Apparatus, process, and platform for activity analysis

Abstract

Embodiments can relate to devices and systems for generating activity data. The system can include a racket-vibration-dampening device. The racket-vibration-dampening device can include a housing, at least one processor, at least one transceiver, and one or more sensors disposed in the housing and configured to generate sensor data. The system can include a computing device in wireless communication with the racket-vibration-dampening device. The computing device can include at least one transceiver and at least one processor for receiving sensor data from the racket-vibration-dampening device, accessing a machine learning model stored in a non-transitory memory to analyze the sensor data, and generating, via the machine learning model, activity data based on the sensor data.

Description

Atty. Ref. No. 0073605-001025Apparatus, Process, and Platform for Activity AnalysisCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This patent application is related to and claims the benefit of priority of U.S. provisional patent application no. 63 / 688,633 filed on August 29, 2024, the entire contents of which are incorporated herein by reference.STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH AND DEVELOPMENT

[0002] This invention was made with government support under Grant No. CNS1909479 awarded by the National Science Foundation. The Government has certain rights in the invention.FIELD OF THE INVENTION

[0003] Embodiments can relate to systems and devices for providing vibration dampening of a striking implement and generating activity data during a user’s use of the striking implement. The devices and systems can include one or more sensors, processors, transceivers, and power supplies for measuring, processing, and / or transmitting inertial movement data of the striking implement and / or the user. The device can be communicatively connected with one or more processors that can employ one or more machine learning models to analyze the inertial movement data and generate activity data based on the inertial movement data that can provide detailed insight into the user’s use of the striking implement.BACKGROUND OF THE INVENTION

[0004] A large industry is growing around sports analytics, where the main goal is to sense, infer, and analyze various sport activities, such as strokes, player gestures, racket movements, player strategies, etc. For example, the iBall proposes a wireless sensor fusion system by utilizing an inertial measurement unit (IMU) and motion models to track a cricket ball’s 3D trajectory and spin by embedding radio sensors inside the ball. Moreover, multiple commercial products such as Zepp, Babolat POP, Head, Qlipp, Sony smart Tennis, and Courtmatics can trackAtty. Ref. No. 0073605-001025 a ball’s motion and a players’ behavior. Consequently, the global market for sport-related wearable devices is booming and expected to reach $106.47 billion by 2028, with an expected compound annual growth rate (CAGR) of 4.18% over the coming years.

[0005] More specifically, in various racket-based sports, such as tennis, badminton, table tennis, etc., comprehending or capturing the stroke dynamics of the racket during use, and particularly, the racket’s interaction with a ball, can be vital for analyzing and improving or enhancing player performance. For example, ball speed, ball spin, ball impact location, and stroke classification are crucial metrics for stroke evaluation in racket sports. Ball speed and spin allow the player to evaluate the strength and effectiveness of their strokes, while ball impact location provides feedback on the precision of a player’s shots. Furthermore, analysis of tennis stroke types enables players and coaches to work on diversifying their shot selection. Analytical data on players’ performance can pinpoint their specific areas of deficiency, enabling targeted improvements and expedited progress.

[0006] With prior vision processing techniques, using cameras such as PlaySight, SwingVision, and Hawk-eye technology can capture ball movement and player’s activity. However, cameras can be susceptible to interference from occlusions and lighting conditions and are very expensive and complex to set up. Therefore, most players do not have access to these systems. In addition, these systems lack the capability to capture finer grain information, such as impact location, as the human body tends to obstruct the camera view during tennis play. In contrast to cameras- based systems, motion sensors-based systems are more affordable, power efficient, easy to set up, and prone to bad lighting conditions, while being capable of capturing multi-dimensional information including ball speed, impact location, and stroke type. Lots of recent works in the field of ubiquitous computing focus on accessing player performance. However, these systems disrupt players’ natural grip and playing style due to bulky form factors caused by sub-optimal PCB design. In addition, they encounter resistance within the tennis community as an unusual tennis accessory while suffering from a dearth of stroke information due to the long distance between sensor placement and hitting area. On the other hand, there are some commercial products available on the market to evaluate the performance of players. However, these systems are closed without any details about the software or hardware and the are no APIs available to access the raw sensor data. In contrast to such works, we design an open sourceAtty. Ref. No. 0073605-001025 smart dampener platform that integrates the capability of evaluating ball speed, impact location, and stroke types altogether while sharing full details of hardware, software, and firmware. SUMMARY OF THE INVENTION

[0007] Embodiments can relate to a racket-vibration-dampening device and a system for generating analytical data. The racket-vibration-dampening device can include a housing that is removably attachable to one or more strings of a racket and can dampen vibrations transmitted via the strings of the racket. The device can detect and record user activity via one or more sensors and processors disposed within the housing, and can transmit the sensor data to one or more remote processors in wireless communication with the device for data generation (e.g., activity data), processing, and / or analysis. The system can include the racket dampening device and one or more machine learning models to receive the sensor data from the racket dampening device and generate activity data describing a user’s activity based on the sensor data.

[0008] An exemplary embodiment can relate to a racket-vibration-dampening device configured to generate analytical data. The racket-vibration-dampening device can include a housing. The housing can be configured for removable attachment to one or more strings of a racket, such that the racket-vibration-dampening device can be configured to dampen vibrations transmitted via the one or more strings of the racket. The racket-vibration-dampening device can include one or more sensors disposed within the housing and configured to generate sensor data. The racketvibration-dampening device can include one or more processors disposed within the housing and in communicative connection with the one or more sensors and configured to receive the sensor data from the one or more sensors. The racket-vibration-dampening device can include at least one transceiver disposed within the housing and in communicative connection with the one or more processors. The at least one transceiver can be configured to transmit the sensor data to at least one remote processor in wireless connection with the at least one transceiver. The racketvibration-dampening device can include at least one power supply disposed within the housing. The at least one power supply can be in electrical connection with, and can be configured to transmit electrical energy to, the one or more sensors, the one or more processors, and the at least one transceiver.

[0009] In some embodiments, the racket-vibration-dampening device can include a flexible printed circuit board fixedly disposed within the housing. In some embodiments, the one orAtty. Ref. No. 0073605-001025 more sensors, the one or more processors, and the at least one transceiver can be disposed on the flexible printed circuit board.

[0010] In some embodiments, the housing can include thermoplastic polyurethane.

[0011] In some embodiments, the one or more sensors can include at least one inertial measurement unit sensor.

[0012] In some embodiments, the at least one power supply can include at least one rechargeable battery.

[0013] In some embodiments, the racket-vibration-dampening device can be further configured to transmit the sensor data in real-time or near real-time, such as via the at least one transceiver.

[0014] Another exemplary embodiment can relate to a system for generating activity data. The system can include a racket-vibration-dampening device. The racket-vibration-dampening device can include a housing. The housing can be configured for removable attachment to one or more strings of a racket, such that the racket-vibration-dampening device can be configured to dampen vibrations transmitted via the one or more strings of the racket. The racket-vibration- dampening device can include one or more sensors disposed within the housing and configured to generate sensor data. The racket-vibration-dampening device can include at least one first processor disposed within the housing and in communicative connection with the one or more sensors and configured to receive the sensor data from the one or more sensors. The racketvibration-dampening device can include at least one first transceiver disposed within the housing and in communicative connection with the at least one first processor and configured to transmit the sensor data. The racket-vibration-dampening device can include at least one power supply disposed within the housing. The at least one power supply can be in electrical connection with, and can be configured to transmit electrical energy to, the one or more sensors, the at least one first processor, and the at least one first transceiver. The system can include at least one computing device in wireless communication with the racket-vibration-dampening device via the at least one first transceiver. The at least one computing device can include at least one second transceiver configured to receive the sensor data from the racket-vibration-dampening device via the at least one first transceiver. The at least one computing device can include at least one second processor in communicative connection with the at least one second transceiver. The at least one second processor can be configured to receive the sensor data from the at least oneAtty. Ref. No. 0073605-001025 second transceiver. The at least one second processor can be configured to access one or more machine learning models stored in a non-transitory memory. The one or more machine learning models can be configured to analyze the sensor data. The at least one second processor can be configured to input the sensor data to the one or more machine learning models. The at least one second processor can be configured to generate activity based on the sensor data via the one or more machine learning models.

[0015] In some embodiments, at least one of the one or more machine learning models can include a support vector regressor.

[0016] In some embodiments, the at least one computing device can include a mobile user device having at least one display.

[0017] In some embodiments, the at least one second processor can be further configured to display, via the at least one display, the activity data via one or more graphic user interfaces.

[0018] In some embodiments, the at least one second processor can be further configured to generate and / or display the activity data in real-time.

[0019] In some embodiments, the activity data can include at least one of ball speed data, impact location data, or stroke classification data.

[0020] In some embodiments, the stroke classification data can include a plurality of stroke type classifications.

[0021] In some embodiments, the plurality of stroke type classifications can include a serve stroke classification, a forehand groundstroke classification, a backhand groundstroke classification, a forehand volley classification, a backhand volley classification, and an overhead classification.

[0022] In some embodiments, the at least one second processor can be further configured to pre- process the sensor data input to the at least one machine learning model.

[0023] In some embodiments, the pre-processing of the sensor data can include data segmentation and cubic spine interpolation.

[0024] Further features, aspects, objects, advantages, and possible applications of the present invention will become apparent from a study of the exemplary embodiments and examples described below, in combination with the Figures, and the appended claims.BRIEF DESCRIPTION OF THE DRAWINGSAtty. Ref. No. 0073605-001025

[0025] The above and other objects, aspects, features, advantages and possible applications of the present innovation will be more apparent from the following more particular description thereof, presented in conjunction with the following drawings. Like reference numbers used in the drawings may identify like components.

[0026] FIG. 1 depicts a cross-section view of an exemplary racket-vibration-dampening device.

[0027] FIG. 2A depicts a real-world racket mounting configuration of an exemplary racket- vibration-dampening device.

[0028] FIG. 2B depicts a 3 -dimensional rendering of a racket mounting indication of an exemplary racket-vibration-dampening device.

[0029] FIG. 3 depicts a schematic diagram of a flexible printed circuit board and circuitry components of an exemplary racket-vibration-dampening device.

[0030] FIG. 4 depicts a perspective view of the exemplary racket-vibration-dampening device of FIG. 1.

[0031] FIG. 5 is a schematic block diagram of an exemplary system for generating activity data.

[0032] FIG. 6 depicts an illustrated overview of the exemplary system of FIG. 5 and accompanying workflow process.

[0033] FIG. 7 depicts a size / shape comparison of the exemplary racket-vibration-dampening device of FIG. 1 as compared to traditional tennis vibration dampeners.

[0034] FIG. 8 depicts a real-time demonstration of a graphical user interface displaying activity data generated by the exemplary system for generating activity data of FIG. 5.

[0035] FIG. 9A depicts a headband attachment configuration of the exemplary racket-vibrationdampening device of FIG. 1 for generating activity data.

[0036] FIG. 9B depicts a glove attachment configuration of the exemplary racket-vibrationdampening device of FIG. 1 for generating activity data.

[0037] FIG. 10 depicts a badminton racket handle attachment configuration of the exemplary racket -vibration-dampening device of FIG. 1 for generating activity data.

[0038] FIG. 11 shows a graphical representation of accelerometer, gyroscope, and magnetometer data generated by an IMU sensor during a tennis rally.

[0039] FIG. 12 shows a graphical representation of exemplary cubic spline interpolation of gyroscope data generated by an IMU sensor.Atty. Ref. No. 0073605-001025

[0040] FIG. 13 shows a graphical representation of gyroscopic x, y, and z data generated by an IMU sensor during a single tennis stroke.

[0041] FIG. 14A depicts an illustration of gyroscopic axis rotation based on ball impact location.

[0042] FIG. 14B depicts an illustration of gyroscopic axis rotation based on ball impact location.

