This invention relates to detecting and analysing the biomechanical motion, for instance the gait, of a human or animal.

When humans are running or walking it can be useful to be able to detect and analyse their gait. For example, information about gait can help to improve the performance of sportsmen. It is known to detect certain information about a runner's gait using sensors such as accelerometers located in shoes. (See US 8,375,784). However, this has a number of disadvantages, notably that it requires the user to be wearing special shoes, or shoes equipped with suitable sensor attachments.

A growing trend is for users to undertake sports activities whilst carrying a smartphone. The smartphone may include a satellite positioning sensor which can sense the user's location, and accelerometers that can sense motion of the phone. The smartphone may also communicate with accessories carried by the user, such as sensors attached to a shoe that can sense motion of the user's feet. In that way the smartphone can gather information about the motion of the user's feet. Whilst the smartphone is a convenient device for gathering data, this system still requires sensors to be attached to the user's shoes.

US 2013/0190657 discloses measuring the acceleration of a runner over time, and outputting the instantaneous or time-averaged values of those parameters.

Another way to monitor a person's gait is to have the person walk or run on a treadmill whilst capturing a video of their feet. The video can then be replayed to analyse the person's gait. This process can only be adopted in a gym or other static environment.

"Activity Recognition from User-Annotated Acceleration Data" (Ling Bao and Stephen S. Intille, PERVASIVE 2004, LNCS 3001 , pp. 1-17, 2004) describes discerning a range of activities undertaken by a user by means of multiple accelerometers worn simultaneously on different parts of the body. This system is inconvenient for casual users due to the number of accelerometers that are used.

It would be desirable to be able to monitor the gait of a user whilst exercising at an arbitrary location, with fewer constraints on the equipment to be carried by the user.

According to one aspect of the present invention there is provided an arrangement for analysing the biomechanical motion of a wearer, the arrangement comprising: an acceleration sensing unit attached to the upper body of the wearer for sensing acceleration of the wearer's upper body and generating a series of acceleration values indicative of that motion; and a processor communicatively coupled to the acceleration sensing unit for processing the acceleration values, the processor being configured to gather real-time acceleration values from the acceleration sensing unit and to compare the real-time acceleration data with one or both of: (a) historic acceleration data derived from the acceleration sensing unit and (b) predetermined acceleration features indicative of biomechanical motion quality to form an output indicative of the current state of the biomechanical motion of the wearer.

According to a second aspect of the invention there is provided an arrangement for analysing the biomechanical motion of a wearer, the arrangement comprising: an acceleration sensing unit suitable for attachment to the upper body of the wearer for sensing acceleration of the wearer's upper body and generating a series of acceleration values indicative of that motion; and a processor communicatively coupled to the acceleration sensing unit for processing the acceleration values, the processor being configured to gather real-time acceleration values from the acceleration sensing unit and to compare the real-time acceleration data with one or both of: (a) historic acceleration data derived from the acceleration sensing unit and (b) predetermined acceleration features indicative of biomechanical motion quality to form an output indicative of the current state of the biomechanical motion of the wearer; and directions for attaching the acceleration sensing unit to the upper body of a human. According to a third aspect of the invention there is provided an arrangement for analysing the biomechanical motion of a wearer, the arrangement comprising: an acceleration sensing unit comprised in a treadmill for sensing vertical acceleration of a user ambulating on the treadmill and generating a series of acceleration values indicative of that motion; and a processor communicatively coupled to the acceleration sensing unit for processing the acceleration values, the processor being configured to gather real-time acceleration values from the acceleration sensing unit and to compare the real-time acceleration data with one or both of: (a) historic acceleration data derived from the acceleration sensing unit and (b) predetermined acceleration features indicative of biomechanical motion quality to form an output indicative of the current state of the biomechanical motion of the user.

The acceleration sensing unit may comprise one or more accelerometers. The acceleration values may be values derived from one or more accelerometers.

The acceleration sensing unit may be comprised within an earpiece adapted to be attached to the ear of a wearer. The earpiece may have an earbud configured to be lodged in the ear canal of a wearer. The earpiece may be configured to wrap at least partly around the pinna of a wearer. The processor may be comprised in the earpiece. The processor may be comprised in a housing separate from the earpiece. The earpiece and the processor may be provided with wireless communication devices whereby the acceleration values can be transmitted from the earpiece to the processor. The earpiece may comprise a loudspeaker.

The arrangement may comprise a memory, for example for storing sensed acceleration data.

