particularly suitable for geo-tagging applications, where the position and/or direction information is typically not needed immediately. By delaying the processing, complexity of a portable device can be minimised, since the data can be analysed by a more powerful processor at a different time and place; battery power in the portable device can also be conserved.

According to a further aspect of the invention, there is provided apparatus adapted to estimate the azimuthal orientation of a camera, the apparatus comprising: an antenna attachable to the camera and having a predetermined directional radiation pattern which is anisotropic in the azimuthal plane when the camera is held in a normal orientation for photography; a receiver, electrically connected to the antenna and adapted to receive signals from each of a plurality of transmitters; and processing means, adapted to: measure the strength of the received signals; obtain the position of each of the plurality of transmitters; obtain the position of the camera; predict * ** the expected strength of each of the received signals; and to estimate the * i5 azimuthal orientation of the camera based on the obtained positions of the *... camera and the plurality of transmitters, the measured signal strengths, the * * * antenna radiation pattern, and the predicted signal strengths. *..

The invention will now be described by way of example, with reference *. : 20 to the accompanying drawings, in which: * S. Figure 1 is a flow-chart of a direction-estimation method according to an embodiment; Figure 2 is a block diagram of an apparatus according to an embodiment; Figure 3 illustrates the principle of the direction-estimation method and apparatus according to an embodiment using one directional antenna; Figure 4 illustrates the principle of the direction-estimation method and apparatus according to an embodiment using two directional antennae; Figure 5 is a diagram of a further example configuration for two directional antennae; and Figure 6 illustrates the principle of the direction-estimation method and apparatus according to an embodiment using one directional antenna and a further isotropic antenna; The inventors have recognised that, in the context of geo-tagging photographs, the location of the subject of the photo is more important than the location of the camera itself. For example, a photographer may take two photos while standing in the same location, but the content of the photos may be very different, because the camera was pointing at different subjects in to each case. Conventional geo-tagging will assign the same location to these photos. Conversely, a photographer may take many photos of a single subject of interest, for example, by walking around a statue to obtain different views.

* Conventional geo-tagging will assign a different location to each image.

In order to differentiate these different situations, it is necessary to Se..

is record not only the location of capture of an image, but also the direction in *:::: which the camera was pointing. This will enrich the geo-tagging experience of the photographer and allow more advanced searching and browsing functions *..

* for image-collections. The orientation or pointing direction of a camera at a given position has two degrees of freedom: the azimuthal angle, in the horizontal plane; and the elevation angle, in a vertical plane. The azimuthal angle is more important for the photography application, since this is the "compass direction", or orientation in the plane of the earth's surface.

Knowledge of the azimuthal angle is therefore essential to determine what the camera was pointing at, from a given known location. In any case, elevation angle can easily be determined by accelerometer sensors if desired.

The current invention allows the azimuthal angle of orientation to be determined, based on measurements of signals received by the camera. The signals used for direction estimation are preferably the same signals used for determining the position of the camera; therefore, a positioning receiver integrated in or attached to the camera can provide direction information as well as location. Furthermore, there is no need to provide additional hardware components, such as a separate electronic compass comprising magnetic field sensors, or an inertial navigation system (INS) comprising gyroscopic sensors.

This allows the cost of the camera, or camera accessory, to be minimised, while providing enhanced geo-tagging features. For example, the direction of taking each photograph could be displayed on a map, or the subject of the photograph could be automatically identified by reference to a Geographic Information System (GIS) database. The identification of the subject of the photograph can be made more accurate by also recording information about the distance from the camera to the subject. This data may be available either directly, for example from a range-finder in the camera (used for auto-focus), or indirectly, for example from lens focal-length information.

