GB2618776A

GB2618776A - An impact measurement system

Info

Publication number: GB2618776A
Application number: GB2206877.9A
Authority: GB
Inventors: Kimberley Michael; Wurr Tod
Original assignee: Cambridge Impact Tech Ltd
Current assignee: Cambridge Impact Tech Ltd
Priority date: 2022-05-11
Filing date: 2022-05-11
Publication date: 2023-11-22
Also published as: WO2023218184A1

Abstract

There is described an impact measurement system for use as a training aid adapted to provide metrics of strikes or other impacts made against a padded target. The system comprises an impact sensor and a processing system. The impact sensor, e.g. one or more accelerometers, is arranged to output impact signals in response to strikes made upon the padded target by a striking object, e.g. a fist or foot. The processing system receives signals from the sensors and filters them to distinguish impacts from other movements of the target. The system then executes predictive models trained to calculate an impulse value for the striking object, an impact duration and an impact speed and use these calculated values to calculate a force value for the impact. The models are initially trained using impacts from objects having a known momentum. The calculated metrics are then displayed to a user.

Description

An Impact Measurement System The invention relates to an impact measurement system, particularly but not exclusively as part of a sports training aid.

Padded targets are commonly used in the training of various contact sports such as martial arts, rugby and American football, to improve participants' fitness and technique.

It is desirous for training aids to provide feedback to the participant including metrics of strikes or other impacts made against the target. For example, the force and/or accuracy of a kick or punch.

The following are examples of prior art impact measuring systems intended for use in or for the training of contact sports: EP3488903 KR102221287, US9084924 US2003216228, US2012220430, US2017304140 and US2018001141.

Padded targets can be held in a variety of ways. For example, they may be mounted against a wall, suspended from a chain or rope or held by a training partner. The method used will often depend on the facilities available at the training centre but also the particular training exercise being undertaken at the time. Thus a participant may strike the same target held in a variety of different ways.

A problem with prior art systems is that the impact metrics they output differ depending on the way the target is supported. This makes it difficult for a participant to compare strikes on the target supported in one way, e.g. mounted on a wall, with strikes when the target is held in a different way, e.g. hand held. -2 -

The inventors have identified two reasons for this. The first is that the duration of an impact is dependent on the manner in which the padded target is held, which affects the force of the impact. This is why a strike on a hand-held target feels "softer" to a participant than a comparable strike on a wall mounted target.

The second reason is that impact readings from sensors are often complex and will differ markedly depending on the manner the target is held in ways that are attributable to the difference in force. This makes accurate calculation of metrics from readings difficult because although it is possible to accurately calibrate a device based on readings when the target is held in a specific way, e.g. held rigidly against a concrete wall, it is not possible to do this for every possible way the target can be held. For example, when hand held, the readings will depend on the size and weight of the person holding the padded target as well as their posture and the way they hold the padded target.

The present invention was conceived to provide a sports training aid that ameliorates the aforementioned problems According to a first aspect of the invention there is provided an impact measurement system comprising: a target; an impact sensor adapted to output signals indicative of a characteristic of an impact on the target by a striking object; a processing system adapted to receive output signals and configured to: using the output signals, execute a first predictive model trained to calculate an impulse value for the striking object consequent of the impact; and use the calculated impulse value determined to calculate a force value for the impact. -3 -

The invention can also be described in terms of a method and thus according to a further aspect of the invention there is provided a method for determining a force value of an impact by a striking object on a target, the method comprising: using an impact sensor to obtain signals indicative of a characteristic of the impact on the target by the striking object; executing a first predictive model trained to calculate an impulse value for the striking object consequent of the impact from the signals, and using the calculated impulse value to calculate a force value of the impact.

The following applies to either aspect of the invention.

The impulse (change in momentum) equates to the integral of a force of the impact over the time the impact occurs. It represents change in momentum of the striking object without reference to time and thus its value is independent of the duration of an impact and thus also the manner in which the target is supported.

Although in principle impulse is independent of the way of the target is supported, in practice the manner in which the target is held still has an effect on the outputs from the impact sensor.