The processor may be configured to detect periodicity in a portion of the real-time acceleration data sensed for a wearer, quantify one or more predefined metrics as applied to that portion of the acceleration data, compare the quantified metrics with reference data and thereby estimate a measure of the wearer's biomechanical motion. The reference data may be derived from real-time acceleration data previously sensed for the wearer.

The processor may be configured to adapt the reference data over a period of time in dependence on real-time acceleration data sensed on multiple occasions for the wearer.

The predetermined acceleration features may be indicative of one or more biomechanical anomalies. The predetermined acceleration features may be indicative of any one or more of fatigue, gait asymmetry, stumbling and adaptation of gait to uneven terrain. At least one of the predetermined acceleration features may be indicative of fatigue. The system may comprise a display for providing a visual output to the wearer. The processor may be configured to cause the display to display an indication of fatigue in response to detection of that one of the acceleration features in the real-time data.

At least one of the predetermined acceleration features may be indicative of gait asymmetry. The system may comprise a display for providing a visual output to the wearer. The processor may be configured to cause the display to display an indication of fatigue in response to detection of that one of the acceleration features in the realtime data.

The predetermined acceleration features may be indicative of one or more predetermined forms of physical activity. The predetermined forms may include one or more of walking, running, swimming and cycling.

The processor may be configured to compare the real-time acceleration values with the predetermined forms, determine which of the predetermined forms best matches real-time acceleration data and store an indication of that predetermined form in respect of the time associated with that real-time acceleration data. The arrangement may comprise an attachment specifically adapted to attach a housing comprising the accelerometers to the upper body of the wearer.

The processor may be configured to gather real-time acceleration values from only the acceleration sensing unit.

The arrangement may comprise a second acceleration sensing unit attached to the upper body of the wearer for sensing acceleration of the wearer's upper body and generating a series of acceleration values indicative of that motion. The processor may be configured to gather the real-time acceleration values additionally from the second acceleration sensing unit.

The real-time acceleration data for each time may comprise data representative of accelerations in multiple non-parallel axes.

The processor may be configured to compare the data representative of accelerations in different axes to form the output indicative of the current state of the biomechanical motion of the wearer.

The processor may be configured to process the data representative of accelerations so as to estimate the current angular attitude of the user's head.

The processor may be configured to process the data representative of accelerations so as to estimate the current angular attitude of the user's spine.

The processor may be configured to process the data representative of accelerations so as to estimate the current pose of the user.

The arrangement may comprise a user output device. The arrangement may be configured to provide an output for indicating to a user an output representing the user's current head attitude, spine attitude and/or pose. The acceleration sensing unit may be adapted to be worn by an animal.

The device may provide an indication of the posture of a wearer, or may classify the activity of the wearer into one of a series of predetermined classes and may then contemporaneously display an indication of that class, or may store that class in association with the current time.

The present invention will now be described by way of example with reference to the accompanying drawings. In the drawings:

Figure 1 is a schematic diagram of a first embodiment of gait capture equipment.

Figure 2 shows examples of apparatus configured for attaching a sensing device to a user's upper body.

Figure 3 is a schematic diagram of a second embodiment of gait capture equipment. Figure 4 illustrates a process for forming a model of gait. Figure 5 illustrates a process for analysing real-time gait data. Figure 6 illustrates a user interface for posture training.

The equipment to be described below takes data from one or more accelerometers carried in a single unit borne on the upper body of a user, for example the user's torso, neck or head. By sensing those accelerations at a series of moments over time a time-series of accelerations is produced. Accelerations sensed most recently (e.g. during a window extending back between 1 and 10 seconds from the latest sensed acceleration) can be compared with one or both of: (i) specific acceleration features derived from the analysis of previously logged data for the same and/or other users and identified as correlating with certain gait artefacts; and (ii) comparisons between assessments of the recently sensed accelerations according to one or more metrics and predetermined thresholds for those assessments. The metrics could, for example include the period of each step; the difference between the period of each step; the period of the immediately preceding step; the difference between the period of each step and the period of the last but one step; and the variability of step properties within a window of data. The thresholds could be determined in dependence on average accelerations previously obtained for the same user, or by means of other processes that are described below. In that way changes of walking or running technique or form can be identified. Using the approaches described below it has been found that useful data about a user's gait data closely indicative of fatigue, changes in ambulation style, stumbling and adaptation to different terrain can be derived from accelerometers contained in just a single unit borne on the user's upper body. Accelerometers carried in multiple locations are not required.