Fig. 1 is a flowchart of a method according to one embodiment of the invention. In this example, GPS satellite positioning signals are used both for obtaining position and the estimating azimuthal angle of orientation. The *:*::* method is suitable for determining the orientation of a camera provided with a directional GPS antenna. * * *S. *

In step 10, the strength of signals received from several satellites is **** : measured. The GPS constellation comprises at least 24 satellites, in different orbits, continually transmitting a data signal. A GPS receiver can produce a position fix anywhere on earth if it can successfully calculate its range from : 20 four satellites. The accuracy of the position fix improves with increasing *..: numbers of visible satellites. Measuring the strength of each of the satellite signals is easier than complete calculation of a range; therefore, it may be possible to measure signal strength for more satellites than are used for position estimation -for example, errors in decoding the satellite data message may prevent range calculation, but a signal-strength estimate may still be available. Signal strength can be measured by any appropriate means, for example using Power Signal to Noise Ratio (PSNR). Absolute signal strength values are not essential for the direction estimation process; it is the relative strength across the set of satellites that is important.

At step 20, the satellite positions are obtained. This is done as part of the conventional GPS position estimation procedure and so this information will be available before the position of the camera is known. Different GPS receivers may obtain the satellite positions in different ways. A common approach is to decode the data message in the satellite broadcast itself to obtain ephemeris and almanac data, describing the satellite trajectories.

Alternatively, in "Assisted" GPS (AGPS) this information is provided via a separate channel, such as a cellular telephony network. In another GPS implementation, known as "Capture and Process Later" (hereinafter "Capture and Process"), the ephemeris and almanac data are not necessarily provided to the receiver. Instead, short sequences of the satellite transmission are received and stored by the receiver; later, these can be uploaded to another lo device for processing to derive position estimates. The device performing the later processing can obtain the satellite position information from a central database of historical ephemeris data, for example, via the internet. The current direction estimation method is not limited to any particular means of obtaining satellite positions in step 20, and is therefore applicable to any GPS implementation. *...

When the satellite position data are to be obtained from the satellite : data, there are a number of alternatives, allowing different levels of accuracy and robustness. In the best case, each satellite data signal can be decoded completely, giving access to the ephemeris, or detailed orbital trajectory : * 20 information. However, some satellite signals may be too weak (noisy) for complete decoding (although signal strength may still be used in the direction estimation process, as noted above). Even if the ephemeris cannot be decoded from a given satellite data message, it may still be possible to decode the almanac. This provides approximate trajectory data and may be sufficiently accurate in practice to support the direction estimation. Indeed, the almanac information about all satellites is broadcast by all satellites, so that it is not essential to receive a given individual satellite signal strongly in order to be told, via the almanac messages of other satellites, where it is. At step 30, the position of the receiver (that is, the position of the camera) is obtained.

Again, this is done in the normal sequence of GPS processing, and so no additional processing effort is required. The relevant GPS position estimation methods are well known and widely available. Again, the current method has no limitation to any particular type of GPS implementation.

Although steps 10, 20, and 30 need not be performed in any particular order, in a real-time GPS implementation it may be beneficial to perform signal-strength measurement step 10 first, followed by positioning steps 20 and 30. This is because the signal strength measurement may be more robust to interference from the camera than the positioning functions. Digital cameras typically generate electro-magnetic (EM) interference in the processes of image-capture and storage to memory which can degrade the SNR of received satellite signals. Likewise, in a capture-and-process implementation,, it may be beneficial to perform signal strength measurements first (while the camera is "noisy"), before capturing and storing the signal samples necessary for later position estimation.

At step 40, the method predicts the signal-strength that can be expected :.:: 15 to be observed at the location of the receiver at the time that the photograph S...

was taken, given the location of each of the satellites. The signal power : originally transmitted is known and/or assumed to be equal for all satellites.

The attenuation of the signals will depend on the distance from each satellite to the receiver; and atmospheric effects will also cause the attenuation to vary : * 20 for satellites at different elevations in the sky, as observed by the receiver. The model used for prediction of expected signal power can be more or less sophisticated. The more accurate the model, the greater the likely accuracy of the eventual direction estimate. For example, a trivial model would assume that approximately equal power should be received from all satellites. A more complex model might take into account the elevation in the sky of each satellite. As an alternative to purely theoretical prediction, actual signal measurements can be taken into account, or used on their own. In any case, the goal is to provide a set of reference signal levels against which actual signal strength measurements, from the directional antenna, can be compared.