Favourably, this effect is minimised by using a first predictive model, e.g. a first machine learning model, trained to minimise sensitivity to the duration of the impact. To do this, the first predictive model may be trained using sensor data collected from impacts of known impulse values on a test target of a structure similar or identical to the padded target. The impacts are performed on the test target supported with different degrees of rigidity. -4 -

In order that the impulse value for each impact is accurately known during the training phase, the target may be struck by a mechanical impact device with a striking head of a known weight. The speed of the striking head immediately before each impact with the test target may be determined using a speed detector device, e.g. using a pair of light gates that the striking head passes through before impact.

The training process comprises striking the test target with many different permutation of weight of striking head and speed of the striking head to provide sensor data for different impacts having a large range of impulse values The processing system may be configured to determine an impact duration value for the impact. The processing system may be configured to use the determined impact duration value with the determined impulse value to calculate a force value for the impact. For example, the impulse value may be divided by the impact duration value to provide an average force value for the impact because force = change in momentum over time The processing system may be adapted to: using the output signals, execute a second predictive model, e.g. a second machine learning model, trained to determine the impact duration value.

The second predictive model may be trained using at least in part the same training data collected in the training phase for training the first predictive model.

The processing system may be adapted to determine a normalised force value for the impact, the normalised force value corresponding to an expected force value for the impact had it been made against a first reference target system, which may be, for example, the target supported by a first reference mount. -5 -

The normalised force value allows for meaningful comparison of the force of a strike performed against a target held in one way with a strike performed against the same target held in a different way.

The reference target system has a known spring constant. Its value may be based on a real target system, e.g. a substantially identical target mounted on a concrete wall. It may correspond to a combination of target and mount that was used during the training phase for the first machine learning model. Alternatively, it may have a notional value. It may have a value equating to a target of a different construction. For example, the normalised force value may be used to provide an indication of a force value for a strike had it been performed upon a target that would not ordinarily be suitable to be struck by a user, for example an unpadded force plate or other unforgiving object such as a concrete block, because of the likelihood of causing injury to the user.

To calculate the normalised force value, the processing system may be adapted to divide the calculated impulse value by a normalised impact duration value for the impact (rather than the actual impact duration). The normalised impact duration value corresponds to an expected duration value for the impact had the strike occurred on the reference target system.

This method takes advantage of the fact that the impulse of the measured strike is independent of the duration of the impact, and thus of the object being hit.

This means the impulse value of the strike can be used to accurately calculate the normalised force value.

The system may be adapted to determine a second normalised force value for the impact, the second normalised force value corresponding to an expected force value for the impact had it occurred on a second reference target system having a different effective spring constant to the first reference target system, e.g. because it represents the target supported on a second reference mount of a -6 -different rigidity to the first reference mount and/or a target of a different construction.

One or more of the multiple normalised force values can be selected by, or otherwise presented to, the user to provide equivalent force values for the impact for different mount configurations or target constructions.

The impact system may be adapted to output the determined force value and/or, where applicable, normalised force value to the user e.g. via a user interface There are various methods by which the normalised duration value can be determined. A number of these may require the processing system being adapted to determine a speed of the striking object at a time that it impacted the target. For example, the processing system may be adapted to: using the output signals, execute a third predictive model, e.g. a third machine learning model, trained to determine the speed of a striking object at substantially the time that it impacted the target.

Because a speed model can be trained without the need to know the apparent mass of the striking object, the third model may be trained using sensor data collected from impacts performed by human rather than a calibration machine. This provides better speed predictions for human strikes.

Additionally, the processing system may be adapted to calculate an apparent mass of the striking object from the impulse value and the determined speed of the striking object.

The processing system may be adapted to use the determined speed of the striking object and the determined apparent mass of the striking object to determine the normalised duration value for the impact. For example, the processing system may be adapted to use the determined speed of the striking -7 -object and the determined apparent mass of the striking object as inputs to a mathematical model that provides, as an output, the normalised duration value for the impact.

In one example the processing system may use a look-up table or other array comprising data from, and/or derived from impacts on a target mounted on the first reference mount Additionally or alternatively, the processing system may solve a formula in which the determined speed of the striking object and the determined apparent mass of the striking object are variables.

In certain embodiments, the processing system may be configured to use a support indicator to determine the duration of the impact and/or the normalised duration of the impact. In the case of determining duration of the impact, the value of the support indicator is indicative of the manner in which the target was supported when the target was impacted. In the case of normalised duration, the value of the support indicator is indicative of a spring constant of a reference or notional target system.