[0043] FIG. 15 depicts an illustration of player motion of a Forehand Groundstroke and a Backhand Ground stroke.

[0044] FIG. 16A depicts an exemplary racket placement of the racket-dampening-device of FIG. 1 and an x, y, z, coordinate system for determining ball impact location.

[0045] FIG. 16B depicts a racket having a stencil mark created by an ink-soaked tennis ball during ball impact location calibration and testing.

[0046] FIG. 17A is a graphical representation of a median of error for user-dependent and userindependent ball speed estimation of a machine learning model as compared across 15 different player types.

[0047] FIG. 17B is a graphical representation of a median of error for ball speed estimation of a machine learning model as compared across 6 different stroke type classifications.

[0048] FIG. 17C is a graphical representation of a median of error for ball speed estimation of a machine learning model as compared across 3 different proficiency levels.

[0049] FIG. 18A is a graphical representation of a ball speed estimation accuracy percentage of a machine learning model as compared across 7 different ball speed ranges of outgoing ball speed.

[0050] FIG. 18B is a graphical representation of the cumulative distribution function (CDF) of ball speed estimation error probability of a machine learning model as compared to ball speed in miles-per-hour.

[0051] FIG. 19A is a graphical representation of a median error for user-dependent ball impact location detection of a machine learning model as compared across 15 different player types.

[0052] FIG. 19B is a graphical representation of a median error for user-independent ball impact location detection of a machine learning model as compared across 15 different player types.

[0053] FIG. 20A is a graphical representation of a median error of ball impact location detection of a machine learning model as compared across 3 different proficiency levels.Atty. Ref. No. 0073605-001025

[0054] FIG. 20B is a graphical representation of the cumulative distribution function (CDF) of ball impact location detection error probability of a machine learning model as compared to error distance in centimeters.

[0055] FIG. 21 depicts images of exemplary ball impact location results.

[0056] FIG. 22 depicts an illustration of ball impact location detection error distribution across different zones of a racket.

[0057] FIG. 23A is a graphical representation of user-dependent and domain adaptation stroke classification accuracy percentage of a machine learning model as compared across 15 different players.

[0058] FIG. 23B is a graphical representation of a stroke classification accuracy percentage of a machine learning model as compared across 7 different ball speed ranges of outgoing ball speed.

[0059] FIG. 23C is a graphical representation of a stroke classification accuracy percentage of a machine learning model as compared across 3 different proficiency levels.

[0060] FIG. 24 depicts a graphical representation of user experience survey results for the exemplary racket-dampening-device of FIG. 1 as compared with alternative sensing platforms.

[0061] FIG. 25 depicts a graphical representation of user experience survey results for the exemplary racket-dampening-device of FIG. 1.

[0062] FIG. 26A is a graphical representation of ball speed and stroke type classification accuracy of the exemplary system for generating activity data of FIG. 5 as compared to a number of sessions.

[0063] FIG. 26B is a graphical representation of a median of error of ball impact location detection of the exemplary system for generating activity data of FIG. 5 as compared to a number of sessions.

[0064] FIG. 27A is a graphical representation of ball speed, ball impact location detection, and stroke type classification accuracy of the exemplary system for generating activity data of FIG. 5 as compared to racket weight.

[0065] FIG. 27B is a graphical representation of ball speed and stroke type classification accuracy of the exemplary system for generating activity data of FIG. 5 as compared to a number of sessions.Atty. Ref. No. 0073605-001025

[0066] FIG. 28A is a graphical representation of ball speed and stroke type classification accuracy of the exemplary system for generating activity data of FIG. 5 as compared to an amount of continuous use time.

[0067] FIG. 28B is a graphical representation of a median of error of ball impact location detection of the exemplary system for generating activity data of FIG. 5 as compared to an amount of continuous use time.DETAILED DESCRIPTION OF THE INVENTION

[0068] The following description is of exemplary embodiments that are presently contemplated for carrying out the present invention. This description is not to be taken in a limiting sense, but is made merely for the purpose of describing the general principles and features of the present invention. The scope of the present invention is not limited by this description.

[0069] Referring to FIGS. 1-4, an exemplary embodiment can relate to a racket-vibrationdampening device 100 configured to generate analytical data. The racket- vibration-dampening device 100 can include a housing 102. In an exemplary embodiment, the housing can be a 3-D printed housing 102 made of Thermoplastic Polyurethane (TPU). The housing 102 can be made of any suitable material such that the housing 102 can provide sufficient strength and rigidity for the device 100, such that the device 100 can withstand repeated impacts (e.g., impacts from tennis balls), and such that the overall weight of the device 100 can be within a desired weight range (e.g., 5g -7g), such that the device 100 does not interfere with or otherwise burden a user’s typical use of the racket. For example, the housing 102 can be made of any suitable material, such as plastic, acrylic, PVC, TPU, TPE, silicon rubber, etc. The housing 102 can be made via any suitable manufacturing process, such as 3-D printing, additive manufacturing, molding, casting, etc.

[0070] The housing 102 can be configured for attachment to one or more racket strings 116. For example, the housing 102 can include one or more channels 114 configured to receive and securely attach to the one or more racket string 116. For example, the housing 102 can include a first channel 114a positioned on a first side of the housing 102 and a second channel 114b positioned opposite the first channel 114a, such that the device 100 can be inserted intermediate a first racket string 116a and a second racket string 116b. To attach the device 100 / housing 102 to the racket strings, a user can displace the first racket string 116a and second racket string 116bAtty. Ref. No. 0073605-001025 from their typical positions relative to the racket frame, that is, a user can manipulate the first string 116a and / or the second string 116b to increase the distance between them such that the housing 102 can be inserted intermediate the racket strings 116a / b. Upon insertion of the housing 102, the first and second racket strings 116a / b can secure the housing via friction between the racket strings 116 and the channel 114 as the racket strings 116 realign with / return to their original position prior to manipulation by the user. Friction and / or physical contact between the racket string 116 and the housing 102 can dampen vibrations transmitting along one or more racket strings 116 by dispersing the energy of the vibrations via a transfer of energy transmitted from the racket strings 116 to the housing 102 / device 100 via the contact / friction.

[0071] In some embodiments, the one or more channels 114 can be sized such that they can be configured to receive / attach to racket strings of varying thickness. For example, the width / diameter of the channels 114 can vary as the channel extends from an outer portion 118a of the housing 102 toward an inner portion 118b of the housing 102. That is, the width / diameter of the channel 114 at the outer portion 118a can be larger than the width / diameter of the channel at the inner portion 118b, such that positionally securing fiction between the channel 114 and the racket string 116 can be achieved for multiple thickness of racket strings 116. In some embodiments, the housing 102 can include a plurality of channels 114 such that the housing 102 can be attached intermediate two racket strings 116 oriented in a first direction (e.g., horizontally oriented relative to the racket frame) and / or can be attached intermediate two racket strings 116 oriented in a second direction different from the first direction (e.g., vertically oriented relative to the racket frame).

[0072] In some embodiments, the device 100 can include one or more mechanical means for attachment to one or more racket strings 116, such as one or more clips, fasteners, twist and lock mechanisms, etc. It should be understood that the above description is merely exemplary, and that the housing 102 can include any suitable materials, physical configurations, and / or means of attachment such that the device 100 can be securely attached to, and dampen vibrations of, one or more racket strings 116.

[0073] The racket-vibration-dampening device 100 can include at least one sensor 104 disposed within the housing 102 and configured to generate sensor data 201 (FIG. 6). In an exemplary embodiment, the sensor(s) 104 can include an inertial measurement unit (IMU). For example,Atty. Ref. No. 0073605-001025 the sensor(s) 104 can include an IMU having an accelerometer, a gyroscope, and / or a magnetometer, which can measure a combined effect of acceleration and gravity vectors, angular velocity, and the direction of a magnetic field, respectively. The sensor data 201 (FIG. 6) generated by the sensor(s) 104 can be used to determine an object’s movement, orientation, and / or changes of position (e.g., 9-axis IMU data). In some embodiments, the device 100 can include at least one IMU and one or more additional sensors, such as microphones, pressure sensors, force sensors, etc. It should be understood that the above is merely exemplary and that the device 100 can include any suitable sensor(s) for generating sensor data (e.g., sensor data 210, see FIG. 6) that can be used to determine an object’s movement, orientation, and / or changes in position.

[0074] In some embodiments, the racket-vibration-dampening device 100 can include at least one processor 106 disposed within the housing 102 and in communicative connection (e.g., wired and / or wireless) with the sensor(s) 104 and configured to receive data (e g., sensor data 201) from the sensor(s) 104 and / or to transmit data (e.g., via one or more transceivers) to one or more remote processors (e.g., a mobile device, smartphone, computing devices, servers, etc.). In an exemplary embodiment, the processor(s) 106 can include an ARM Cortex-M4 processor. As described below, the processor(s) 106 can include any suitable processors.

[0075] In some embodiments, the racket-vibration-dampening device 100 can include at least one transceiver 108 disposed within the housing 102 and in communicative connection (e.g., wired and / or wireless) with the processor(s) 106. The transceiver(s) 108 can be configured to transmit data (e.g., sensor data 201 received by the processor(s) 106 from the sensor(s) 104) to at least one remote computing device (e.g., a mobile device, remote computing device, server, etc.) in wireless communication with the at least one transceiver 108.

[0076] In some embodiments, the racket-vibration-dampening-device 100 can include at least one power supply 110 disposed within the housing 102. The power supply(ies) 110 can be in electrical connection with, and configured to transmit electrical energy to, the sensor(s) 104, the processor(s) 106, and / or the transceiver s) 108. For example, the power supply 110 can be a rechargeable battery 110, such as a coin cell battery. In some embodiments, the device 100 can include means for securely positioning the power supply (ies) 110 within the housing 102, such as a power supply (e.g., battery) retainer, as depicted in Fig. 3. It should be understood that theAtty. Ref. No. 0073605-001025 above description is merely exemplary, and that the device 100 can include any power supply suitable for powering the device 100 and / or the components thereof.

[0077] In some embodiments, one or more of the above-described components of the device 100 (e.g., sensor(s) 104, processor(s) 106, transceivers 108, etc.) disposed within the housing 102 can be disposed on a printed circuit board (PCB) 112 positioned within the housing 102. In an exemplary embodiment, the PCB 112 can include a flexible PCB (FPCB) 112. As depicted in FIG. 3, the FPCB 112 can be shaped in accordance with the shape of the housing 102. In some embodiments, the FPCB 112 can be a two-sided FPCB having a first side 112a and a second side 112b opposite the first side 112a. In some embodiments, the FPCB 112 can have a first portion 112c and a second portion 112d that are mirrored in size and shape, as depicted in FIG. 3, and the FPCB 112 can have a third portion 112e positioned intermediate the first portion 112c and the second portion 112d.

[0078] In some embodiments, one or more of the components of the device 100 (e.g., sensor(s) 104, processor(s) 106, transceiver(s) 108, etc.) can be disposed on the first portion 112c of the first side 112a of the FPCB 112 and the power supply(ies) 110 can be disposed on the second portion 112d of the first side 112a of the FPCB 112 such that each of the components and / or the power supply(s) 110 can be in electrical connection with each other via the FPCB 112. As depicted in FIGS. 2-4, in some embodiments, the third portion 112e can be sized and shaped such that the first portion 112c and the second portion 112d can be positioned within the housing in a folded configuration, such that the first portion 112c can be positioned to overlap and / or mirror the second portion 112d to reduce the overall size footprint of the components / FCPB relative to the housing, while maintaining an electrical connection between each of the components (e.g., sensor(s) 104, processor(s) 106, transceivers 108, etc.) and / or the power supply(ies) 110. It should be understood that the above described configuration is merely exemplary, and that the components (e.g., sensor(s) 104, processor(s) 106, transceivers 108, etc.), the PCB / FPCB 112, and / or the power supply(ies) 110 can be arranged within the housing in any suitable arrangement / configuration.