At step 50, the direction in which the receiver (camera) antenna is pointing is estimated. The signal strengths expected to be observed and the signal strengths actually observed are related by the receive-or gain-pattern (also known as the radiation pattern) of the directional antenna and the direction in which it was pointing when the signal strength measurements were taken. The radiation pattern can be obtained and expressed in a variety of ways. These include prior explicit quantitative measurements of gain at different beam angles; approximation by a default theoretical or numerical model (such as an ideal patch antenna); or by reference to the known design-characteristics of the particular antenna being used. Knowing the characteristics of the radiation pattern, the remaining unknown parameters are those related to camera orientation. The estimation of orientation is therefore achieved by comparing the predicted and measured signal-strengths, with reference to the radiation pattern, and solving for the orientation parameters.

Fig. 2 shows an example apparatus suitable for implementing the above method. This includes a GPS receiver 200 attached to a camera 100. The attachment of the receiver may be internal (integrated) or external. If the :.:: Is receiver is external (for example, if the receiver is an accessory, detachably *S..

attached to the camera) then the camera hot-shoe may preferab'y be used for : attachment. It is essential, in any case, that the orientation of the receiver-antenna 210, with respect to the camera, is known.

The GPS receiver 200 of this embodiment is a Capture-and-Process : * 20 receiver, It includes a GPS radio-frequency (RF) front-end 220 for down-converting the satellite signals and sampling the resulting intermediate frequency (IF) signals. It also has a microprocessor 230 which receives satellite signal samples from the front-end 220 and stores them in a memory 240. Later, when the receiver 200 is connected to a personal computer (PC) 300, the stored samples are uploaded for processing. The PC therefore carries out the usual GPS processing for position estimation, as well as the method of the current invention. As an alternative, in some Capture-and-Process implementations, the PC will, in turn, upload data to a server (not shown) for processing. In this case, the direction estimation may be carried out on the server instead.

As will be clear from the foregoing description, the method is not limited to OPS receivers with any particular hardware implementation or configuration (apart from the requirement mentioned above for fixation of the antenna to the camera in a consistent orientation).

Fig. 3 illustrates the principle of the invention in greater detail. This shows the camera 100, with its attached directional antenna 210. The radiation pattern of the antenna is directional (anisotropic) in the horizontal plane. That is, when the camera is held in its normal orientation for taking a photograph, the gain of the antenna varies with the azimuthal angle. The main lobe of the directional antenna 210 is crudely illustrated in Fig. 3 by the fan-shaped beam 450. The constraints on the antenna orientation can be fulfilled in several ways. Preferably, the antenna lobe points in the same direction as the camera.

This is beneficial for GPS signal reception, since the front of the camera will usually not be obscured when a photo is taken. Furthermore, if the "beam" 450 of the directional antenna is taken to be a cone shape, then the orientation constraint above is satisfied when the camera is in portrait or landscape orientation. That is, the antenna directionality will be independent of roll along * *** the optical axis of the camera. Of course, in practice, a perfect circularly symmetrical antenna gain pattern will not be achievable, and the radiation ** pattern will be altered by 90 degree rotation of the camera. This can be taken in to account by providing a sensor, such as an accelerometer, which can * * 20 detect the roll of the camera with respect to the horizontal plane. Such sensors are commonly provided in digital cameras to indicate the format of images captured. Other cameras may have a manual portrait-mode setting: this can also be used to infer the antenna radiation-pattern.

According to the geometry indicated by Fig. 3, antenna 210 will receive a relatively strong signal from satellite 401 and a weak signal from satellite 402. Thus, by comparing the measured and expected signal strengths, with reference to the radiation-pattern, the method can deduce that the camera is pointing more in the direction of satellite 401 Knowing the position (and therefore relative direction) of this satellite, the azimuthal orientation can be estimated as "North".

Fig. 3 provides a simplified plan-view of the geometry of the satellites with respect to the antenna. In practice, the elevation of the satellites 401, 402 in the sky will impact on the results. Satellites vertically overhead provide little or no directional information. Satellites near the horizon are most likely to be obscured by arbitrary land-based obstacles (trees, mountains, buildings). As a result, the best information for estimating direction comes from measurements for satellites at moderate elevations in the sky.

For photo geo-tagging applications, it will usually be sufficient to annotate images with points of the compass, rather than a numerical angle.