Figure 1 shows a first example of equipment that can be carried by a user to capture gait information. The equipment comprises a self-contained primary unit 10. The primary unit comprises sensors for sensing motion of a user wearing the unit and a processing engine for processing the sensed data to extract features from the sensed data and for analysing those features so as to alert a user to certain types of feature.

The primary unit 10 of figure 1 comprises a housing 1 1 . Within the housing are a battery 12, one or more accelerometers 13, a processing section 14 and a memory 15. The primary unit also comprises a user interface constituted by a display 16 and a series of user-actuable switches 17 which are exposed on the exterior of the housing. The display and switches could be combined into a touchscreen device.

The battery 12 stores energy for powering the other components.

Collectively, the accelerometers 13 are preferably capable of sensing acceleration in any direction. To that end the accelerometers could be provided by set of three integrated or discrete accelerometers arranged to sense acceleration along respective orthogonal axes. Alternatively, the accelerometers could be arranged to sense acceleration along one or more specific axes that are orientated in a predetermined fashion with respect to the user's body: for example there could be only a single accelerometer, which could be arranged with respect to the user's body so as to sense acceleration in a vertical or substantially vertical direction when the user is walking or running; or there could be only two accelerometers, which could be arranged with respect to the user's body so as to sense acceleration in a vertical or substantially vertical direction and in a lateral or substantially lateral direction when the user is walking or running. In a preferred example the accelerometers provide a time series of acceleration values. The accelerometers could be mechanical accelerometers. Preferably the accelerometers measure acceleration directly.

The outputs of the accelerometers 13 are passed to the processing section 14. The processing section could perform its functions using dedicated hardware, using a general purpose processor executing software code, or using a combination of the two. In this example, the accelerometers provide a digital output representative of acceleration, and a processor 18 executes software code stored in a non-transient way in software memory 19 in order to perform its functions. The processor is arranged to analyse the data received from the accelerometers so as to detect features of interest in the data. Some examples of how this can be done are described below, but in general the processor may optionally filter the sensed acceleration data to remove noise. The processor may then compare a set of recent acceleration samples with one or both of (i) a reference set of data based on samples previously collected by the processor, representing a normal gait of the user (and potentially other users) carrying the device 10; and (ii) one or more predetermined data features or patterns representative of certain specific gait anomalies. Based on a predetermined model defined in the software stored in memory 19 and that the processor 18 runs, the processor may then determine whether the recently detected samples deviate from the reference data in a predefined way, or match one of the predetermined data features. If such a deviation or match is detected then the processor may alert the user, store a record of that event in memory 15, or transmit a message to a remote device so that it can log the event or generate an alert. The operations performed by the processor will be described in more detail below. The reference set of data for a user may be updated over time, as the equipment gathers information on the motion of the user. This process allows a comparison of the user's current motion with the reference motion to more sensitively indicate short- term deviations of the user's current motion from the user's normal motion.

The processing section 14 can write historic acceleration data to the memory 15, from where it can be retrieved by the processor to help with future analysis or transmitted to a remote entity for viewing or additional analysis.

The primary unit 10 could include other sensors and interfaces. For example, it could include any one or more of:

a satellite positioning sensor which could provide information on the location and/or altitude of the primary unit to the processing section 14;

a wired or wireless communication interface for allowing communication between the processing section 14 and a remote device, for example a cellular telephony interface, a Wi-Fi (IEEE 802.1 1 ) interface, a Bluetooth interface or a USB (universal serial bus) interface;

an environmental sensor such as a temperature sensor or a humidity sensor, which could provide information on the environment around the primary unit to the processing section 14;

gyroscopic sensors, for example one or more rate gyro sensors;

a loudspeaker, by which alerts could be provided to a user; and

a microphone, which in combination with the loudspeaker could permit the primary unit to function as a telephone.

It is desirable for the housing 1 1 to be carried on the upper body of the user: that is to say the torso, neck or head. This has been found to provide for useful measurement of the user's gait. Examples of ways in which the housing can be carried are as follows:

around the user's waist, for example on a waist strap attached to the housing, or by the housing being clipped to or integrated in the waistband of a garment being worn by the user; around the user's chest, for example on a chest strap attached to the housing; attached to the user's head, for example by means of a strap attached to the housing or by the housing being integrated into an article of headgear such as a pair of glasses, an earpiece, a helmet, a cap or a sweatband or headband;

attached around the user's neck, for example by a neck strap or collar;

attached to another garment such as a vest or bra.