For example, the processing system may hold a set of mathematical models that describes, for a given impact speed and apparent mass, an expected duration of an impact. Each model of the set being adapted for a different value of the support indicator. The processing system may be configured to use the support indicator to select the most appropriate mathematical model from the set of mathematical models and input the determined speed of the striking object arid the determined apparent mass of the striking object into the selected model to determine the impact duration value.

Additionally, the processing system may be configured to select a further mathematical model from the set of mathematical models and input the -8 -determined speed of the striking object and the determined apparent mass of the striking object into the further selected model to determine the normalised impact duration value The system may comprise a user interface (e.g. comprising a touch screen or switch) configured to receive the support indicator value from the user, which may correspond, for example, to whether the target is mounted onto a fixed structure, such as a wall, or hand-held.

Alternatively and/or additionally, the processing system may be configured to use the output signals from the impact sensor(s) to determine the manner in which the target is mounted.

The sensor output signals may be inputted to the first predictive model unmodified. Alternatively, the output signals may be processed by the processing system before being inputted. For example, the processing system may be configured to determine other variables from the output signals which are used as the input to the first predictive model.

Similarly, the output signals may be inputted to the second predictive model and/or third predictive model unmodified or modified.

The impact sensor may comprise an accelerometer. In many applications this is the preferred type of sensor because they are relatively small, cheap, rugged and can take the necessary number of samples per second needed to determine the characteristics of an impact, including duration, with the desired accuracy.

Favourably the impact sensor comprises a tri-axial accelerometer providing output signals indicative of the acceleration of the target in three orthogonal axes This allows rotational movement of the target to be determined during an -9 -impact, from which an indication of the accuracy and/or direction of a strike on the target can be gauged.

Where an accelerometer is used, the output signal may comprise an indication of the change in acceleration of the target.

The impact measuring system may output to the user, through the user interface, impact metrics of one or more impacts. This may be performed through one or more visual, aural or haptic means. For example, the user interface may comprise one or more of a display (e.g. including a touch panel) and a speaker.

The impact measuring system may be adapted to display to the user, via the user interface, one or more impact metrics comprising or derived from the following: the impact force value; the impact duration, the speed of the striking object at the moment of impact; the apparent mass of the striking object; the impulse value of the strike; the kinetic energy (Joules) arid power (watts) of the impact, and, where applicable, one or more normalised force values.

Favourably the impact measuring system is adapted to communicate to the user, via the user interface the force value and, where applicable, the first and/or second normalised force values.

The target may be a padded target comprising a resiliently compressible material to absorb at least a portion of the force of the impact on the target According to another aspect of the invention there is provided a sports training aid comprising the impact measurement system of any claim 1-16 as variously described above with the padded target.

In one embodiment the processing system is mounted within and/or against the target and the target comprises a transmitter to transmit (favourably wirelessly) strike metrics calculated by the processing system, e.g. via a wireless communication link, to the user interface, located remotely from the target.

Alternatively, the processing system could be remote from the target, optionally collocated with the user interface, and the transmitter may instead be adapted to receive the output signals from the sensor and transmit them, e.g. via a wireless communication link, to the processing system. This embodiment is less preferred because: it significantly increases the quantity of data that needs to be wirelessly transmitted, which may reduce the maximum strike cadence; the display unit would need to have the computing power to process the data, limiting the variety of devices that could perform the function of the user interface.

Irrespective of the implementation chosen, where the user interface includes a display, it is preferred that the display is separate from the target so that the display can be located in a position remote from the target that will avoid it being damaged by strikes to the target, and where it can conveniently relay strike metric information to the user or the user's trainer.

Additionally or alternatively, at least a part of the user interface may be implemented within or against the padded target. An example may be a beeper or a ruggedized simple LED display adapted to withstand the repeated mechanical shocks from impacts to the device.

The invention will now be described by way of example, with reference to the following figures in which: Figure IA is a schematic of a sports training aid that incorporates an impact measuring system;

-H -

Figure 1B is a perspective view schematic of a padded target incorporating an impact sensor and transmitter module of the impact measuring system; Figure 2 is a flow chart illustrating processes carried out by the processing system of the impact measuring system to determine strike metrics including force of impact and a normalised force of impact; and Figure 3 is a flow chart illustrating a variant process flow carried out by the processing system to determine strike metrics including force of impact and normalised force of impact.