This means that the accuracy requirements for the current method are not particularly high. For example, direction can be estimated according 8 points of the compass (north, north-east, east, south-east...) with an expected error of +1-22.5 degrees.

In such applications, an antenna with an approximately hemispherical radiation (gain) pattern is advantageous. This means that the gain is substantial over one half of a three-dimensional sphere and significantly lower in the opposite hemisphere. This pattern provides a good trade-off between *.S.

the need to receive many signals (for example, for accurate position : estimation) and the need for directionality of the antenna (required for direction estimation). A patch antenna can provide a reasonable approximation to this hemispherical gain pattern. This type of antenna can be modelled as having : * * 20 gain which declines gradually with angular deviation from an axis normal to the *.s. plane of the patch. That is, the maximum gain is in the direction normal to the patch and rolls off to a lower value in the plane of the patch (90 degrees to the normal axis). For a given angle of deviation from the axis, the gain is approximately constant with rotation about the axis. That is, the gain pattern is circularly symmetric in the plane of the patch.

Further directional antennae can be added to the apparatus, as shown in Fig. 4. This increases the number of signal strength measurements available (one for each satellite, for each antenna) and can therefore increase the accuracy and/or the robustness of the direction estimate. In the example shown in Fig. 4, a second antenna 210a is shown, with a gain pattern or beam 450a oriented at 90 degrees to the beam 450 of the first antenna 210. The second antenna 210a will receive a strong signal from satellite 402 and a weak signal from satellite 401 These measurements are added to the set of constraints provided by the signal measurements at the first antenna 210.

They can be used to validate or refine the azimuth estimate.

There are many possible approaches to combining the information provided by multiple directional antennae. The following examples are presented in order of increasing complexity: A) use the measurements for one antenna, chosen from among the different antennae in different orientations. For example choose the likely best antenna based on camera orientation (landscape or portrait).

B) use the multiple results of A) (each an independent estimate) and merge the estimates, to generate a more accurate or robust result. Examples of this include averaging or taking a median value. The merging step may include a quality or confidence measure, such as a standard deviation of the estimates from their mean.

is C) use the signal strength measurements from each signal and each antenna *I..

jointly. For examp'e, an exhaustive search can be used to estimate direction, as described in greater detail below. In such a method, estimates of the expected signal strength for each signal and each antenna can be produced for a given hypothesis of the azimuthal orientation. These estimates can be compared and combined across multiple antennae in the same manner as for a single directional antenna.

Some particularly advantageous configurations of multiple antennae can be considered: 1) Antennae pointing at 90 degrees bearing relative to one another, as already described above -for example, one forward and one left. This arrangement gives good satellite coverage and diversity of directional coverage 2) Two antennae, at a mutual angle of 90 degrees as in case 1), but at oblique angles with respect to the camera optical axis, as shown, for example, in Fig. 5. One antenna 210c may point 45 degrees to the right (R) of the axis, and the other 210b may point 45 degrees to the left (L). The antennae are shown with notional radiation patterns 450c and 450b, respectively. This configuration retains the benefits of 1), but concentrates the antenna gain to the front of the camera, where the strongest signals are likely to be detected. The gain patterns of the two antennae may overlap to a significant extent in this scenario -this can help to ensure that satellites are not "lost" in a low-gain region between the lobes of multiple antennae.

s 3) Three antennae: two as in 2) and the third pointing vertically upwards (U) like a conventional GPS camera antenna. In landscape mode this vertical antenna gives a benchmark or reference signal level, and helps positioning (discussed in greater detail below). When in portrait mode, the antenna U becomes horizontal -pointing either left or right, depending on the camera roll-direction (clockwise or anti-clockwise). One of the left and right antennae, 450b 450c, points obliquely upward-forward, and the other points obliquely downward-forward. The combination of U and R, or U and L (typically choosing the antenna, L or R, pointing upward-forward) can then be used for direction finding.

When multiple antennae are included, their use can be selected, * *..

according to the orientation (landscape, portrait rotation) of the camera, which : is sensed in many conventional digital cameras. So, for example, in case 3), * either: L and R; L and U; or R and U can be chosen depending on the orientation of the camera. This selection can be made in real-time, by making : * * 20 measurements only from the selected antenna, or off-line in later processing, by storing measurements for all antennae, but only using those selected in accordance with the sensed orientation.