It has also been found that mounting the housing 10 on the user's upper arm can yield useful results. That may be done by means of an arm strap attached to the housing. The housing 10 could be provided with a mechanical connector 20 by means of which it can be releasably connected to a range of apparatus for attaching it to a user's body. For example, figure 2 shows the housing 1 1 attached to a chest strap 21 and to a waist band clip 22. Whatever the way in which the housing 1 1 is attached to the body of the user, it is desirable for it to be firmly attached to the body so that it moves together with the part of the body to which it is attached. In that way the data gathered by the accelerometers can be more representative of the motion of the user's body.

Figure 3 shows an alternative embodiment. In this embodiment the components shown in figure 1 are distributed between a sensor unit 30 and a base unit 31 . The sensor unit 30 is in the form of an earpiece. It comprises the accelerometers 13, a local battery 32 and a first wireless communication device 33 for communicating with the base unit. The base unit comprises a second wireless communication device 34 for communicating with the first wireless communication device 33 and the battery 12, processing unit 14 and memory 15 of the unit 10 shown in figure 1 . In this embodiment the accelerometers 13 sense motion of the earpiece 30, and the sensed data is provided to the first wireless communication device 33. The first wireless communication device transmits the sensed data to the second wireless communication device 34, which provides it to the processing unit 14 for processing as in the embodiment of figure 1 .

The earpiece 30 can also include a loudspeaker 35. The processing unit 14 of the base unit may be capable of replaying audio from data stored in the memory 15 to the loudspeaker 35 via a wireless communication link between transceivers 33 and 34. Alternatively the earpiece itself may be capable of replaying audio through the loudspeaker, using an audio processor integrated in the earpiece.

The earpiece may attach to the user's ear by a formation 36 of its housing that wraps around the ear of a wearer and/or by virtue of an ear bud 37 that can be inserted into and lodge in the ear canal of the wearer. These mechanisms allow the earpiece 30 to be attached firmly to the user's head, so that it moves with the user's head. This allows the accelerometers 13 in the earpiece to gather accurate data about the motion of the user's head. It has been found that mounting the accelerometers in an earpiece is a particularly convenient approach since it provides good data about the user's gait and because many users find an earpiece to be a convenient form in which to carry a device attached firmly to their body. Furthermore, when the earpiece can also replay audio: for example music, to a user the user can enjoy that audio whilst exercising.

In an alternative embodiment the processing section 14 and/or the memory 15 of the base unit may be incorporated in the earpiece. Then there need not be a separate base unit 31 .

The processing performed by the processing section 14 will now be described in more detail with reference to figures 4 and 5.

One way in which the processing section 14 can detect elements of interest in the live acceleration being sensed from the motion of a user is by comparing it with features that have been previously determined to be representative of certain types of gait anomaly. Such anomalies can include stumbling, asymmetric motion, loss of form attributable to fatigue or lack of co-ordination, adjustment of gait to uneven ground and so on. It is desirable to identify such features before the processing section 14 operates on live data, by pre-processing a set of reference data. The reference data can be processed offline, for example using a server, to develop a model. Elements of the model can then be incorporated in the algorithms implemented by the processing section 14. Figure 4 shows steps in the development of the model. Sample users perform activities such as running or walking whilst wearing personal devices similar or substantially identical to those in which the model will be used (e.g. devices 10, 30 and 31 ) and whilst also being monitored using other more detailed techniques: for example by means of a series of accelerometers or position sensors mounted on the users' shoes and/or legs. These generate two pools of time-series source data: a set of aggregated running data 40, derived from the more detailed sensing, and a set of training data 41 , derived from the replica personal devices. The aggregated running data represents the subjects' gaits in considerable detail. The aggregated running data can be analysed to isolate instances of anomalies of the types described above. (Step 42). The timings of these anomalies in the aggregated running data are known. The training data 40 from the corresponding times can then be analysed (step 43) to identify features in the training data that occur at the same time as the anomalies. To do this, parts of the training data from time windows around each anomaly can be analysed either to find absolute features or by comparing their content with time- averaged training data. The time-averaged training data could itself be formed by identifying periodicity in the training data on a scale of 0.3 to 1 .5 seconds, representative of the period of the paces of a typical user, and then averaging between the accelerations at corresponding points in each cycle. The period of that periodicity could be 0.3 to 1 .0 seconds for running data or 0.7 to 2.0 seconds for walking data. Such periodicity could be identified in windows of data of total duration that may, for example, be from 3 to 10 seconds. In this way features can be identified that occur in the training data at times associated with the occurrence of anomalies in the aggregated running data. The features can then be classified, for example by multilayer proceptron (MLP) and/or naive Bayes (NB) bagging in order to classify them as relatively representative or unrepresentative of the anomalies. (Step 44). Where features in the training data are found to have a strong correlation with the anomalies they can form part of a model (45) of the behaviour of the training data with user activity. For example, the model may include:

the correlation of an increase over time of the maximum vertical deceleration during a stride with increasing fatigue of a user; the correlation of an aperiodicity in successive strides of greater than a certain proportion of the stride period with the onset of a stumble; or

the correlation of a difference of greater than a certain percentage between (i) the peak leftward acceleration during a pace (two successive strides) and (ii) the peak rightward acceleration during that pace with a kinematic asymmetry of gait.

the correlation of an increase in variability of stride period, mean stride period (or other metric) within a window of data with increasing user fatigue.

Figure 5 shows how the model 45 can be implemented in processing section 14. The software stored in memory 19 includes definitions of features contained in the model. Live running data 50 is gathered from the accelerometers 13. The live running data is a time series of acceleration samples. The live running data is compared (step 52) with historic acceleration data 51 derived from the same user and stored by the processing section 14 in memory 15. This comparison allows the processing section to detect both features that are composed of instantaneous characteristics (such as the feature of a forwards acceleration greater than a predetermined threshold value) and features that are composed of changes over time in the collected acceleration data (such as an increase over time of the maximum vertical deceleration during a stride). The comparison of the live running data with the stored data may be performed by detecting a pace/stride period in a window of live data and comparing the accelerometer data for that period with averaged data for a pace/stride period in the stored data, or by means of classified data as described in more detail below. The live running data 50 and the comparison of the live running data with the historic data derived at step 52 are both compared at step 53 with the features of the model, as defined in the software stored in memory 19, that have been deemed to be representative of anomalies. The output of step 53 is an indication of whether or not the live running data includes anomalies, as indicated at 54, 55. That output can be used to provide information on gait. (Step 56). The information may be provided in a number of ways, for example:

1 . The processing section can cause a device carried by the user to provide a visual, audible or other form of alert when a feature indicative of an anomaly of a certain type is detected in the live data. For example, when the live data includes a feature indicative of excessive fatigue the processing section can alert the user so that the user can choose to stop and avoid risking injury. Or as the live data indicates increasing fatigue an indicator of the estimated level of fatigue can be presented to the user. The display 16 illustrated in figure 1 includes a readout of estimated fatigue level.

2. Information indicating the timing of the occurrence of a feature indicative of anomaly can be stored in memory 15. For example, the times when the user's gait is representative of running on uneven ground can be logged in the memory 15. That information may later be uploaded to separate analysis software, for example via the internet, so that the user can better review their performance.

3. The processing section can cause an alert to be sent to a remote device when a feature indicative of anomaly of a certain type is detected in the live data. For example, when the live data includes a feature indicative of stumbling or unsteadiness the processing section can cause an alert to be sent to a predetermined address by means of a wireless communication interface associated with the processing section. In this way, if the device 10, 30, 31 is being worn by an infirm person, a carer can be alerted when the user might have fallen over.

In one embodiment a single housing could contain the accelerometer(s) and a processor configured to process the data gathered by the accelerometer(s) by comparing it with historic acceleration data and/or predetermined acceleration features to generate an output that provides information on the wearer's gait. In another embodiment, the accelerometers could be in one housing and the processor could be in a portable device communicatively coupled with the unit containing the accelerometer(s). This would be the case when the data is analysed by a mobile phone carried by the wearer of the accelerometer(s). In another embodiment the processor could be housed in a fixed, non-portable installation remote from the wearer of the accelerometer(s) but still communicatively coupled with the accelerometer(s). This would be the case when the data is analysed by a remote data server.

In a second processing path, the features identified in steps 52 and 53 can be stored in memory 15. (Step 57). This allows them to be analysed later, for example to help the user improve their gait or to help refine the model 45. This can help provide the user with a personalised analysis of their running performance, for example to improve their running style. Additional data may also be stored in memory 15. For example, the memory may store data relating to any one or more of the position, altitude, heart rate or respiration rate of the user at specific times. That data can then be uploaded to a remote computer for display or analysis.