With reference to Figs IA and 1B there is shown a sports training aid 1 The sporting training aid 1 comprises a padded target 2 and an impact measuring system 3 for accurately measuring human impacts such as punches, kicks, shoulder charges etc., made against the padded target 2 and which provides feedback to the user comprising strike metric data, to help the user improve their technique, power and/or fitness Referring to Fig 1B, the padded target 2 has a front-facing side 2A for receiving impacts from the user and a rear facing side 2B about which the padded target may be supported, e.g. by a human trainer or a wall. In variant embodiments, the padded target 2 may be supported about other sides, e.g top side, such when taking the form of a suspended punch bag.

The padded target 2 comprises a first board 21, a second board 22, a first resiliently compressible padding layer 23 and a second resiliently compressible padding layer 24.

The first and second boards 21, 22 are relatively rigid compared with the padding layers 23, 24.

The first padding layer 23 is mounted to a front facing side 21A (as viewed in Fig 1B) of the first board 21. The first padding layer is adapted to compress in response to an impact to soften the real force experience by the user to reduce injury. The second padding layer 24 is arranged between a rear facing side 21B of the first board 21 and a front facing side 22A of the second board 22 An example of a suitable material for the first board 21 is a resin-impregnated fibre board.

The second board 22 is not required to be made from a material with as great a flexural strength as the first board 21. As such the second board 22 may be constructed from cheaper materials such as plywood or similar.

An example suitable material for the padding layers 23, 24 is a solid foam synthetic plastics material, many examples of which are readily available and already commonly in use for the purpose.

The impact measuring system 3 comprises one or more impact sensors 31, e.g. tri-axial accelerometers; a processing system 32; a transmitter 33; and a user interface device 34, comprising a display 34A and receiver 34B.

The one or more impact sensors 31 are adapted to sense changes in acceleration of the padded target 2 and output signals indicative thereof to the processing system 32. The processing system 32 is adapted to receive and process the output signals from the sensors 31 to determine metrics for one or more impacts (and favourably each impact) inflicted upon the target 2. The output metrics are passed to the transmitter 33 that transmits them by a wireless communication protocol, to the user interface 34 via receiver 34B, which communicates the metrics to the user via the display 34A.

The one or more impact sensors 31 are mounted against (or optionally within) the first board 21. The processing system 32 and the transmitter 33 are favourably mounted against the second board 22 where they are protected from the force of impacts by the second padding layer 24. A wire 35 is provided between the (or where multiple, each) impact sensor 31 and the processing system 32 to carry the output signals from the (each) impact sensor 31 to the processing system 32. Where multiple impact sensors 31 are used, they may be connected to the processor through a daisy-chain link.

The processing system 32 comprises a processor 32A and a non-volatile computer readable store 32B communicatively coupled to the processor 32A, holding one or more computer programs for execution by the processor 32A to carry out the processing steps described below. The store 32B also holds mathematical models including machine learning models used in said processing steps.

Figure 2 is a flow chart illustrating the processing steps carried out by the processing system 2 when operating to calculate and communicate strike metrics to a user using acceleration samples from sensors 31 Acceleration samples from the sensors 31 are transmitted (100) to the processing system 32 The processing system 32 carries out a filtering process (101) on the received samples to identify a set of samples from the received samples that relate to an impact on the padded target 2 by a striking object (e.g. a user's foot, fist or shoulder through being punched, kicked, shoulder charged etc), as opposed to as a result of other movement of the padded target. This may be achieved in a variety of ways known to those skilled in the art of signal processing, including application of low and/or high pass filters and/or application of pattern recognition algorithms.

The processing system 32 executes (102) a first machine learning model (MLM). The first MLM uses the set of accelerometer samples identified at (101) as an input and from them calculates the impulse, i.e. change in momentum of the striking object that occurs as a consequence of the impact.