Multiple receivers (that is, multiple GPS RF front-ends) may be provided to support the multiple antennae. However, multiple antennae can also share a GPS RF front end. This can be achieved using a switching circuit, which allows a simpler and cheaper hardware implementation. Signal strength of each of the antennae can be measured in sequence (that is, consecutively, instead of concurrently). In theory, there is a risk that the camera may be moved or re-oriented while this sequence is being executed. Thus, the signal strength of all antennae should be sampled as quickly as possible, so that the variation caused in the measurements by camera movement should not be significant.

This will be especially true if the measurements are made instantly when the shutter-release button is depressed, since the photographer will be holding the camera still to take the photograph. By way of example, it may take between lOOms and is to make signal-strength measurements for a single antenna (or to record signal-samples sufficient to allow the signal strength measurement to be performed). Switching from one antenna to the next may take, for example, between Ims and is.

Note that highly-directional reception of GPS signals conflicts with the usual requirement to receive signals from as many satellites as possible, in order to estimate position. For this reason, it may be advantageous to provide io a further, omni-directional antenna in addition to the directional antenna. In this context, omni-directional means that the antenna's radiation-pattern is substantially isotropic, at least in the azimuthal plane. This criterion is fulfilled by the upward-facing antennae used in most existing in-camera GPS implementations. Fig. 6 shows an example of an apparatus with this modification. In this arrangement, the directional antenna 210 of Fig. 3 is augmented by a further antenna 215 with an isotropic radiation-pattern 455.

Antenna 215 will receive signals from both satellites 401, 402. The output of this antenna will therefore be more appropriate for position estimation in steps and 30. S. *. * * S

* * 20 In practice, it can be beneficial to use both the directional antenna 210 and the further, isotropic antenna 215 for direction estimation. For example, the isotropic or upward-directed further antenna can provide a reference signal level for each transmitter. This can be used to supplement or replace the theoretical prediction of signal level at step 40, discussed above. That is, the signal strength received at the omni-directional antenna can be used as a reference or predictor for the signal strength measured at the uni-directional antenna. This may be beneficial, since it can avoid over-reliance on a theoretical model, whose assumptions may not be satisfied in practice. For example, although a theoretical model will not be able to predict signal attenuation due to obstructions, a signal measurement at the omni-directional antenna will implicitly take account of real reception conditions of this kind.

For completeness, an example of a signal-strength prediction method for use in step 40 will now be described. According to one embodiment, the method relies on the obtained positions of the satellites and the camera, obtained in steps 20 and 30, respectively. From these positions, it is straightforward to compute the azimuth and elevation of each satellite, relative to the camera. In this embodiment, the received signal strength is assumed to depend on satellite elevation (that is, angle above the horizon). A theoretical relationship can be derived between these two variables, based on the known characteristics of the shaped beam pattern transmitted by GPS satellites (see, for example, the well know text: Kaplan, "Understanding GPS: Principles and Applications", Artech House); however, other effects may also be taken into account, such as the likelihood of obstruction and attenuation of the satellite signal in flight. Sample values for the variation of received signal power with elevation are as foJiows: * ** * * * I* Elevation Angle Gain : (degrees) (dB) 0 -8 -6 -4 * . 25 -2 -1 0 For elevations above 50 degrees (that is, from 50 to 90 degrees), the gain decreases again from 0. In other words, the maximum signal strength is expected for satellites at moderate elevations. As can be seen from the above values, the received signal strength varies weakly around these moderate angles.

The gain values can be applied to a nominal signal-to-noise ratio (SNR) to give expected values for received signal strength at varying elevations. A nominal (default) SNR of 47dB has been found to give good results in practice.

At this stage, the prediction of signal strength is independent of any antenna characteristics. For example, on the basis of the figures given above, the predicted received SNR for a satellite at an elevation of 25 degrees would be 45dB.