It is desirable to simplify the processing of the sensed data about a user's gait so as to reduce the time and energy required to perform the processing and thereby improve the battery life or reduce the size of the device(s) carried by the user. In one approach the sensed data gathered from the movement of a user could be compared directly with time-averaged data gathered from the movements of the same user or with data gathered from the movements of the same user when he or she is fresh. Then a deviation of a certain type of the instantaneous data from the time-averaged data could signify a stumble, and a deviation of the instantaneous data from the when-fresh data could indicate a deterioration in form. However, this approach requires a considerable amount of processing. It has been found that highly significant information about a user's gait can be derived by a much simpler approach. That approach can be summarised as:

1 . Detecting a periodicity in the sensed acceleration data derived from the movements of a user. That periodicity may correspond to the period of a single step or pair of steps.

2. For the sensed acceleration data during each period, quantifying a series of predetermined metrics.

3. Comparing the quantified value for each of those metrics to one or more thresholds.

4. Assessing the current status of the user's gait on the basis of those comparisons. This approach avoids the need to retain detailed information about the acceleration patterns of previous strides. The metrics can be selected so that the quantification step can be performed based on a simple analysis of the acceleration data during a single short window of data, which also avoids the need for detailed analysis. The pattern of vertical or net acceleration of a runner's body follows a generally sinusoidal pattern, with a stride corresponding to one cycle of the pattern. The step of detecting periodicity may be based on the identification of maxima or minima of the signals from individual accelerometers (e.g. from an accelerometer aligned in an approximately vertical axis), or from the overall acceleration determined by a vector sum of the accelerations measured by multiple accelerometers. To overcome noise or short-term variability in the acceleration data the data may be smoothed, e.g. over a period less than 0.1 s. The maximum may be sought in a window extending between one instant when the relevant acceleration crosses through a mid or zero value and the second subsequent such crossing.

The predetermined metrics may, for example, include any one or more of the following:

- Step duration features:

- step duration;

- difference in duration between one step and the previous step;

- difference in duration between two consecutive steps and the previous two consecutive steps;

- odd step asymmetry, i.e. the difference between successive steps;

- even step asymmetry, i.e. the difference between successive steps with the same leg.

- Frequency domain features, e.g. as determined by an FFT of the acceleration data for a step:

- amplitude of frequency spectrum in one or more predetermined frequency bands, e.g. 0 to 1 Hz, 1 to 2Hz, 2 to 3Hz, in each axis along which acceleration is directly measured and/or in relation to the vector sum of the measured accelerations;

- the frequencies of the top n peaks in the frequency spectrum of the acceleration data, where n may for example be five.

- Time domain features as determined based on acceleration data over the period of a step:

- mean acceleration in each axis along which acceleration is directly measured and/or in relation to the vector sum of the measured accelerations; - RMS acceleration in each axis along which acceleration is directly measured and/or in relation to the vector sum of the measured accelerations;

- standard deviation of acceleration in each axis along which acceleration is directly measured and/or in relation to the vector sum of the measured accelerations;

- coefficient of variance of acceleration in each axis along which acceleration is directly measured and/or in relation to the vector sum of the measured accelerations;

- cross-correlation between the acceleration values sensed for motions on different axes.

The model, which takes into account a set of pre-determined metrics as listed above, used to classify the live running data may be determined from (i) analysis (e.g. averaging) of historical data for the respective user; (ii) analysis (e.g. averaging) of data for a set of reference users; or (iii) an estimate of a normal gait for the user based on a set of data for reference users and data entered by the user indicating his or her physical characteristics (e.g. age, height, running ability, weight). The historical data for a particular user may be determined over a period of 0 to 10 seconds prior to the instantaneous data and/or over a period of 10 to 300 seconds prior to the instantaneous data and/or over a period from the start of the current exercise activity and/or over data including one or more previous exercise activities. In this way the system can learn what behaviour is normal for a user, e.g. by averaging the quantified values of the metrics over the period in question. The comparison of the instantaneous quantified metrics against such averages may be performed by identifying whether an instantaneous metric is above or below the corresponding average, or whether the instantaneous metric deviated from the corresponding average by greater than a predetermined threshold.