The first MLM is trained using sensor data collected from impacts for which the momentum of the striking object at the point of impact is known. The impacts are performed on a test target of a structure similar or identical to the padded target. Each impact brings the striking object to a stand-still and so reduces its momentum to zero. The change in momentum of the striking object corresponds to the impulse value During the training phase, the impacts are performed on the test target supported with different degrees of rigidity in order to make the first MLIVI resilient to the manner in which the target is mounted, in other words such that the first MLM can accurately predict the impulse of an impact regardless of the manner in which the target is supported In order that the impulse value for each impact is accurately known during the training phase, the target may be struck by a mechanical impact device with a striking head of a known weight. For example the impact device may take the form of a hammer head on a swinging arm. The speed of the striking head immediately before each impact with test target may be determined using a speed detector device, e.g. using a pair of light gates that the striking head passes through before impact To provide a large and varied dataset, each target with a different degree of rigidity, is stuck many times using many different permutations of weight of striking head and speed of the striking head to provide impacts having a large range of impulse values The processing system 32 executes (103) a second machine learning model (MLM), The second N4LN4 is adapted to use the samples identified at 101 as an input and from them calculate the duration of the impact. The duration of the impact being the time taken to effect the change in the momentum of the striking object as a consequence of the impact, i e which for a body part against a padded target, is the time taken to bring the body part to a complete stop.

The second MLM is trained using sensor data collected from the same impacts used to train the first MLM but in which each impact is labelled for impact duration rather than momentum. To obtain duration data for each collision during training, the time taken from initial contact with the target to reversal of direction of the striking object is measured for each strike. This may be done, for example, using a rotary encoder to measure the change in angle of a swinging hammer arm used to strike the target.

The processing system 32 executes (104) a third machine learning model (MLM) The third MINI is adapted to receive the set of samples as an input and from them determine the speed of the impacting object when it initially contacted the padded target 2.

The third MLM is trained using sensor data collected from impacts for which the speed of the striking object at the point of impact is known. The impacts are performed on a test target of a structure similar or identical to the padded target. Because the third MLNI can be trained without the need to know the apparent mass of the striking object, the third MLM is trained using impacts performed by one or more humans rather than by a calibration machine. This provides better speed predictions for human strikes.

The apparent mass of the striking object is calculated (105) by dividing the impulse calculated at (102) by the impact speed calculated at (104) (Apparent Mass = Impulse/ Impact Speed).

An average force value of the impact is calculated (106) by dividing the value of the impulse calculated at (102) by the value of the duration of the impact calculated at (103) The effective spring constant (i.e. stiffness) of the padded target 2, when impacted, and thus the duration of impacts upon it, is dependent upon the form of construction of the padded target 2, including the amount and type of padding material used, as well as the manner in which the padded target 2 is held, e.g. wall mounted or hand held.

For the same padded target 2 (ignoring any change through wear) any variation in spring constant is solely attributable to the manner in which the padded target 2 is held.

In addition, or alternatively to determining the force of the impact at (106), the processing system 32 may be configured to input the impact speed value calculated at (104) and the apparent mass value for the impact calculated at (105) into a mathematical model configured to determine, from these variables, a normalised collision duration for the impact (107). The normalised collision duration corresponds to the collision duration for a notional impact by an impacting object of the same speed and apparent mass as the measured impact, on a target 2 held by a reference support which together constitute a system with a known spring constant.

The mathematical model may comprise a look-up table or an array that comprises impact data derived from testing strikes carried out on a test target supported on the reference mount for training one or more of the first, second and third NTLIVIs. The look-up table comprises expected impact durations for combinations of speed and apparent mass of the striking object when using the reference mount The impulse determined at (102) is divided (108) by the normalised impact duration determined at (107) to determine a normalised impact force.

Because the normalised impact force is independent of the manner that the padded target 2 was actually held when the impact took place, its value for a specific impact can be meaningfully compared with an earlier or later impact that took place on the padded target held in a different manner.

This is because two strikes with identical speed and apparent mass of the striking object, the first performed on a wall-mounted target and the second on a hand-held target, have the same impulse values and the same normalised impact duration, and therefore the same normalised impact force values, notwithstanding that the actual force of the first strike would be greater -often to an extent noticeable to the user -because of its shorter actual impact duration than the second strike.

As mentioned above, the effective spring constant of the system depends upon the construction of the target. As such, it is also possible to provide, for a given impact, a normalised force corresponding to the force of that impact had it been made on a target of a different construction. For example, it could be used to provide an equivalent force for impacts against an unpadded force plate -a form of target that would be unsuitable to be used as a training aid due to the likelihood of injury to the user. Thus, it also allows meaningful comparison of impacts made against different targets with differing construction.

As an alternative to a look-up table, the mathematical model may comprise one or more formulae that relate impact duration in terms of impact speed and apparent mass, The determined actual force of the impact and/or the normalised force of the impact are displayed to the user via display 34A.