[0079] In some embodiments, one or more of the above described electrical components of the device 100 (e.g., sensor(s) 104, processor(s) 106, transceiver(s) 108, etc.) can be configured in a system-on-chip (SoC) configuration. For example, in some embodiments, the processor(s) 106,Atty. Ref. No. 0073605-001025 the transceiver s) 108, and / or the sensor(s) 104 can be disposed within a single chip disposed on the FPCB 112, such a microcontroller SoC. In some embodiments, the device 100 can include additional components related to the operation of the device 100, such as a multi -function switch / button (e.g., power button, wireless communication setup button, etc.) for controlling one or more operations of the device 100, visual indicators (e.g., LED lights, etc.) for indicating operational statuses or functions of the device 100 to a user, etc.

[0080] Referring now to FIGS. 5 and 6, an exemplary embodiment can relate to a system for generating activity data 200. The system 200 can include the above-described device 100, the details of which will not be repeated here in the interest of brevity, and at least one computing device 202 in wireless communication with the device 100, e.g., via the transceiver(s) 108. In some embodiments, the computing device(s) 202 can be a user mobile device, such as a laptop, tablet, mobile phone, smartphone, etc. The computing device(s) 202 can include at least one transceiver 208 configured to transmit and / or receive data, for example, to and / or from the transceiver(s) 108. The computing device(s) 202 can include at least one processor 206 in communicative connection (e.g., wired and / or wireless) with the device 100 and configured to transmit and / or receive data to and / or from the device 100, e.g., via the transceivers 108 / 208.

[0081] The processor(s) 206 can be configured to receive data, e.g., sensor data 201, from the device 100. The processor(s) 206 can be configured to access at least one machine learning model (MLM) 204 to process the received sensor data 201. In some embodiments, the MLM(s) 204 can be stored in at least one memory 210 (e.g., a non-transitory memory) in communicative connection with the processor(s) 206, such that the processor(s) 206 can access the MLM(s) 204. In some embodiments, the memory(ies) 210 can be a part of the computing device(s) 202. In some embodiments, the memory(ies) 210 can be a part of another computing device in communicative connection (e.g., wired or wireless) with the computing device(s) 202. The MLM(s) 204 can be configured to receive (e.g., as input) data from the processor(s) 206, such as the sensor data 201 generated by the device 100.

[0082] In some embodiments, the processor(s) 206 can be configured to pre-process the sensor data 201 before inputting the sensor data 201 to the MLM(s) 204, as described below. For example, the processor(s) 206 can be configured to access a set of instructions, algorithm, etc., from the memory 210 (e.g., non-transitory memory) that can cause the processor(s) 206 toAtty. Ref. No. 0073605-001025 segment and / or interpolate the sensor data 201. The sensor data 201 can be segmented into a plurality of segments, with each segment representing an individual stroke. For example, the sensor data 201 can be segmented and timestamped into individual strokes based on a predetermined magnitude threshold, where an individual stroke is determined based on one or more values of the sensor data 201 (e.g., gyroscopic y-axis magnitude) exceeding the predetermined threshold. In some embodiments, the sensor data 201 can be interpolated (e.g., by the processor(s) 206), such as via cubic spline interpolation to recover a saturation gap to a spike in the sensor data 201 (e.g., magnitude spike), to resample the sensor data 201 with an even time gap (e.g., based on timestamp data), and / or to up-sample the sensor data 201 to a higher sample rate.

[0083] The processor(s) 206 can be configured to generate activity data 214, via the MLM(s) 204, based on the sensor data 201, as described in detail below. For example, the MLM(s) can be configured to receive the pre-processed sensor data 201, process one or more segments of the segmented sensor data 201 (e g., individual strokes), and generate activity 214 for each individual segments, such that each individual stroke can have activity data 214 that describes one or more aspects of the stroke, e.g., ball impact location, stroke type, and / or ball speed. In some embodiments, the MLM(s) 204 can include any suitable MLM for processing the sensor data 201 and / or generating the activity data 214, such as linear regression, polynomial regression, neural network, and / or support vector regression MLMs. In an exemplary embodiment, the MLM(s) 204 can be a support vector regression (SVR) MLM 204.

[0084] In some embodiments, the activity data 214 can include a plurality of types of activity data related to a user’s use of a racket to which the device 100 is attached, for example, during a tennis match, practice, etc. For example, in some embodiments, the activity data 214 can include stroke type classification data, ball speed data, and ball impact location (e.g., impact point of a ball on a racket) data. In some embodiments, the stroke type classification data can include one or more of the following stroke types: Serve, Forehand Groundstroke, Backhand Groundstroke, Forehand Volley, Backhand Volley, and Overhead. In some embodiments, the stroke type classification data can include additional stroke types, such as drop shot, lob, half volley, etc. It should be understood that the above is merely exemplary, and that the activity data 214 can include any suitable data relating to the activity of a user during use of the device 100. ForAtty. Ref. No. 0073605-001025 example, in some embodiments, the activity data 214 can include data such as total stroke count, total time of a user’s use session, average ball speed, top ball speed, etc.

[0085] In some embodiments, the computing device 202 can include one or more displays 212. For example, in some embodiments, the computing device 202 can be a user mobile device (e.g., smartphone) including the display(s) 212. In some embodiments, the processor(s) 206 can be configured to display the activity data 214 on the display(s) 212. For example, the processor(s) 206 can be configured to access (e.g., from memory 210) and run a software application configured to generate an interactive graphical user interface (GUI) to display the activity data 214 such that a user can view and / or interact with the activity data 214. For example, as depicted in FIG. 8, the processor(s) 206 can display on the display(s) 212, e.g., via the software application running on the computing device 202, the activity data 214 in a graphic format that is easily viewed, comprehended, and / or interacted with by a user / analyst / etc.

[0086] In some embodiments, the system 200 can be configured to generate the activity data 214 in real-time or near real-time. For example, the device 100 can be configured to transmit the sensor data 201 in real-time or near real-time to the computing device 202, and the computing device 202 can be configured to receive and pre-process the sensor data 201 in real-time or near real-time, such that the computing device 202 can generate and display the activity data 214 in real-time or near real-time. In some embodiments, the system 200 can be configured to generate the activity data 214 asynchronously. For example, the device 100 can be configured to store the sensor data 201 in a memory (e.g., non-transitory memory) that is a part of the device 100 (e.g., a memory of the processor 106, a memory communicatively connected to the processor 106, etc.). In some embodiments, the system can be configured such that the device 100 can store the sensor data 201 in a memory and can transmit the sensor data to the computing device 202 based on the occurrence of a predetermined prompting event, such as a predetermined amount of time passing, the end of a use session, no activity being detected for a predetermined amount of time, etc., such that the activity data 214 can be generated at any point in time after the initial generation of the sensor data 201. For example, in some embodiments, the system 200 can be configured such that the device 100 can store the sensor data 201 in a memory (e.g., non- transitory memory) and can asynchronously transmit the sensor data 201 to the computing device 202 at a periodic time interval (e.g., every 30 seconds).Atty. Ref. No. 0073605-001025

[0087] The processor(s) 106 / 206 can be any of the processors disclosed herein. The processor(s) 106 / 206 can be part of or in communication with a machine (logic, one or more components, circuits (e.g., modules), or mechanisms). The processor(s) 106 / 206 can be hardware (e.g., processor, integrated circuit, central processing unit, microprocessor, core processor, computer device, etc.), firmware, software, etc. configured to perform operations by execution of instructions embodied in algorithms, data processing program logic, artificial intelligence programming, automated reasoning programming, etc. Use of processors 106 / 206 herein can include any one or combination of a Graphics Processing Unit (GPU), a Field Programmable Gate Array (FPGA), a Central Processing Unit (CPU), etc. The processor(s) 106 / 206 can include one or more sub-processors or processing modules. A sub-processor or processing module can be a software or firmware operating module configured to implement any of the method steps disclosed herein. The sub-processor or processing module can be embodied as software and stored in memory 210, the memory 210 being operatively associated with the processor(s) 106 / 206. A sub-processor or processing module can be embodied as a web application, a desktop application, a console application, etc.

[0088] The processor(s) 106 / 206 can include or be associated with a computer or machine readable medium. The computer or machine readable medium can include memory 210. The computer or machine readable medium can be configured to store one or more instructions thereon. The instructions can be in the form of algorithms, program logic, a model, etc. that cause the processor(s) 106 / 206 to perform any of the functions described herein.

[0089] Any of the memory 210 discussed herein can be computer readable memory configured to store data. The memory 210 can include a volatile or non-volatile, transitory or non-transitory memory, and be embodied as an in-memory, an active memory, a cloud memory, etc. Embodiments of the memory 210 can include a sub-processor or processor module and other circuitry to allow for the transfer of data to and from the memory 210, which can include to and from other components of a communication system. This transfer can be via hardwire or wireless transmission. The communication system can include transceivers, which can be used in combination with switches, receivers, transmitters, routers, gateways, wave-guides, etc. to facilitate communications via a communication approach or protocol for controlled and coordinated signal transmission and processing to any other component or combination ofAtty. Ref. No. 0073605-001025 components of the communication system. The transmission can be via a communication link. The communication link can be electronic-based, optical-based, opto-electronic-based, quantumbased, etc.

[0090] The processor(s) 106 / 206 can be in communication with other processors of other devices (e.g., a computer device, a desktop computer, a laptop computer, a computer system, etc.). Any of those other devices can include any of the exemplary processors disclosed herein. Any of the processors 106 / 206 can have transceivers or other communication devices / circuitry to facilitate transmission and reception of wireless signals. Any of the processors 106 / 206 can include an Application Programming Interface (API) as a software intermediary that allows two applications to talk to each other. For example, use of an API can allow software of the processor 106 of the racket-vibration-dampening device 100 to communicate with software of the processor 206 of the other device(s).

[0091] Any data transmission between the processor(s) 106 / 206 and memory 210, between the processor(s) 106 / 206 and a database, and between the processor(s) 106 and processor(s) 206 of other devices, etc. can be via a pull operation (e.g., the processor(s) 106 / 206 can pull the data) or a push operation (e.g., the data can be pushed to the processor(s) 106 / 206). The processor(s) 106 / 206 can receive the data in streaming format, or store it in memory 210 before being processed. In addition, embodiments of the algorithm, model, etc. disclosed herein can be developed as an application software (an “App”) to be implemented on a processor(s) 106 / 206 of a device. The App can be sent via a streaming format, or the App can be sent and stored on a memory associated with or accessed by the device.

[0092] As noted herein, the processor(s) 106 / 206 can be configured to be a component of, used in combination with, or in communication with another device / system - e.g., this can include the processor(s) 106 / 206 being part of the device / system, the device / system being part of the processor(s) 106 / 206, the processor(s) 106 / 206 in communication with the device / system, etc. “Being part of’ can include being on the same substrate or integrated circuit. For instance, the processor(s) 106 / 206 can be a component of, used in combination with, or in communication with a predictive modeling system, a decision support system, an automated control system, etc. processor(s) 106 / 206 can use a model (e g., machine learning model) or algorithm disclosedAtty. Ref. No. 0073605-001025 herein or provide the model or algorithm to the device / system to assist with or augment the performance of these devices / sy stems.

[0093] EXAMPLES

[0094] The following are exemplary compositions, devices, methods, and implementations of the embodiments disclosed herein. While the examples may focus on one implementation, it is understood that this is exemplary and the embodiments disclosed herein are not limited thereto. Additionally, it should be understood that the terms “smart dampening device,” “smart dampening system,” “smart dampening device / system,” and “system for generating activity data” refer to the above-described device 100 and system 200, and may be used interchangeably.