For completeness, an example of an estimation algorithm suitable for use in step 50 will now be described. In this embodiment, the antenna points along the line of sight (optical axis) of the camera. The directional radiation pattern of the antenna is modelled on a circularly symmetric hemisphere where gain varies according to the angular deviation of the line-of-sight from the optical axis. Sample antenna gain values for a patch antenna are as follows: Angle of Deviation Antenna Gain (degrees) (dB) * S.

S S S * S. -0.5

* *SS ________________________ ______________________ -1.0

________________________ ______________________ * . .

S. * --4.0 -6.0 -7.0 * 85 -8.0 -9.0 -9.5 -9.0 -8.0 -7.5 -7.0 A deviation of 0 degrees means that the line of sight to the satellite coincides with the optical axis (the pointing direction) of the camera. A deviation of 90 degrees would mean that the signal is being received in the plane of the patch antenna (that is, at its edge). The relationship between the angle of deviation, A, and the elevation, B, and azimuthal angle C is given by: cos A = cos B cos C The azimuthal angle, C, is the difference in the horizontal bearing between the azimuthal orientation of the camera and the azimuth of the satellite.

The example method proceeds by testing hypotheses against the data and the model. That is, the method performs a search over a number of discrete azimuth angles, comparing the expected and measured signal SNR values for each case. The angle which gives the best match of the measured data to the theoretical model is determined as the actual angle of azimuthal orientation. In the current embodiment, the method evaluates 8 cases, being the 8 points of the compass: N, NW, W, SW, S, SE, E, NE. For each case, and for each satellite, the angular deviation of the satellite is calculated from the * ,** elevation angle, B, and azimuthal difference, C, as described above. Based on **.* * SS.

the angular deviation, the antenna gain can then be read from the table. The gain value is added to the predicted SNR determined in step 40 (as described above). As will be recalled from the above description -in the current embodiment, the SNR prediction produced in step 40 is independent of * " antenna gain. Thus, combining the isotropic" SNR with the directional gain provides a specific estimate for the expected SNR for that satellite, if the camera were pointing in the direction being tested. This final estimate is compared with the measured SNR for the same satellite and the error for all satellites is combined. In the current embodiment, the absolute differences between final estimate and measured value are added together. After this calculation has been repeated for each of the 8 directions, the direction giving the minimum total absolute difference is chosen as the direction estimate.

For multiple directional antennae, as noted previously, the absolute differences between estimated and measured SNR can be summed across all antennae being used.

Clearly, exhaustive search is not the only optimisation strategy by which a direction estimate can be produced. The scope of the invention is thus not limited to any particular optimisation algorithm or class of algorithms.

Note that predicted signal levels may be produced for satellites which are not detected in practice. Thus, real signal strength measurements may not be available for all the satellites which are predicted by the model. These satellites could be ignored; however, the absence of expected received signals is useful information for the direction estimation algorithm. This optional information can be included simply by arbitrarily setting a relatively low io "threshold" SNR for the missing satellites. For example, in the current embodiment, a "threshold" SNR of 15dB is used. The method then proceeds as before, comparing the expected received SNR with the threshold value, in place of a real, measured SNR value.

Another approach -which implicitly avoids the problem of differing is numbers of satellites -is to compute the mean absolute difference, or other * ..s normalised value, in place of summed absolute values.

The estimation method described relies on a relatively basic model of antenna gain, with one degree of freedom (angular deviation from the axis).

More accurate direction estimation may be possible if the antenna gain characteristics are available in greater detail. For example, the variation of gain * with angle of elevation may differ from the variation in the azimuthal direction -that is, the gain pattern may not be circularly symmetric about the axis. If the gain pattern is different in these two dimensions, then rotation of the camera between portrait and landscape orientations will affect the predicted received signal strength. Other effects, such as the effect of the photographer's head and/or hand near the camera, can also be included in the model. Nonetheless, the method described above has been found to give good results in practice. It is also simple to implement, which is an advantage.

The embodiments described above use an antenna which faces in the same direction as the camera lens. This is preferable for the reasons already outlined above. However, if it is not possible to provide a forward-facing antenna -for example, due to design constraints -the method can still function with antennae in other orientations, provided the antenna is arranged so as to be directional in the azimuthal direction when the photograph is taken.