The user may select, via an input means (e.g. a touch screen, and/or one or more buttons etc) of the user interface 34, for display on display 34A, any one or more of the metrics mentioned above, namely: force of impact, normalised force of impact, speed, impact duration, normalised impact duration, apparent mass, impulse, kinetic energy and power.

Additional look-up tables may be provided to provide additional normalised force values for additional reference mounts or target constructions that represent targets with a variety of different effective spring constants Each of the additional look-up tables or arrays is compiled from impact data derived from testing strikes carried out on the respective reference mount or target construction.

Figure 3 describes a variant process flow. Samples from the accelerometers are transmitted to the processing system 32 (200). The processing system carries out a filtering process (201) to identify a set of samples from the received samples that relate to an impact on the padded target 2 by a striking object.

The processing system 32 executes (202) the first IVILM to calculate the impulse of the striking object that occurs as a consequence of the impact.

The processing system 32 executes (203) the third MLM to determine the speed of the impacting object, e.g. fist or foot, when it initially contacted the padded target 2.

As before, the apparent mass of the impacting object is calculated (204) by dividing the impulse calculated at (202) by the impact speed calculated at (203) The processing system 32 determines a firmness indicator which represents the firmness of the manner in which the padded target is held (205). This may be received as an input by the user via the user interface 34A (205A). [This will require the user interface to comprise means to receive input from the user interface, e.g. a touch panel, and the transmitter 33 and receiver 34B will need to be replaced with transceivers.] Alternatively, the processing system may derive the indicator by executing (205B) a fourth I\TLM adapted to receive the accelerometer samples as an input and to output the firmness indicator.

The fourth MILNI is trained from the same impacts used for the first MILM but the label to learn is changed from impulse to firmness indicator.

In one example the firmness indicator is a firmness score having a value between 1 and n, n equates to the number of degrees of resolution to which firmness is determined. As discussed above, the firmness of the hold will depend on whether the padded target 2 is fixed to a wall or handheld. In the case of the former it may also be affected by the way it is mounted and/or the size and construction of the wall, and for the latter, the size, weight & posture of the person holding the target.

The processing system 32 holds, in store 32B, a set of n number of mathematical models that relate impact duration to the apparent mass and speed of the impacting object. Each mathematical model of the set corresponds to a different firmness score between 1 and n.

The firmness indicator derived at (205) is used to select (206) the appropriate mathematical model from the set. The speed and apparent mass determined at (203) and (204) are inputted to the selected model to derive the actual duration of the impact (207). The derived duration value is then used to determine the actual force of the impact (208).

-20 -Alternatively, or additionally, the normalised force of the impact may be determined. To determine the normalised force of the impact, the speed and apparent mass of the impacting object determined at (203) & (204) are first inputted (209) to a model of the set that represents the firmness of the reference mount to determine the normalised impact duration. The impulse determined at (202) is then divided (210) by the calculated normalised impact duration to provide the normalised impact force.

As before, any of the calculated impact metrics can be selected to be displayed (211) via display 34A.

It will be appreciated that the order of certain steps may vary from that described in relation to Fig 2 and Fig 3 The padded target could incorporate impact sensors other than, or in addition to accelerometers, for example load cells The transmitter 33 and the receiver 34B within the user interface device 34 may be replaced by transceivers to allow two-way communication between the two.

Where so, the interface device 34 may, for example, indicate which strike metrics have been selected by the user. This information can be used by the processing system 32 to select or restrict the data sent via the link to the user interface device.

Examples of suitable models that can be used to implement one or more of the first, second, third and further MLMs of either embodiment are: Fully-Connected Artificial Neural Network; 1 or 2-dimensional Convolutional Neural Network; Long Short-Term Memory Model.

-21 -Other types of model may also be used Alternatively, one or more of the first, second and third MLN4s may be replaced by other, non-machine learning, predictive models such as, for example, linear or logistic regression.

Other data preparation steps may be performed to the raw signal data from the sensors before model execution, e.g. application of low, high or band-pass filters, or extraction of certain data features, e.g. peak and trough values.

Although less preferred, a wireless connection between transmitter 33 and receiver 34 may be replaced with a wired connection.

In a variant embodiment, the padded target may additionally include one or more impact sensors mounted elsewhere within the target, e.g. against or within the rear board, such as on the front face 22A of the rear board 22. One or more of the predictive models may be trained using outputs from all sensors, which may improve the ability to determine the manner in which the target is mounted and to predict impulse independently from the mariner in which the target is mounted.