[0095] Introduction

[0096] Designing a smart dampening device can be challenging for a number of reasons: (i) The use of large electronic devices is categorically impractical due to their potential interference with player movements (ii) At professional or semi-professional levels, any physical change to the racket is almost unacceptable, thus necessitating any electronic sensors modality to be integrated into existing racket components or accessories, (iii) The device system must possess mechanical compatibility to allow for effortless integration into existing rackets and mechanical reliability to sustain high intensity of player activity, (iv) There is no state-of-the-art data collection method for collecting impact location, stroke types, and speed simultaneously, (v) The interaction between a tennis ball and tennis racket is characterized by a highly transient and complex dynamic process, which results in noisy or under-sampled sensor data, (vi) Players with different proficiency levels and dominant hands have diverse swing characteristics. Players may also use either side of the racket to counterattack an incoming ball. A smart dampening device must provide consistent accuracy across diverse variations.

[0097] Towards solving the above challenges, FIG. 2 depicts an overview of a smart dampening device and system for generating activity data. The smart dampening device can exploit a number of opportunities in hardware and multi-stage optimized algorithms, such as: (i) a custom micro-controller and sensing based on system-on-chip (SoC) architecture, which can decrease the form factor size while satisfying the requirements of embedded sensing and electronics, (ii) Considering that most tennis players attach vibration dampeners to their racket to reduce the vibration of racket strings, the smart dampening device can exploit flexible printed circuitAtty. Ref. No. 0073605-001025(FPCB) technology that can bend PCB within the form factor of a widely adopted dampener, as shown in FIG. 1. The deliberate selection of this specific form factor can allow professional or semi-professional players to seamlessly incorporate the smart dampening device, without encountering any adverse effects on their game performance. For example, the form factor of the smart dampening device (100) as compared to traditional commercially available racket dampeners (300) is depicted in FIG. 7. (iii) The smart dampening device can exploit the form factor design of existing conventional vibration dampeners and 3D-printed technology to customize an easy-to-use smart dampener. Its housing can be made of Thermoplastic Polyurethane (TPU) material to provide comparable dampening properties to a conventional dampener made of rubber or silicone, (iv) Leveraging on the data collection method of badminton’s impact point, a dataset can be collected to implement and evaluate the smart dampening device, (v) With signal processing and machine learning techniques applied, the system for generating activity data can have multi-stage algorithms to separate out and process each stroke event, then perform analysis on them accurately, (vi) An extensive user study encompassing diverse proficiency levels, genders, and playing habits was conducted to validate the robustness and precision of the smart dampening device and system.

[0098] An extensive user experience study was also conducted, in which the smart dampening device satisfied the users with high ratings across multiple dimensions such as acceptance, weight, and ease of use, demonstrating the high acceptability of such dampener-based form factors in the community. For validating sports analysis performance, a systematic study with diverse users achieves an accuracy of 3.59mph for ball speed estimation, a precision of 3.03cm for estimation of ball impact location, and an accuracy of 96.75% for stroke classification of 6. Furthermore, the performance is consistent over various players, sensor placement, and longer duration of the real-playing experiment. In the realm of tennis stroke analytics, the smart dampening device and system exhibits high precision and maintains solid robustness across a diverse range of users, player skill levels, and other variables.

[0099] We enumerate our contributions below: (i) To our best knowledge, embodiments disclosed herein relate to the first design of a sensor-embedded smart dampener by exploiting opportunities in a form factor design of existing real vibration dampener, 3D printing material, flexible PCB, and low-cost manufacturing to reliably allow continuous ball motion and player’sAtty. Ref. No. 0073605-001025 activity tracking for a long time while ensuring wide acceptance within tennis community, (ii) Demonstrated feasibility, precision, and stability of analyzing tennis strokes, ball speed, and ball impact location for sports analysis with SmartDampener in the shape of dampener and design of signal processing and analytic models, that identifies opportunities in correlations between racket motion, stroke characteristics, and player activity, (iii) Conducted a systematic user study across varied users to validate the community acceptance and comfort levels of SmartDampener as well as the performance of ball motion and player activity tracking with robustness to user diversity, changes in sensor placement, and ability to perform under real-play conditions, (iv) We will open-source SmartDampener for the community to explore it with novel capabilities in hardware and use cases. Although this discussion takes place primarily in the context of tennis, the core ideas and the platform can be generalized to other sports analytics use-cases as well.

[0100] RELATED WORK

[0101] This section relates to the review of previous work on assisting tennis with wearables, computer vision, and machine learning, as well as other sports analytics applications.Table 1. Summary of Related WorkAtty. Ref. No. 0073605-001025

[0102] Wearables

[0103] Table 1 summarizes a comparison between the smart dampening device and system and prior works in the field of wearable computing. Several sensors-based systems mount motion sensors on the wrist for tennis stroke detection and classification. Furthermore, some researchers exploit the interaction between the racket and the ball by mounting the sensor on the handle. TennisEye introduces an approach of placing an IMU sensor above the handle for stroke classification and tennis ball speed estimation. Pei et al. proposes to mount an IMU sensor on the handle to detect 6 tennis strokes. Nevertheless, such positionings of the sensors limit their sensing range in these systems because their placement on the wrist or handle is significantly distant from the area of impact. While prior works adopt certain electronic components for rapid prototype development, this results in a bulky form factor because such components can be suboptimal for overall device size when integrated together. Conversely, the smart dampener device embeds a motion sensor inside a dampener housing mounted on racket string, as depicted in FIG. 1, to exploit insight into the shots for stroke types, speed, and impact point. Moreover, the fusion of electronic components within a tennis dampener, a widely utilized accessory in tennis, enables the tracking of ball motion and player activity, meanwhile serving the dual function of mitigating string vibrations, akin to conventional dampeners. Among commercially available tennis sensors, Qlipp and Courtmatics are perhaps the closest to the smart dampening device / system. Nevertheless, the technical details are closed-source due to the proprietary nature of the hardware and there is no access to the raw sensor data for developers. To our best knowledge, the smart dampener device is the first open-source dampener-form factor sensors that fulfill all of the design requirements discussed in paragraph

[0040] ,

[0104] Computer Vision

[0105] In computer vision technology, single camera-based methods propose to track tennis ball trajectories from low-quality single-camera videos. Fei Yan et al. proposes a tennis ball tracking algorithm from a single camera by exploiting advances in modified particle filters. Calandre et al. introduces a system for 3D table tennis ball trajectory analysis using a single camera. Tracknet and Reno et al. track the position of a tennis ball from broadcast videos. However, only partial and limited ball position data can be extracted from single-camera videosAtty. Ref. No. 0073605-001025 or broadcast videos. To collect more data, recent researchers use multiple dimensional video data to analyze tennis ball motion, such as Wei et al., Zhao et al., and Cant et al. There have been several commercial products for multiple dimensional tennis ball tracking, such as SwingVision, Hawk-eye technology, PlaySight, and AccuTennis. However, the majority of tennis courts lack these systems as cameras are considered invasive to privacy and require optimal lighting, high resolution, intricate installation, and significant financial investment in hardware that often amounts to tens of thousands of dollars. In contrast, the smart dampening device can be ubiquitous and low-cost while being robust to ambient conditions.

[0106] Algorithms for Sport Analytics

[0107] A variety of algorithms have been applied for tennis analytics, including Hidden Markov Models, regression models, SVMs, Convolutional Neural Networks (CNNs), and Deep Learning Models. TennisEye proposes two models including a physical model and a regression model to estimate ball speed. TennisMaster applies Hidden Markov Model to segment a tennis serve into 8 phases. Recently, MonoTrack proposes a Recurrent neural network (RNN) for shot segmentation and 3D shot trajectory estimation for badminton without human intervention. Conversely, the smart dampener device and system for generating activity data integrates a differential shot segmentation technique, a machine learning algorithm for identifying the racket side, and three distinct support vector models dedicated to the final objective of tennis analytics. The smart dampener device and system for generating activity data can incorporate the design of various machine learning based algorithms and data processing techniques, forming a processing pipeline to achieve multiple tennis analytic tasks.

[0108] THE SMART DAMPENING DEVICE AND SYSTEM FOR GENERATING ACTIVITY DATA PLATFORM

[0109] In this section, the design of the smart dampening device and system for generating activity data for sports analytics in tennis are described. First, the design principles of the smart dampening device and system for generating activity data are elaborated one. Later, the form factor, hardware, and software are discussed.

[0110] Design Principles

[0111] One main goal of developing the smart dampening device and system for generating activity data is to design a general-purpose hardware sensing platform for the wearable researchAtty. Ref. No. 0073605-001025 community that allows for the discovery of sensing capabilities on dampener form factor devices. The following principles were used as guidelines when designing the smart dampening device and system for generating activity data.

[0112] Aesthetics and Community Acceptance

[0113] The smart dampening device should ensure an elegant appearance and pleasing aesthetics as a real tennis dampener, maintaining an appearance that the tennis community is familiar with and widely accepts. Existing commercial platforms such as Sony Smart Tennis, Zepp, Qlipp, Courtmatics, and Head have attracted a lot of attention. Gimenez-Egido et al. conducted an extensive study and validated that Zepp Tennis 2 improves the performance of players in tennis at low cost. These studies demonstrate that the integration of sensors within pre-existing tennis accessories familiar to tennis players offers significant utility and gamers community acceptance. Therefore, it is envisioned that the smart dampener device can be readily and rapidly acceptable to a broad range of players.

[0114] Long Battery Lifetime and Easy Charging

[0115] Energy constraints come along with inevitable form factor constraints since a tennis game may last for several hours; continuous sensing must be feasible during the entire session with a tiny battery. The smart dampening device can be designed with a low-power microcontroller and a rechargeable battery that can be utilized for a long time on a single charge. This allows players to continuously use the smart dampening device without the hassle of frequent battery charging.

[0116] Wireless Communication

[0117] The smart dampening device / system can be designed to wirelessly communicate with mobile devices, providing convenience of connection during gameplay for users. This wireless communication with minimal disruption maintains the common appearance of a real tennis dampener. The smart dampening device can be composed of a SoC integrated with BLE radio while maintaining low-power consumption and miniature form factor.

[0118] Coverage of a Wide Range of Applications

[0119] The smart dampening device / system can have various applications that provide valuable insights into user performance. For example, the smart dampening device / system can capture data on shot quality such as tennis strokes, ball speed, and ball impact location. While smartAtty. Ref. No. 0073605-001025 dampening device / system can provide insightful information to users for shot analysis to improve performance, it could also offer other applications like training aid, match analysis, and social interaction.

[0120] Openness and Extensibility

[0121] The research community should be able to easily extend the smart dampening device’s / system’s hardware and software platform. Given the small size of the smart dampening device, the device currently only integrates an IMU sensor, which can still cover a wide range of sports analytics. However, other sensors such as microphones, pressure sensors, and force sensors can be integrated depending on the design need of a use case. It is believed that opensourcing the smart dampening device / system will promote extensibility.

[0122] Reliability and Compatibility of Mechanical Design

[0123] The performance of the smart dampening device is designed to be robust in real play conditions, while maintaining in minimal interruption to the player. Because it is unavoidable that the ball will hit the housing of the smart dampening device in the context of a real-world scenario, the mechanical design of the smart dampening device should be reliable in real play conditions such as ball-hitting conditions. In addition, the smart dampening device is designed to have user-friendly functionality and broad compatibility in daily life. Players can seamlessly integrate the smart dampening device into using their existing racket without the need for any additional adjustments or equipment. This feature can allow players to continue using their preferred racket effortlessly. Players can easily mount or dismount the smart dampening device as commonly used tennis dampeners, enhancing the overall ease of use. It is particularly important to maintain compatibility with different racket models and string setups when designing the smart dampening device. This compatibility can ensure that a wide range of players can readily adopt the smart dampening device without requiring significant changes to their existing equipment.