The algorithms described above enable a unit carried by a user to detect features indicative of normal or anomalous gait in real time. That detection is performed by comparing live data on the accelerations experienced by a device attached to the user's upper body with previously determined behaviours established during the collection of reference time series data. The analysis of the user's gait may be employed for a range of purposes. 1 . The user's gait may be monitored over the course of an exercise session, in the manner described above. As the session proceeds a standard gait pattern (i.e. the pattern of accelerations detected over the course of a pace/stride, or the typical or average values of one or more of the predetermined metrics) for the user can be established. If the gait pattern in current live data differs from the standard gait pattern in respect of one or more features previously determined to be indicative of fatigue, and whose details have been stored in memory 19, the device can present the user with a corresponding output, for example an indication that the user is estimated to be fatigued. In that way, the current gait pattern for a user can be compared with a set metric. In an alternative arrangement, the current gait pattern for a user can be compared with a metric that evolves over the course of time. That allows the system to advise a user whether their gait is improving or deteriorating. In one way to implement that arrangement the processor repeatedly compares the live acceleration data against one or more metrics to determine a value indicative of the fit between the live acceleration data and the metric(s). The metric(s) may be indicative of one or more characteristics of sound biomechanical activity (e.g. a balanced activity pattern between left and right over a cycle such as a pair of successive strides, a pattern previously determined to be representative of the motion of an expert performer or a regular periodicity of pattern) or of unsound or anomalous biomechanical activity. That degree of fit can then be stored, and when the analogous degree of fit is subsequently compared the device can output an indication of whether the later-determined degree of fit is better or worse than the previously-determined degree of fit. In a second way to implement that arrangement the processor processes live acceleration data to form a set of data representative of one or more characteristics of the live motion: for example by filtering the data in time or frequency. That processed acceleration data can be stored as data 51 (figure 5). The processed acceleration data can then be compared with later-determined live acceleration data to estimate changes in the wearer's biomechanical activity between the two sets of data.

2. The gait pattern in live data may be determined to include a feature that has previously been determined to be indicative of a gait defect, such as imbalance or muscular weakness or instability. If such a feature is detected, or is detected consistently over a period of time, then the device can present the user with a corresponding output, for example an indication that the user has an imbalanced gait. During running the user's body will sway to the right and left as the user's right and left feet hit the ground. When a detected anomaly relates to the behaviour of a specific foot of the user then the relative timing of the occurrence of (a) the features indicative of the anomaly and (b) lateral accelerations indicative of left or right foot strikes can be used to establish whether the anomaly pertains to the user's left or right foot.

3. Features of the gait pattern in live data derived from a runner may indicate elements of the style of the runner: for example whether he or she is striking on their toe or heel, their cadence, their length of stride, whether they are running on smooth or rough terrain, uphill or downhill, their level of fatigue and so on. That information can be stored in the memory 15 for subsequent review and analysis. The increase in the user's level of fatigue and/or any deviation in the style of the user can be detected and reported.

4. The gait information can be used to assist in identification of illness or injury, or the onset thereof, to monitor the progress of a user to recover function after surgery, to monitor and check the fitment of effect of prosthetics, and to monitor the effect of specific shoes or other clothing or accessories on performance.

5. Features of a user's gait may indicate the activity being undertaken by the user. For example, the user's gait may differ as between walking and running, and other aspects of the accelerations that are detected may indicate whether the user is undertaking activities such as climbing stairs, sitting, lying down, riding a bicycle or driving a car. When the detected motion is indicative of such an activity, that information may be stored in memory 15.

6. A device carried by the user may provide an indication of the current state of the user's gait. In that way the user can receive feedback that can permit him or her to maintain a desirable gait.

The information that is output about the current state of the gait of the wearer may relate to one or more of the regularity (i.e. maintenance or not of a consistent time period between successive strides or paces), symmetry (i.e. maintenance or not of a symmetric acceleration profile between successive strides) or poise (i.e. the change in acceleration over time during a single stride or pace). The series of acceleration values provided by the acceleration sensing unit preferably provide a resolution of greater than ten values per second, enabling the profile of accelerations during a single stride to be analysed. The acceleration values are preferably derived from measuring the acceleration of the part of the body to which the acceleration sensor(s) is/are attached, not by differentiating velocity or position measurements.

Instead of monitoring gait (the biomechanical behaviour of a user for bipedal locomotion, such as walking or running), a device using the principles described above could be used to monitor biomechanical performance or behaviour in other activities such as ball sports (e.g. football, rugby, tennis, golf or hockey), cross-country skiing, cycling, swimming, rowing, skating, kayaking. It may also be used to monitor run-ups in athletic or gymnastic activities, and to monitor performance or behaviour on static gym equipment.

Figure 6 illustrates a further example of how the equipment described above may be used. A sensor having one or more acceleration sensors is mounted in an article of headgear, such as an ear-mounted speaker unit or headphone, a hat or a headband. The sensor is capable of sensing the attitude of the wearer's head relative to vertical, about both a transverse axis (i.e. the angle front to back) and a longitudinal axis (i.e. the angle side to side). The sensor is communicatively coupled to a monitor device, which could be a mobile phone 60 (see figure 6). The sensor could be coupled to the monitor device by a wireless channel such as a Bluetooth link. The sensor is configured to transmit sensed acceleration values to the monitor device. The monitor device is configured to process the sensed values by means of software stored non- transiently on the device, and to thereby derive data for presentation to the user, e.g. on a display provided on the monitoring device.