Claims

-22 -Claims 1. An impact measurement system comprising: a target; an impact sensor adapted to output signals indicative of a characteristic of an impact on the target by a striking object, a processing system adapted to receive the output signals and configured to: using the output signals, execute a first predictive model trained to calculate an impulse value for the striking object consequent of the impact, and use the calculated impulse value to calculate a force value for the impact.
2 An impact measurement system according to claim 1 wherein the processing system is adapted to determine an impact duration value for the impact, and calculate the force value using the impact duration value and the calculated impulse value 3. An impact measurement system according to claim 2 wherein the processing system is adapted to: using the output signals, execute a second predictive model trained to determine the impact duration value.4 An impact measurement system according to claim 1, 2 or 3 wherein the processing system is adapted to determine a normalised force value for the impact, the normalised force value corresponding to an expected force value for the impact on a reference target system -23 -An impact measurement system according to claim 4 wherein the processing system is adapted to determine a normalised duration value for the impact, the normalised duration value corresponding to an expected duration value for the impact on a reference target system; and using the normalised duration value and the calculated impulse value to determine the normalised force value.6. An impact measurement system according to any claim 1-5 wherein the processing system is adapted to: using the output signals, execute a third predictive model trained to determine a speed of the striking object at substantially the time it impacted the target.7 An impact measurement system according to claim 6 wherein the processing system is adapted to calculate an apparent mass of the striking object from the impulse value and the determined speed of the striking object.8 An impact measurement system according to claim 5, 6 or 7 wherein the processing system is adapted to use the determined speed of the striking object and the determined apparent mass of the striking object as inputs to a mathematical model configured to provide, as an output, the normalised duration value for the impact.9 An impact measurement system according to any previous claim wherein the first predictive model was trained using output signals from multiple impacts on the target with known impulse values; the training comprising multiple impacts of the target supported in multiple ways having different mounts of different firmnesses and effective spring constants.10. An impact measurement system according to any previous claim wherein the second predictive model was trained using output signals from multiple impacts on the target with known speed values; the training comprising multiple impacts -24 -of the target supported in multiple ways having different effective spring constants 11. An impact measurement system according to any previous claim comprising a user interface adapted to communicate to a user one or more measured metrics associated with the detected impact.12. An impact measurement system according to any previous claim wherein the measured metrics comprise one or more of impact force and normalised impact force 13. An impact measurement system according to any previous claim in which the target comprises a resiliently compressible material to absorb at least a portion of the energy of the impact on the target.14. An impact measurement system according to any previous claim wherein the first predictive model comprises a first machine learning model.15. An impact measurement system according to any claim 3-14 wherein the second predictive model comprises a second machine learning model.16. An impact measurement system according to any claim 6-15 wherein the third predictive model comprises a third machine learning model.17. A sports training aid comprising the impact measurement system of any claim 13 -16.18. A method for determining a force value of an impact by a striking object on a target, the method comprising: -25 -using an impact sensor to obtain signals indicative of a characteristic of the impact on the target by the striking object; executing a first predictive model trained to calculate an impulse value for the striking object consequent of the impact from the signals; and using the calculated impulse value to calculate a force value of the impact.19. A method according to claim 18 comprising determining an impact duration value for the impact, and calculating the force value using the impact duration value and the calculated impulse value.20. A method according to claim 19 comprising executing a second predictive model trained to determine the impact duration value from the signals 21. A method according to claim 18, 19 or 20, comprising determining a normalised force value for the impact, the normalised force value corresponding to an expected force value for the impact on a reference target system.22 A method according to claim 21 comprising determining a normalised duration value for the impact, the normalised duration value corresponding to an expected duration value for the impact on a reference target system; and using the normalised duration value and the calculated impulse value to determine the normalised force value.23. A method according to any claim 18-22 comprising executing a third predictive model adapted to receive the signal as an input and trained to determine a speed of the striking object at substantially the time it impacted the target from the signals.-26 - 24. A method according to claim 23 comprising calculating an apparent mass of the striking object from the impulse value and the determined speed of the striking object.A method according to claim 22, 23, or 24 comprising using the determined speed of the striking object and the determined apparent mass of the striking object as inputs to a mathematical model configured to provide, as an output, the normalised duration value for the impact.