[0124] Manufacturability and Cost-Effectiveness

[0125] To leverage the smart dampening device / system platform for the research and sports community, the device must be easily manufactured at an affordable cost. To achieve this, the smart dampening device can be designed with a composition of low-cost commercial off-the- shelf (COTS) electronic components and housing cases fabricated of 3D printing material.Atty. Ref. No. 0073605-001025

[0126] Platform Design

[0127] Form Factor Design

[0128] One goal is to design a lightweight, community-accepted, and unobtrusive smart dampener that can work as a real tennis dampener while being embedded with electronics and sensors that can sense ball motion and player activity and stream the data wirelessly for sports analytics. Therefore, the following opportunities can be exploited to maintain the form factor of a tennis dampener with embedded sensing and electronics while satisfying the functionality of commonly used dampeners, (i) The electronics can be designed using an FPCB that can bend, as illustrated in FIGS. 3-4. A custom microcontroller and sensing hardware can be assembled into the FPCB, the details of which are explained below. A replaceable and rechargeable coin cell battery (FIG. 3) can be integrated to power the hardware, (ii) The hardware can be enclosed inside the shape of a dampener, whose material can be designed to fit on the string of tennis rackets.

[0129] Hardware and Power Circuitry

[0130] Initially, a schematic design of a micro-controller module was developed to minimize form factor size while combining sensing devices and electronics. Integrated features can include (i) Sensing hardware, which can include an IMU sensor, (ii) A microcontroller (MCU) to collect data from the IMU sensor and stream wirelessly to a mobile / remote device running a companion app. (iii) A BLE module for communication with a wirelessly connected companion app. (iv) Battery circuitry to support diverse requirements of the above hardware components. Various hardware components were carefully assembled into a double-sided FPCB as shown in FIG. 3 to fit the form factor requirements. The MCU can consist of a 2.4 GHz radio frequency (RF) transceiver for BLE and ARM Cortex-M4 32-bit 64 MHz processor with floating-point unit (FPU), which can support multiple interfaces such as SPI, I2C, and UART. The IMU chip incorporated can be an ICM20948, which can provide 9-axis IMU data and can interface with the MCU using the SPI protocol. The whole PCB can be powered with a 70 mAh, 3.7V rechargeable lithium-ion coin battery, supplying 5.8 hours of data streaming. Overall, the power consumption of the hardware is about 12 mA when actively streaming the sensor data to the wirelessly connected companion app.

[0131] Mechanical DesignAtty. Ref. No. 0073605-001025

[0132] FIG. 1 depicts the mechanical design of the smart dampening device. To fulfill mechanical design requirements as discussed above, the opportunity of mechanical design and material of existing vibration dampeners can be exploited. Players can easily affix and disengage the smart dampening device as a real dampener. The smart dampener’s housing can be manufactured using fused deposition modeling (FDM) technology with TPU materials. While TPU allows the smart dampening device to sense string vibration and reduce the amount of vibration as a real dampener, the smart dampening device can be tightly mounted on string, as depicted in FIG. 1. In an exemplary embodiment, the dimensions of the smart dampening device can be 21.4 mm x 27.5 mm * 9.7 mm (W x L * H), which is much smaller than existing commercial products like Qlipp and Courtmatics. In such an embodiment, the smart dampening device can weigh 6.1g, and thus, can be comparable to commercially available dampeners weighing 5g.

[0133] Software Framework

[0134] The smart dampening device / system can be developed for use with iOS and Android using the React Native framework, which can be used for developing cross-platform mobile applications for iOS and Android platforms. The smart dampening device / system can communicate with the mobile / remote applications via BLE using a custom GATT profile with a sampling rate of 100 Hz. FIG. 8 depicts a real-time demo of an exemplary smart dampening system mobile application interface on a smartphone device implemented with ML modules for stroke classification, ball speed, and ball impact location. More details on data processing are discussed below.

[0135] Price Breakdown

[0136] The retail and wholesale unit price for exemplary major hardware components of the smart dampening device is summarized in Table 2. All firmware packages and programs developed for the smart dampening device / system can be made at no cost. The total cost to produce a single smart dampening device can be as low as $9.42, much lower than the price of existing commercialized smart dampeners like Qlipp and Courtmatics.Atty. Ref. No. 0073605-001025Electronic Component Unit Price (Retail) [U$] Unit Price (Wholesale) [U$]MCU (NRF52832) 5.02 2.23IMU Sensor (ICM20948) 26.45 5.62Chip Antenna 0.59 0.23Voltage Regulator ICs 1.47 0.80B attery Protecti on ICs 0.32 0.18Case 0.01 0.01Battery 1.57 0.35Total 35.43 9.42

[0137] Firmware and Application

[0138] The firmware of the smart dampening system / device can include two main components: (i) Collecting the data from the IMU sensors at the microcontroller; and (ii) Streaming data over BLE connection to a mobile / remote device. C++, Arduino, and BLE libraries can be used for implementing the functionalities. The firmware and the mobile / remote application can use popularly available Arduino and Android frameworks, and thus, can be easily extensible by developers to incorporate additional features.

[0139] USE CASE ANALYSIS

[0140] This section provides a brief overview of potential use cases of the smart dampening device / system.

[0141] Smart Sport Training

[0142] With the increasing involvement of people in sports activities like tennis, the need for coaching and training in sports activities is increased. Many players struggle to hire a professional sports trainer because of various challenges, including financial constraints. The use of wearable technology in daily sports activities is popular nowadays as it offers an affordable and convenient solution for players to improve their training performance in sports events. It is believed that the smart dampening device and system can help players improve training performance at an affordable cost while providing valuable information to users such as shot analysis, ball speed, sweet spot precision, stroke statistics, and impact point detection.

[0143] Mixed Reality in Sports

[0144] In the AR / VR environment, even though there is no physical impact, haptic feedback can potentially be integrated into the smart dampening device to provide a feeling of hitting aAtty. Ref. No. 0073605-001025 real ball to the user who is in an AR / VR environment. The sensing data collected from real plays using the smart dampening device / system can inform the actuators in the haptic system to provide similar feels for the user in an AR / VR environment. The raw IMU data from the smart dampening device can still be used for stroke analysis in AR / VR environment. In general, the application of virtual reality (VR) technology to sports has attracted intense interest recently. For example, VR can play the role of virtual trainer when playing racket sports for exercise and training. A combination of a Unity game engine and digital glove can enhance players’ physical feedback in the application of baseball training software in VR or AR environment. To enable the integration of VR into the smart dampening device / system, the smart dampening device can integrate haptic feedback mechanisms, such as a vibration motor and / or piezoelectric actuators, which can be seamlessly integrated into the smart dampening device to generate haptic feedback to users. For example, to emulate the sensation of a tennis ball striking a racket, a vibration motor could be strategically embedded within the smart dampening device to simulate the tactile feedback associated with such impact, even though the impact between the ball and the racket did not physically occur. With the combination of haptic technology and the smart dampening device and system, users can engage in peer-to-peer tennis matches that closely replicate the sensory experience of real-life gameplay. It is also envision that the smart dampening device can recognize the motion of users as an input device such as a smart glove, as depicted in FIG. 9B, for a number of VR applications such as education and training.

[0145] Injury Prevention

[0146] Users can often experience heavy vibrations and shocks during play, which can potentially cause discomfort or injury. The smart dampening device can capture vibrations at impact, warning the user of possible discomfort. The smart dampening device / system can exploit IMU data to detect player motions that may potentially result in injury, providing corrective suggestions. It is envisioned that the smart dampening device could be incorporated into a headband, enabling it to monitor potentially hazardous movements while the wearer participates in various sports activities. Such an integration can aim to alert users to potential injury risks during sports engagement.

[0147] Extensibility to Other SportsAtty. Ref. No. 0073605-001025

[0148] With careful consideration during the design process, the smart dampening device has a lightweight and a miniature form factor, as discussed above, that is capable of being adapted to other sports such as badminton, golf and table tennis, as depicted in FIG. 10. The smart dampening device can be easily mounted on instruments related to a variety of sports due to its adaptable design. Additionally, the smart dampening device can be attached to clothing and / or smart clothing, as depicted in FIG. 9A, Ito monitor player performance. It is envisioned the smart dampening device / system can support diverse applications in sports.

[0149] Sports Activity Visualization

[0150] The smart dampening device / system can provide users with intuitive and comprehensive insights into their performance via visualization techniques, allowing them to easily interpret and analyze their playing patterns. As users play, the smart dampening device / system collects data on various aspects of their performance, such as shot speed, spin rate, and impact location. After the session, the player can accesses the companion mobile / remote application, which presents visualizations of users’ performance data. For example, the companion mobile / remote application can show a heat map of the ball’s impact location on the racket, with the ball speed and stroke type, as shown in FIG. 8. By interpreting this visualization, users can identify patterns in their shot placement, such as favoring one side of the tennis racket or consistently hitting shots with high speed. With this insight, users can adjust their strategy and focus on areas of their game that may need to improve.

[0151] TENNIS STROKE ANALYTICS

[0152] Among the several potential use cases of the smart dampening device / system mentioned above, its capability of analyzing tennis player performance was validated by presenting the feasibility and robustness analysis of detecting multiple characteristics of tennis shots utilizing the IMU sensor integrated into smart dampening device. These characteristics can include stroke type, ball speed, and impact location. In this section, multiple stages of data processing and various machine learning models that were designed to approach the problem will be presented. An overview of an exemplary processing system is shown in FIG. 5.

[0153] IMU Data Pre-processing

[0154] As discussed above, the smart dampening device can be designed to be mounted on the racket at a location where tennis players usually place their vibration dampener. Thus, the IMUAtty. Ref. No. 0073605-001025 sensor in the smart dampening device can precept both the player’s movement and the ball’s impact on the racket. An IMU is composed of an accelerometer, a gyroscope, and a magnetometer. The accelerometer measures the combined effect of acceleration and gravity vector. The magnetometer measures the direction of the magnetic field. The gyroscope measures angular velocity. The accelerometer, magnetometer, and gyroscope sensors can be used to determine an object’s movement, orientation, or changes of position.

[0155] Stroke Segmentation

[0156] FIG. 11 shows a clip of IMU readings during a period of a tennis rally. Spikes that correspond to strokes and lower amplitude components that represent the player’s movement were observed. For further stroke analysis, it can be necessary to segment the IMU time series into individual strokes, as all the events can happen in quick succession. Using the segmented stroke IMU data, stroke type, ball speed, and impact location can be estimated. Gyroscope data can be used for segmentation because its variation is stronger based on empirical observations. In previous works, such as TennisEye, IMU readings are segmented by setting an absolute magnitude threshold to capture the spikes, i.e. the strokes timestamps are determined by whether at time t the gyroscope y’s magnitude GY t exceeded the defined threshold respect to 0.However, the occurrence of an impact is always after a fast back-swing. In this case, the impact spike may originate from non-zero magnitude and the actual absolute spike magnitude is higher. FIG. 13 depicts one such case. The above method may be unable to capture strokes or segment strokes with shifted offsets in similar cases. The temporal magnitude difference method, which uses a sliding window of 2-timestamps wide, was introduced to compare the magnitude at the current timestamp Gtt with the magnitude at the previous timestamp Gt -1 . A timestamp can be determined as the occurrence of a stroke if, on any gyroscope axis, the absolute value of Gt - Gt -1 is higher than a predefined threshold. The moving window can be applied on the x / y / z axis of the gyroscope to ensure impacts from all possible directions are considered. It was found empirically that 400 degrees / s is the best fit for this threshold to capture all spikes in the samples. Afterward, a 2000ms slice of IMU data can be preserved for each stroke, centered to the impact occurrence timestamp. Such a window size was selected since there are multiple stages in a complete player motion for a stroke, including back swing, acceleration, impact, and follow- through, as shown in FIG. 13. In most samples, back swing and acceleration together take atAtty. Ref. No. 0073605-001025 most 1000ms, and follow-through can finish within 1000ms in most cases. The accuracy of successful segmentation using the above method is discussed below.