For example, an antenna facing to the left or right when the camera is in landscape orientation will have the desired directional properties in that orientation. However, when the camera is rolled (clockwise or anticlockwise) into portrait orientation, the directionality requirements will not be fulfilled, because the antenna will point either vertically up or vertically down. In this case, one solution is for the photographer to take at least one landscape photo, to allow direction estimation, when taking portrait photos.

io As noted previously above, the actual measured and predicted signal-strengths are related by the antenna gain-pattern. An additional significant parameter of the antenna is its polarisation. The signals of GPS satellite-transmissions are right-hand circularly polarised. The receiver antenna gain *.* S will therefore vary according to the polarisation of the antenna as well as the is direction of arrival of the signals. The situation is further complicated by the fact that signal-polarisation can be modified by reflection; however, if this is *5* taken into account, it can be used to eliminate reflected signals from the direction-estimation calculation. For example, if the antenna is right-hand circularly polarised, then reflected signals having different polarisation will be relatively attenuated, while signals directly received will be relatively strong.

Once again, like orientation selectivity, this polarisation selectivity conflicts to some extent with the goal of gathering enough satellite signals to perform position estimation. As described above, this drawback can be overcome by providing a further antenna, for GPS signal reception for positioning.

Despite the advantages of the direction estimation method, there will inevitably be circumstances in which direction estimation is flawed, ambiguous, or impossible. This will typically occur because no satellites are detectable (for example, indoors or underground); or because the field of view of the antenna is restricted, such that satellite signals are only received from one direction (for example, when taking a picture out of the window of a building). A particularly difficult case may arise when facing a building: in this case, the antenna radiation pattern (and presence of the photographer) will prevent reception of signals from behind camera; at the same time, the building prevents direct reception of signals to the front. Thus, the only signals received might be signals arriving from the rear but reflected off the building in front. This may result in a direction estimate which is the opposite of the true azimuth orientation.

For these cases and for improved usability in general, it may be beneficial to provide a user interface, to allow the photographer to select among multiple possibilities, correct incorrect direction estimates and/or enter directions manually when no estimate is available.

io The embodiments described above have all focused on an implementation of the method using a satellite positioning system. However, as will be readily apparent to a skilled person, the scope of the method is not limited in its application. The method in general enables direction estimation I...

from a known position, whenever received signal strengths can be measured for a plurality of transmitters whose location is also known. For example, the method may be applied using cellu'ar-telephony base-stations as the reference S..

transmitters. In one embodiment, the location of each base-station is stored in a database. Methods for determining the position of a mobile device in a *:*. cellular network are well known -the position of the receiver can be determined by base-station triangulation, for example. The method then proceeds as described above: predicting the expected signal strengths and comparing them with the measured strengths of the signals from the base-stations; and finally estimating direction using known antenna-pattern parameters.

As another example of a terrestrial implementation of the direction estimation algorithm, the transmitters may be WLAN access points. Again, provided the positions of the APs and the receiver are known or can be determined, the method is the same.

The various terrestrial and satellite-based approaches described above can also be combined, to increase the number and diversity of signal measurements available, thereby further increasing the accuracy of the system.

As will be immediately apparent to the skilled person, no real antenna is perfectly isotropic, in the pure theoretical sense. Thus, it could be said that any practically realisable antenna is to some extent "anisotropic" due to imperfections in the real radiation pattern. In this context, the skilled person will also doubtless understand that references in this description and appended claims to "anisotropic" antennae and "isotropic" antennae are intended to be construed in a practical sense as meaning "substantially anisotropic" and "substantially isotropic", respectively. For example, an isotropic antenna can be defined as one which has a gain-variation of less than 3dB between the maximum and minimum values in the relevant plane. Equivalently, an anisotropic antenna can be defined, for example, as one which has a * difference of more than 6dB between the maximum and minimum gain values over the relevant 360-degree angular range. The invention will provide better *** performance to the extent that the gain variation is large, for the specified anisotropic antennae, and small, for the specified isotropic antenna. That is, the anisotropic antennae should be as directional as possible and the isotropic *.* * antenna should be as directionless as possible. Note also that the performance of the method depends on the effective gain of each antenna in use in a given embodiment. This will depend on factors such as the design of the antenna and camera, as well as the presence of other nearby objects, such as the hand or head of the photographer.