This system may be used as follows. First, the system is trained to a known attitude of the user's head (e.g. vertical about both axes) and a known pose of the user (e.g. sitting upright). When subsequent accelerations/decelerations of the sensor are informed to the monitoring device the monitoring device integrates those values to determine changes in position and angles of the user's head, which are added to the previously known position and angles to provide a current estimated position and angles of the user's head.

Once the monitoring device has been initialised as described above it can determine subsequent attitudes of the user's head and subsequent poses of the user in the following manner.

- A change in attitude of the user's head can typically be detected as an acceleration and subsequent deceleration of the sensor indicating motion along a path that is centred in the region from 5 to 40cm below the sensor, and of a distance not more than 60cm. As described above, the monitoring device keeps an estimate of the current position of the user's head and updates that estimate when acceleration is sensed. The monitoring device can then provide the user with an output indicating the current attitude of their head about either or both axes. (See figure 6 at 61 ).

- When the user's current pose is sitting, a change to a standing pose can be detected through acceleration indicating an upward motion of greater than a threshold distance. The threshold may, for example, be greater than 30cm. A change from standing to sitting can be detected through the opposite acceleration profile. When the user's current pose is standing, a change to walking can be detected through acceleration indicating a horizontal motion of greater than a threshold distance. The threshold may, for example, be greater than 1 m. A change from walking to standing can be detected through that motion ceasing. When the user's current pose is sitting a change to a lying pose can be detected through acceleration indicating downward motion of greater than a threshold distance together with a rotation of the user's head to an approximately horizontal attitude. The threshold may, for example, be greater than 30cm. The head may be determined to be in a lying position when it has rotated between 60 and 120 degrees from the estimated vertical position. The monitoring device can maintain a record of the pose the user is currently estimated to have, and can output that to the user, e.g. using a display of the monitoring device.

- The monitoring unit can also estimate the user's posture: e.g. the net angle of the user's spine relative to vertical. A change in posture can be detected through acceleration indicating a rotational motion about a centre more than a threshold distance below the user's head. The threshold may, for example, be greater than 50cm. The monitoring device can maintain a record of the posture the user is currently estimated to have, and can output that to the user, e.g. using a display of the monitoring device. (See figure 6 at 62).

The monitoring device may maintain a running average of the user's head angle and/or spinal angle (see figure 6 at 63) and/or a record of the total time spent by the user in each of a number of poses (see figure 6 at 64).

The monitoring device and the sensor could be integrated in a single housing or could be separate units.

A system of this type can be used to assist a user to improve their posture and to log the time spent by a user in different activities.

In the examples given above, acceleration of the user is determined by means of one or more accelerometers carried by the user. When the user is using a static treadmill the vertical acceleration of the user can instead be determined by one or more sensors in the treadmill.

Using analogous principles to those described above the system described above can be used to monitor the gait or other biomechanical behaviour of animals rather than humans. For example, to permit it to be worn by an animal the device comprising accelerometers could be attached to a collar (e.g. for a dog), an item of tack such as a saddle (e.g. for a horse) or to an eartag (e.g. for a cow or sheep). A device of the type described above may be used in relation to an animal in order to help monitor or improve its sporting performance, or for veterinary purposes. For example, a change in gait in an animal may be indicative of illness or lameness, of it being about to give birth or of it reaching a certain stage of physical development.

By "upper body" is preferably meant any part of the anatomy of a human or animal wearing the device that is articulated with respect to the both/all legs (including the feet and in the case of a four-legged animal the shoulder). Examples of such body parts include the pelvis, chest, neck and head. Attaching an acceleration sensing device to such a part permits a single device to sense accelerations deriving from both/all legs of the wearer.

The applicant hereby discloses in isolation each individual feature described herein and any combination of two or more such features, to the extent that such features or combinations are capable of being carried out based on the present specification as a whole in the light of the common general knowledge of a person skilled in the art, irrespective of whether such features or combinations of features solve any problems disclosed herein, and without limitation to the scope of the claims. The applicant indicates that aspects of the present invention may consist of any such individual feature or combination of features. In view of the foregoing description it will be evident to a person skilled in the art that various modifications may be made within the scope of the invention.