[0157] Resampling and Interpolation

[0158] Even though a miniature IMU sensor has a wide sensing range, sensor saturation in instances of intense ball impact on the racket was still observed. FIG. 12 illustrates a snippet of raw gyroscope y data in which saturation occurred. It is assumed that such saturation may affect the model fitting of speed estimation and impact location detection. Meanwhile, to maximize the sampling and transmission rate, the smart dampening device / system hardware can be configured to transmit-on-sample, which will not wait until a certain interval to sample and transmit. Thus, the sample rate may vary with real-time connection rate. Eventually, the data transmitted to mobile / remote devices is not evenly sampled. But as the timestamp at each sample is preserved, in the resolution of the above issues, cubic spline interpolation can be chosen to (i) recover the saturation gap to a spike, (ii) resample the time series with an even time gap, and (iii) up-sample the time series to a higher sample rate. An example of cubic spline interpolation result is shown in FIG. 12. Before interpolation, the raw data sample can have a length of 200 and a sample rate of 100Hz. After interpolation, each sample can have a length of 1000 and a sample rate of 500Hz. Performance gain from cubic spline interpolation compared to linear 1-D interpolation is discussed below.

[0159] Data Input for Machine Learning

[0160] After segmentation and interpolation, data is prepared to feed to machine learning algorithms. The input data segments can contain 2000ms of racket motion captured by the IMU, including 1000ms prior to ball impact and 1000ms afterward. Such a span of sampling can preserve complete information of a stroke including player swing, follow-through, and ball impact. The segmented, interpolated IMU data can contain 6 channels of IMU reading including accelerometer x / y / z and gyroscope x / y / z, with a length of 1000 sample points and sample rate of 500 / s.

[0161] Ball Speed Estimation

[0162] To examine possible approaches to estimating tennis ball speed, the physical basis of the energy transfer between the racket and ball at impact was initially studied. Based on theAtty. Ref. No. 0073605-001025 conservation of momentum and coefficient of restitution, a simplified physical model to represent the relationship can be:.fc

[0163] In this system, E is the momentum of the racket before the impact, which is linearly proportional to the accelerometer reading that is perpendicular to the racket face. M and m are the mass of the racket and the ball, vr-init is the racket speed before impact and vr-end is the racket speed after impact, vb-out and vb-in are the incoming and outgoing speed of the ball. For M and m, it can be assumed that they are constants since the consistency of using the same type of racket and ball was maintained during the data collection. Another key equation is:

[0164] Where COR is the coefficient of restitution in racket and ball energy transfer. Using Eq. 1, vb-in can be substituted from Eq. 3 while introducing COR. COR can be affected by several factors, including impact location and string tension, etc. In this case, vb-out is the target value. The resulting representation of vb out is:

[0165] This approach did not fully attach to a physical model, where specific extrinsic factors like player mass and air drag, as well as intrinsic factors such as ball deformation and string extension, need to be considered for the model’s optimal result. Instead, data-driven machine learning was used for speed estimation. This decision is based on the following rationale: (i) Building a fully-fledged physical model with all factors considered can be computationally unfeasible for mobile and ubiquitous devices, (ii) The effort of quantifying factors that have minimal influence on results may not be proportional to the return of accuracy gain, (iii) AAtty. Ref. No. 0073605-001025 reasonable accuracy outcome can be still be reached without computing specifically the above factors, across a wide range of player mass, ambient environment, stroke styles, and racket types. The satisfied accuracy will be discussed below. The scope of this study did not include those factors, but it remains a prospect for future work.Table 3. Comparison of Machine Learning Algorithms for Speed Estimation

[0166] To design the model that best adapts to this task, multiple models were implemented and evaluated, including Linear Regression, Polynomial Regression (degree = 2), Support Vector Regressor (with ‘rbf kernel, penalty parameter C = 10), CNN (2 ConvlD layer, two Fully Connected layers, with BatchNormand ReLU activation between layers). In the initial approach, as it is assumed that the outgoing ball speed can either be the collective effect of accelerations on all axes or rely on the force applied in the Z-axis direction, a Convolutional Neural Networks (CNN) and Linear Regression were employed to predict the relationship between IMU readings and ball speed. However, despite extensive modifications to the structure and parameters, the CNN consistently exhibited overfitting issues. Simultaneously, Linear Regression failed to achieve the required level of precision. Consequently, alternative models were explored, specifically, Polynomial Regression and Support Vector Regression (SVR). It was hypothesized that the underlying relationship was too elementary for CNN, yet too complex for Linear Regression. As delineated in Table 3, a comparative analysis of these machine learning models revealed that SVR outperformed the others, achieving an average error rate of 4.81mph, which corresponds to 9.89% of the average sample speed. Within the SVR framework, various kernel functions were experimented with, including linear, polynomial, sigmoid, and RBF (Gaussian), and the RBF kernel was identified as the most effective in terms of overall performance. Results of detailed performance analysis of the SVM model are discussed below.

[0167] Impact Location DetectionAtty. Ref. No. 0073605-001025

[0168] Since the ball speed of strokes is able to estimated, the viability of detecting the impact location of the ball hitting the racket was also considered. Theoretically, different locations of ball impact should result in magnitude divergence on a particular axis of the IMU reading, e.g., if the impact location is at the far left side of the racket, a higher angular momentum is applied to the gyroscope y axis in the clockwise direction. Similarly, if the impact location is at the upper area of the racket, the rotation will appear on the gyroscope x axis in the clockwise direction. Such effects may combine if, e.g., the impact location may be at the top left, and both gyroscope x and Y axis will have variation. FIGS. 14A and 14B show illustrations of the above cases. Furthermore, the principle of action and reaction, as articulated in Newton’s Third Law, indicates that the reaction force from the ball when hit by a racket accelerated by the user’s swing can also cause such racket rotation, meaning both the impact of the ball and the motion of the player’s swing contribute to changes in racket angular velocity. However, the physical system of tennis ball impact also includes other artifacts such as hand force and inconsistent COR across the racket.

[0169] Constructing a comprehensive physical simulation surpasses the computational capabilities of mobile systems. Moreover, the current capability of motion can sensors fall short of accurately detecting finer factors. To simplify the problem-solving process, multiple machinelearning models were designed to find the best fit for impact location detection. Initially, it was endeavored to construct CNN models utilizing IMU data, which were processed via Short-time Fourier Transform (STFT). STFT was employed to leverage the feature dimension and information gain of IMU data. It is also assumed that unique impact location results in different angular and linear momentum variations for all axes, and thus, the complexity of a CNN will fit this problem. However, it became apparent that STFT overly complicated the dataset, generating higher-order information that was challenging for a CNN to interpret. Additionally, overfitting and inference clustering issues were encountered with the CNN despite various adjustments to its layers and parameters. This led to a pivot towards more straightforward machine learning models, including Linear Regression, Polynomial Regression, and Support Vector Regression. The overall accuracy for the experimented models is as shown in Table 4, including Linear Regression, Polynomial Regression (degree = 2), Support Vector Regression (with£rbf kernel, penalty parameter C = 10), Convolutional Neural Network (2 ConvlD layer, two Fully ConnetedAtty. Ref. No. 0073605-001025 layers, with BatchNormand ReLU activation between layers). The SVR model delivers the best accuracy, which can achieve a median diagonal error of 3.03cm. In the analysis, it became evident that the issue extends beyond a mere linear complexity, as user movements and ball impacts simultaneously influence several IMU axes. When compared with Support Vector Regression (SVR) employing the Radial Basis Function (RBF) kernel, Polynomial Regression demonstrated inferior adaptability to data characterized by imbalance and high variability. Consequently, the SVR model emerged as the more favorable model among those developed, owing to its superior adaptiveness and enhanced performance outcomes.

[0170] To achieve the prediction of impact location on the 2D surface of the racket face, combination of two SVR models was designed to predict for x axis and Y axis of the racket 2D face individually. To curtail the effect of COR variation, a method of data augmentation on ground truth at training time was proposed. The method can be formulated as:C ~ C X ( I F C n) (4) where C is the coordinate on x or Y axis, n is an amplification factor. The best accuracy gain was found when n = 1 for x axis and n = 2 for Y axis. These values correspond to the aspect ratio of the racket face. It was discovered that several players have the habit of spinning their racket during tennis rallies, after which they may hit the ball using either side of the racket face. This situation led to another problem: the smart dampening device / system needs to estimate the location of impacts that hit both sides of the racket. To resolve this issue, a classification model that distinguishes shots of the front side and the back side was implemented. The model is implemented with SVM and achieved an accuracy of 99.69% in 5-fold cross-validation and 99.52% in leave-one-subject-out validation. After the impact side is determined, the reading of the gyroscope x axis and accelerometer z axis can be inverted for a back impact. The accelerometer x axis can also be inverted, considering the possibility that the ball’s incoming direction is not perpendicular to the racket face, resulting in displacements along Y axis. All the above augmentation processes and models combine to be a pipeline for impact location detection. The performance result and evaluation of the impact location model are discussed below.Atty. Ref. No. 0073605-001025Table 4. Comparison of Machine Learning Algorithms for Impact Location

[0171] Stroke Classification

[0172] To approach the problem of stroke classification, the user’s motion when making different strokes was analyzed. It was observed that the player’s motion of racket swings is distinct among various stroke types. FIG. 15 is an example showing the difference in player action between Forehand Groundstrokes and Backhand Groundstrokes. It is believed that such divergence of motion can be used in the classification of strokes. Thus, machine learning methods can be chosen to approach this problem. Existing works and tennis sensor products usually categorize strokes into three types: Serve, Groundstroke, and Volley. The smart dampening device / system can categorize tennis strokes into 6 types: Serve, Forehand Groundstroke, Backhand Groundstroke, Forehand Volley, Backhand Volley, and Overhead. The 6 different classes were encoded into labels of 0 to 5. Various types of machine learning models were designed, using the feature map that pre-processed in 5.1 as input. To adapt the model to players with a left dominant hand, Forehand Strokes and Backhand Strokes were swapped, which were observed have opposite directions of motion compared to right dominant hand players.

[0173] As highlighted above, an investigation revealed distinct variations in player movements across different stroke types. Consequently, it can be imperative to utilize the entire time series of IMU readings to accurately identify the stroke type. This requirement dictates that the proposed model should effectively capture the temporal relationships between successive sensor data frames. Additionally, the model should be adaptable to the varying speeds at which strokes are executed, encompassing both rapid and slower stroke completions. To implement the stroke classification model, several models were designed employing machine learning techniques, which include: AdaBoost, Random Forest, SVM, CNN. Initially, the approach involved the use of CNN and Random Forest, the latter being recommended for this specific task in. However,Atty. Ref. No. 0073605-001025 findings indicated sub-optimal performance from the Random Forest model, likely attributable to variations in sensor placement compared to the referenced study. A CNN was experimented with because their complexity is sufficient to accommodate the temporal elasticity of a swing caused by differing swing speeds. While the CNN demonstrated promising precision, inconsistencies were observed in its performance, particularly when predicting for users that were not included in the training set. To enhance accuracy across a diverse range of players and stroke speeds, the model exploration was expanded to include other potential methods like Adaboost and SVM. Table. 5 demonstrates the following models: AdaBoost with Decision Tree base model, AdaBoost with Random Forest base model, Random Forest (max depth = 50), SVM (with ‘rbf kernel, penalty parameter C = 10), CNN (2 ConvlD layer, two Fully Connected layers, with BatchNorm, with Dropout = 0.8 and ReLU activation between layers), the SVM model exhibited superior accuracy in stroke classification, culminating in an impressive overall accuracy rate of 96.75%.