Various other modifications will be apparent to one skilled in the art.

Claims

CLAIMS1. A method of estimating the azimuthal orientation of a camera having an attached antenna, the antenna having a predetermined directional radiation pattern which is anisotropic in the azimuthal plane when the camera is held in a normal orientation for photography, the method comprising: measuring the strength of signals received at the antenna from each of a plurality of transmitters; io obtaining the position of each of the plurality of transmitters; obtaining the position of the camera; * predicting the expected strength of each of the received signals; and estimating the azimuthal orientation of the camera based on the * .* . obtained positions of the camera and the plurality of transmitters, the measured signal strengths, the antenna radiation pattern, and the predicted signal strengths. **S
2. The method of claim 1, wherein the transmitters are satellites and the *:*. received signals are satellite signals.
3. The method of claim 2, wherein the step of obtaining the position of each of the plurality of satellites comprises at least one of: decoding a data message of a received satellite signal; and obtaining satellite-position or satellite-trajectory information via a communications network.
4. The method of any of claims 1 to 3, wherein the step of obtaining the position of the camera comprises estimating the position of the camera based on signals received from each of the plurality of transmitters.
5. The method of claim 4 wherein the signals for estimating the position of the camera are received at a further antenna, the further antenna having a predetermined radiation pattern which is isotropic in the azimuthal plane when the camera is held in a normal orientation for photography.
6. The method of claim 5, further comprising measuring the strength of the signals received at the further antenna, and wherein the step of estimating the azimuthal orientation of the camera uses the measured signal strengths for the further antenna.
7. The method of any preceding claim, wherein the camera has a plurality of attached antennae, each of the plurality having a different predetermined directional radiation pattern which is anisotropic in the azimuthal plane when the camera is held in a normal orientation for photography; *...,, wherein the method comprises measuring the strength of signals * .** received at each of the plurality of antennae from each of the plurality of transmitters; and * wherein the step of estimating the azimuthal orientation of the camera is *** based on the measured signal-strengths for each of the plurality of antennae, the antenna radiation patterns, and the predicted signal-strengths. ** * *
S S

* o 8. The method of any of claims 1 to 6, wherein the camera has a plurality of attached antennae, each of the plurality having a different predetermined directional radiation pattern which is anisotropic in the azimuthal plane when the camera is held in a normal orientation for photography, wherein the method comprises: selecting one or more antennae from among the plurality; and measuring the strength of signals received at each of the selected antennae from each of the plurality of transmitters, and wherein the step of estimating the azimuthal orientation of the camera is based on the measured signal-strengths for each of the selected antennae, the antenna radiation patterns, and the predicted signal-strengths.
9. The method of any preceding claim, further comprising: * 26 sensing an angle of elevation or an angle of roll of the camera and/or the antenna; and estimating the azimuthal orientation of the camera based on the sensed angle.
10. The method of any preceding claim, comprising estimating the azimuthal orientation of the camera based on a predetermined polarization of the antenna.
11. The method of any preceding claim, wherein the step of estimating the azimuthal orientation of the camera is performed after an intentional non-zero delay after the signals are received.
12. A computer program comprising computer program code means adapted to perform all the steps of any preceding claim when said program is run on a computer. a
13. A computer program as claimed in claim 11 embodied on a computer-readable medium.
14. Apparatus adapted to estimate the azimuthal orientation of a camera, the apparatus comprising: an antenna attachable to the camera and having a predetermined directional radiation pattern which is anisotropic in the azimutha' p'ane when the camera is held in a normal orientation for photography; a receiver, electrically connected to the antenna and adapted to receive signals from each of a plurality of transmitters; and processing means, adapted to: measure the strength of the received signals; obtain the position of each of the plurality of transmitters; obtain the position of the camera; predict the expected strength of each of the received signals; and to estimate the azimuthal orientation of the camera based on the obtained positions of the camera and the plurality of transmitters, the measured signal strengths, the antenna radiation pattern, and the predicted signal strengths. * * * * * * ** S... * . S... S... * .. *5 S I. * . . * . S. * * I * S S.