[0174] It was also proposed and validated that domain adaptation can further improve the accuracy of the model with a minimal amount of adaptation data for unseen users. The performance comparison of the models and detailed evaluation of SVM are discussed below.Table 5. Comparison of Machine Learning Algorithms for Stroke Classification

[0175] PERFORMANCE EVALUATION

[0176] User Study

[0177] Data Set

[0178] A study, approved by the IRD committee, was conducted with 15 players (10 males, and 5 females). The players are aged between 20-50 years and weigh between 50-90 kg. The users wore the sensor device as shown in FIGS. 16A and 16B, with the sensor snugly fitted on the tennis racket. Players were instructed to simply slide the smart dampening device onto the strings of the racket until it sat securely in the desired position, as shown in FIGS. 16A and 16B,Atty. Ref. No. 0073605-001025 akin to the traditional mounting method of a tennis dampener. The subjects were divided into three proficiency levels: beginner, intermediate and advanced. The beginner has played tennis for 0-2 years, intermediate players have played for 3-5 years and advanced players have played for more than 5 years. The players were then instructed to play with the tennis racket which was mounted with the smart dampening device in two user study sessions, one was for stroke type / ball speed and another one was for impact location. To incorporate all ranges of possible playing behavior, players were instructed to play naturally and, where possible, to perform stroke types including serves, forehand groundstrokes, backhand groundstrokes, forehand volleys, backhand volleys, and overheads. Because the ground truth of impact location is different from the ground truth of speed and stroke type, two different methods of data collection were conducted. The details of these two user study sessions are presented below.

[0179] Data Collection Methodology

[0180] Stroke Type and Ball Speed

[0181] Each pair of users took 3 breaks, with removing and remounting the sensor device in between, and the total study time per pair of users is two hours. While the smart dampening device / system provides IMU data for stroke type and ball speed, the SwingVision serves as ground truth for validation as well as provides labels of stroke type and ball speed for training the smart dampening system’s data analysis model. SwingVision, an advanced camera-based solution tailored for tennis, provides stroke type and speed, yet it lacks the capability to precisely determine impact location. Since there is no previous data collection methodology for tennis ball impact location, a novel approach was devised, inspired by the data collection methodologies utilized in a badminton impact location study. The data collection method of tennis impact location will be discussed below. For speed and stroke type, a total of 5526 data points were collected, including 1765 serves, 2626 forehand groundstrokes, 978 backhand groundstrokes, 87 forehand volleys, 55 backhand volleys, and 15 overheads.

[0182] Impact Location

[0183] Since there is no previous work for tennis ball impact location tracking using sensor- embedded dampeners, and SwingVision cannot provide impact location, a novel data collection approach was devised. Each player participated in 6 separate sessions with each session lasting for 30 minutes. Each pair of users took a break, with removing and remounting the sensorAtty. Ref. No. 0073605-001025 device between each session, and the total study time per pair of users is three hours. The tennis ball was sprayed with stencil ink prior to each trial. To facilitate this method, 50-lb white strings were installed on the rackets. After each trial, the tennis racket faced with an inked impact location was photographed as the ground truth of impact location, as shown in FIG. 16B. The impact location was then manually labeled with a utility that was designed with Python and OpenCV library. The racket face was cleaned up after each trial and balls were replaced every 50 trials to avoid the weight gain from ink build-up. For impact location, 3328 data points were collected in total.

[0184] Implementation

[0185] The smart dampening device / system can be implemented on a combination of desktop and mobile devices (e.g., smartphones). The CNN model can be implemented with PyTorch library and other ML models can be implemented with Scikit-Learn library, and the training can be implemented on a desktop, for example, a desktop with an Intel Core i7 9700K CPU, 48GB RAM, and a NVIDIA RTX 3080 GPU. Once trained, the inference can be performed entirely on mobile devices (e.g., Samsung S20 and OnePlus 9 Pro) using ONNX Runtime.

[0186] Performance Results

[0187] To evaluate the performance of the smart dampening device / system, analysis is conducted on the following aspects: (i) Segmentation and Interpolation; (ii) Stroke type classification; (iii) Ball speed estimation; and (iv) Impact location detection. In each part, userindependent analysis and multi-aspect comparison and evaluation is conducted. To evaluate the performance of the smart dampening device / system, a baseline comparison was provided with the performance of state-of-art work. For ball speed estimation and stroke type, the accuracy of TennisEye was chosen as the baseline. TennisEye is a state-of-art sensor-embedded solution of estimation speed and stroke type. The smart dampening device / system can achieve estimation of ball speed with a median error of 3.59 mph, in percentage is 9.89%, better than the accuracy of ball speed of TennisEye with an error of 5.6 mph, in percentage is 10.8%. The smart dampening device / system can achievs estimation of stroke type with an overall accuracy of 96.75%, better than the accuracy of stroke type of TennisEye with an overall accuracy of 96.2%. Since the smart dampening device / system is the first design of a sensor-embedded smart dampener capable of providing insights of impact location, there is no baseline for estimation of tennis impactAtty. Ref. No. 0073605-001025 location. One potential off-the-shelf competitor is the Zepp Tennis sensor. However, as of the time this research is conducted, the companion mobile application of Zepp is no longer accessible. Thus, it is unable to determine the performance of the Zepp Tennis sensor through either experiment or other reliable sources.

[0188] Stroke Segmentation and Interpolation

[0189] The stroke segmentation methods described above reached an average accuracy of 96.69% across all samples, meaning near-to-no strokes would be missed for analysis during tennis gameplay. The performance gain was calculated with the use of cubic spline interpolation, as compared to a linear 1-D interpolation with the same output length. The performance leverage of cubic spline interpolation on the stroke analysis is shown in Table 6.Table 6. Comparison of Final Accuracy with Cubic Spline Interpolation and 1-D Linear Interpolation

[0190] Ball Speed Calculation Accuracy

[0191] As previously presented in Table 3, the designed SVR model can achieve a median error of 3.59mph, in a percentage of 7.76%. Compared to similar works in tennis ball speed estimation, the physical model proposed in TennisEye achieved an overall error of 10.8%. FIG. 18B shows the CDF of ball speed estimation error. At the 50th percentile, the error is 3.5mph, at the 90th percentile the error is lOmph.

[0192] Accuracy vs. Users

[0193] A user study for ball speed estimation was conducted among 15 users. As shown in FIG. 17A, in the 5-fold cross-validation, the smart dampening device / system can contain the median error between 3.15mph and 4.10mph for all players. In the leave-one-subject-out cross- validation, players can still have median errors of less than 4.12mph in all cases, which means the speed estimation model of the smart dampening device / system can adapt to unseen players without perceptible loss of precision. FIG. 17C also showed that the smart dampeningAtty. Ref. No. 0073605-001025 device / system can achieve a stable median error of less than 3.61mph across all player proficiency levels.

[0194] Accuracy vs. Stroke Types

[0195] The ball speed estimation performance of the smart dampening device / system is presented under the following stroke types: Serve(SE), Forehand Groundstroke(FG), Backhand Groundstroke(BG), Forehand Volley(FV), Backhand Volley(BV) and Overhead(OH). The speed estimation models perform the best with Forehand Groundstroke and Backhand Groundstroke. For these strokes, the speed estimation achieved a median error of less than 3.29mph, as shown in Fig. 17B. For other stroke types, the error still resides between 4.48mph and 5.60mph. It is believed that, for Forehand Groundstroke and Backhand Groundstroke, the performance of the smart dampening device / system is better as a result of their higher population in the training sample. In tennis, Groundstrokes are the most prevalent strokes. The accuracy can be further improved with more training samples for particular stroke types.

[0196] Accuracy vs. Speed

[0197] FIG. 18A is presented to show that, in most speed ranges, the smart dampening device / system can achieve an accuracy of greater than 90%. Slightly lower accuracy was observed for samples in higher speed ranges. It is assumed the reason could be the lower population of high-speed samples in the training data set. Typically, only professional players can make consistent high-speed strokes. In most tennis rallies, the outgoing ball speed from a stroke is mostly seen between 30mph and 80 mph.

[0198] Ball Impact Location Detection Accuracy

[0199] Overview

[0200] As previously shown in Table 4, the best accuracy delivered by SVR achieved a median diagonal error of 3.30cm. This result is as small as the width of 3 sectors of string grid on common tennis rackets.

[0201] Front Back Classification

[0202] The classification model designed to determine the front / back side of impact achieved an accuracy of 99.69% in 5-fold cross validation and above 98.0%> for all players in leave-one- subject-out validation.

[0203] Accuracy vs. UsersAtty. Ref. No. 0073605-001025

[0204] As depicted in FIG. 19A, the smart dampening device / system can achieve stable accuracy across all players with minor variations. In the 5-fold cross-validation, the error of individual players can be as low as 3.04cm and the difference of error between users varies in a range of no more than 0.7cm. As demonstrated in FIG. 20A, the smart dampening device / system maintained consistent accuracy among players of differing expertise. Therefore, it is believed that the smart dampening device / system can estimate impact location across players with consistent accuracy with diversity in gender, body masses, sizes, player proficiency, etc. FIG. 20B shows the CDF of impact location detection error.

[0205] Qualitative Result

[0206] FIG. 21 depicts the qualitative results of impact location estimation. In FIG. 21 and all other evaluations in this section, a racket having a width of 26cm and a height of 35cm is used. The estimation of impact location is compared with the ground truth imprinted by stencil ink on the racket, as discussed above. FIG. 22 shows the distribution of impact location detection error across different zones on the racket. Evidently, the smart dampening device / system is capable of estimating impacts across a wide range of locations on the tennis racket with decent accuracy. It is believed that these results are promising in the context of a wide range of applications of sport analysis. Slight accuracy decay was also observed when the impact location is closer to the racket frame. It is believed that is because (i) COR varies from racket center to frame; and (ii) there are a relatively small amount of samples in which a ball impacts adjacent to the frame.

[0207] User Dependent vs. User Independent

[0208] Leave-one-subject-out validation was performed across the players in the samples and the performance results are shown in FIG. 19B. All 15 players’ median errors are contained between 2.60cm and 3.88cm. Over half of the users have errors of less than 3cm. In that case, it is believed that the smart dampening device / system can be capable of impact location estimation even on unseen subjects, delivering similar accuracy to user-dependent results even for new players.

[0209] Stroke Classification Accuracy

[0210] Overview

[0211] As previously shown in Table. 5, an overall accuracy of 96.75% was achieved with SVM. Compared to other works, similar stroke classification accuracy to TennisEye (96.2%)Atty. Ref. No. 0073605-001025 can be achieved. However, TennisEye only has 3 stroke types, as compared to the 6 stroke types of the smart dampening device / system.

[0212] Accuracy vs. Users

[0213] As shown in FIG. 23 A, the stroke classification implemented with the smart dampening device / system can achieve an accuracy between 99.25% and 90.0% across all 15 players in a 5- fold validation on the SVM model. The accuracy distribution among different player proficiency is presented in FIG. 23C, which displays above 95.03% accuracy for all player proficiency levels from beginners to professionals. It is believed that the smart dampening device / system can classify tennis strokes with a stable accuracy for players of different proficiency, age, gender, body weight, etc. Additionally, Player 5 of FIG. 23 A is a right-dominant-hand user. On this player, the smart dampening device / system showed equally high performance as it showed on other players, which means the smart dampening device / system can adapt to players with different dominant hands.

[0214] Accuracy vs. Speed

[0215] According to FIG. 23B, the smart dampening device / system shows a stable accuracy of above 93.85% for stroke classification across a wide range of speeds. This means the smart dampening device / system can achieve a consistently high accuracy under different game-play intensities to classify strokes.

[0216] Domain Adaptation

[0217] FIG. 23A also shows evaluation results of multi-user models with domain adaptation. User adaptation was implemented for the multi-user model of stroke classification with 30 userspecific data samples for each user, based on each model that is pre-trained in the leave-one- subject-out cross-validation. Based on Fig. 23 A, it is believed that, with an insignificant amount of user-specific data, the smart dampening device / system can achieve stroke classification accuracy that is similar to the user-dependent result for unseen players.

[0218] Performance

[0219] Usability Study

[0220] To assess the user experience of the smart dampening device / system, an extensive usability study was conducted. The study aimed to achieve the following objectives: (i) to conduct a comparative analysis of the smart dampening device / system against alternative sensingAtty. Ref. No. 0073605-001025 platforms, considering factors such as community acceptance, weight, and ease of use, (ii) to investigate the mounting experience of the smart dampening device / system compared with traditional tennis dampeners, (iii) to investigate how users were satisfied performance of the smart dampening device / system, and (iv) to explore users’ understanding and interpretation of stroke type, ball speed, and impact location facilitated by the smart dampening device / system. Users who participated in the usability study were asked to complete a questionnaire on their experience with the smart dampening device / system once the user study experiment was complete. The questionnaire was designed to solicit feedback on users’ experiences with the smart dampening device / system. The questionnaire encompassed inquiries aligned with the aforementioned objectives.

[0221] User Experience in Comparison with Existing Wearable Platforms

[0222] FIG. 24 depicts the results of user experience on the smart dampening device / system as compared with three alternative sensing platforms. To compare the user experience among these platforms, each player used all four platforms separately. The smart dampening system was compared with the other three platforms, including TennisEye, wrist-mounted sensor, and Zepp. It should be noted that only a dummy prototype of TennisEye was created based on the size and weight specification in the paper because the design details of the platform are not fully available to the public. Participants rated all devices anonymously based on community acceptance, weight, and easy to use from 0 to 10. The higher the rating, the better the usability of the device. The wrist-mounted sensor turned out to be rigid and not easy to use for a long time, while TennisEye and Zepp received a higher rating than the wrist-mounted sensor. However, with TennisEye and Zepp, the players noticed that these platforms modified either the weight or length of tennis rackets, thus making them a bit uncomfortable. The smart dampening device / system secured the highest scores in community acceptance, ease of use, and weight shown in FIG. 24. It is envisioned that the smart dampening device / system can gain ubiquitous acceptance within the tennis community.

[0223] User Experience on Mounting Instructions

[0224] To evaluate the user experience of mounting the smart dampening device, users were instructed to simply slide it onto the strings of the racket until it sits securely in the desired position akin to traditional installation methods observed with real tennis dampeners. Afterward,Atty. Ref. No. 0073605-001025 they were asked to compare the similarity of the mounting of the smart dampening device with the mounting of a traditional dampener. About 92% of the participants considered mounting of the smart dampening device similar to a traditional tennis dampener, whereas the remaining 8% users noted that the smart dampening device needs to be recharged periodically, thus making it slightly more effort to use it in comparison with traditional dampeners.

[0225] User Experience on Satisfaction of Tennis Metrics Determined by The Smart Dampening Device / System

[0226] To investigate how users were satisfied by the performance of the smart dampening device / system, participants (tennis players and tennis coaches) who participated in the study were asked to complete a questionnaire on their experience with the smart dampening device / system. The survey included questions on the users’ perspective of the smart dampening device’s / system’s performance regarding stroke types, ball speed, and impact location.Regarding the rating scale of the smart dampening device’s / sy stem’s performance, a rating scale from 1 to 10 corresponding with “Very Unsatisfied” to “Very Satisfied” was used. As shown in FIG. 25, more than 94%, 95%, and 92% of the users reported that they were satisfied with the performance of the smart dampening device / system regarding stroke type, ball speed, and impact location. Considering that the smart dampening device / system can be open-sourced to the community, it is anticipated that its accuracy could exhibit enhancement as a result of increased data availability and community contributions in the future. Overall, users were satisfied by the accuracy particularly given that the smart dampening device / system is low-cost and embeds sensors in a popular tennis accessory in the form of a dampener. The alternatives such as cameras either cannot track metrics such as impact location, or come with a high-cost barrier and installation / maintenance overheads.

[0227] Experience in Interpreting Data

[0228] To investigate how users can interpret stroke type, ball speed, and impact location, interviews were conducted with participants who have participated in the user study. After conducting interviews with tennis players regarding their interpretation of data from the smart dampening system, several key observations were made. Using stroke type, participants assessed their stroke selection and versatility on the court by understanding stroke type. For example, participants became more cognizant of their strengths and weaknesses with respect toAtty. Ref. No. 0073605-001025 stroke selection and could try to dedicate more time in training towards improving weaker parts of their game. Using ball speed, participants could understand how much power is behind their shots, which helps them to fine-tune their strength and control. For example, a shot with high speed has to be given sufficient topspin to keep it inside the court. Using the impact location, participants could evaluate the precision of each shot. Hitting the ball with the sweet spot on the racket is shown to improve the accuracy of shots, and users believed they can use the impact location feature of the smart dampening device / system to train with hitting on the sweet spot of the racket. Overall, participants agree that metrics such as stroke type, ball speed, and impact location can improve their game.

[0229] Robustness of Sensor Placement[00230J The variations in accuracy of stroke, ball speed, and impact location across different sessions are steady, as depicted in FIGS. 26A and 26B, with FIG. 26A depicting impact location accuracy and FIG. 26B depicting stroke and ball speed accuracy. In each session, the sensor device was removed and remounted across sessions thus helping evaluate any effects of changes in sensor position with respect to players’ habits when playing tennis with a real dampener. The accuracy is stable across various sessions because any minor variation in position across sessions is very small, compared with the hardware noise floor. Therefore, the impact of sensor placement on the accuracy of stroke, impact location, and ball speed is negligible.

[0231] Robustness to Ball Hit

[0232] Table. 7 depicts the performance of the smart dampening device / system under ballhitting conditions. To determine the reliability of the mechanical design of the smart dampening device under ball-hitting, studies were conducted with 15 players in real play scenarios. Users are instructed to allow a tennis ball to make contact with the sensor. The results are depicted in Table. 7. As observed, the accuracy of performance of the smart dampening device / system does not degrade under ball-hitting conditions. Therefore, the accuracy of stroke, impact location, and ball speed is robust under ball-hitting conditions.Table 7. Performance under Ball-hitting Conditions (Error, Mean)Atty. Ref. No. 0073605-001025

[0233] Robustness to Racket Diversity

[0234] To validate the impact of different weights on tennis rackets, a supplemental user study was conducted with various additional rackets. Rackets of three different weights (270g, 285g, 315g), and three different sizes (100in2, 103in2, 110in2) were used, which cover the majority of racket diversity in the real world. In the supplemental user study, the users are instructed to use the rackets as mentioned above in real play. The models are trained using the large dataset collected previously using a racket of medium weight and size. For evaluation, the models are tested against a newly gathered dataset that includes data from different types of rackets. The resulting accuracy of ball speed, stroke type, and impact location relative to racket weight is depicted in FIG. 27A, and the resulting accuracy of ball speed, stroke type, and impact location relative to racket size is depicted in FIG. 27B. Findings indicate that the performance of the smart dampening device / system on rackets with varying weights (light and heavy) and head sizes (small and large), without explicit domain knowledge, was comparable to its performance on rackets of medium weight and size, whose training and validation is upon same knowledge domain. It is believed that the consistency in performance can be attributed to the relatively stable differences in weight and size across various racket types. The weight and size of rackets can fluctuate by approximately 10% between different classes, and thus, the physical characteristics of rackets may not exhibit perceptible influence on the smart dampening device / system performance. In future works, it is also possible to further improve racket adaptability and overall accuracy by (i) open-sourcing the smart dampening device / system for the community to increase dataset amount and diversity and (ii) specifying racket parameters in models.

[0235] Lifespan

[0236] According to a statistic performed on the impact location dataset, the average frequency of the smart dampening device being hit by a ball is 1 ,2 / h. An average professional tennis player may play approximately 1300 hours a year. Accordingly, the approximate number of times per year that the smart dampening device will be hit by balls is 1560 (1.2 times hourly and for 1300 hours). A durability test was conducted on the system hardware by placing a tennis ball launcher against a fixed racket with the smart dampening device equipped. The system hardware is hit byAtty. Ref. No. 0073605-001025 machine-launched balls repeatedly for over 1500 times. Afterward, the system still functions as before the experiment, the nature of IMU readings, noise distribution, and vibration dampening was similar to that before. This indicates that the durability of the smart dampening device would be at least 12 months with respect to handling impacts from balls. The actual lifespan could be higher because no measurable degradation due to impacts was found. The electronic components in the smart dampening device are similar to other products, like Apple Airtag, that have high durability. The casing material of the smart dampening device (e.g., TPU) is also widely used in sports equipment such as shoe out-soles. Thus, it is believed that the smart dampening device / system is durable. In the future, since the smart dampening device / system platform will be open source, its durability and lifetime over a much longer duration (decades) can be further validated by the community with more experiments and data collected using the smart dampening device / system.

[0237] Power Consumption and Latency

[0238] The power consumption of the smart dampening device itself is discussed above. Here, the power consumption of executing analytic models in the smart dampening system on mobile devices (e.g., smartphones) is analyzed, using Batterystats and Battery Historian tools to profile the energy of the analytic models model. The latency of execution of all analytic models together on a Samsung S20 and a OnePlus 9 Pro are around 60ms for both devices, sufficient for real-time applications. The real-time power discharge rate of the smart dampening system implemented on the mobile device is 6.08% and 5.25% per hour for Samsung S20 and OnePlus 9 Pro, respectively, with a duty cycle of one execution every 2.5 seconds, simulating the interval of real tennis rally.

[0239] Longer Session Experiment

[0240] To study long-term effects, such as potential drifting, studies were conducted under real play conditions with 15 players for 2 hours for stroke types classification and speed estimation, and 3 hours for impact location. Players continuously used the tennis racket equipped with the smart dampening device. A 30-minute session was conducted, per the user study protocol described above. As depicted in FIGS. 28A (speed and stroke type accuracy vs. time) and 28B (ball impact location accuracy vs. time), the accuracy of the smart dampening device / systemAtty. Ref. No. 0073605-001025 does not degrade with time because the smart dampening system does not integrate long-term sensor data.

[0241] DISCUSSION AND FUTURE WORK

[0242] Extensibility to Incorporate Additional Sensors

[0243] The above discussion shows the feasibility of estimating ball’s speed, impact location, and sweet spot and stroke type via vibration captured by the IMU sensor embedded in the current design of the smart dampening device. However, it is believed that the potential of the design has only begun to be explored. The smart dampening device / system can incorporate additional sensors such as acoustic and camera for applications in spin tracking and slice angle estimation. While integrating all sensors in a single dampener might be limited by the current space and power consumption, these potentials are being explored and it is hoped that the smart dampening device / system can extend to new possibilities.

[0244] Adaptability to Various Sport Applications

[0245] While this description focuses on the sport of tennis, it is believed that the smart dampening device / system can be extensible to applications in various sports, such as field tracking and athlete analytics, due to the lightweight design and small size form factor. For example, attaching the smart dampening device to the clothes of athletes can enable the monitoring of the athlete’s physical load and fatigue. Such monitoring can protect athletes from injuries.

[0246] Human Body Pose Detection

[0247] The smart dampening device / system shows the feasibility of tracking players’ stroke types while estimating the ball’s characteristics. Based on these promising results, it is believed that the smart dampening device / system can be utilized in full-body motion tracking, such as through the measuring of motion segmentation and / or position information.

[0248] CONCLUSION

[0249] The smart dampening device / system shows the feasibility of estimating ball’s speed, impact location, sweet spot, and stroke type with a design of a smart dampener in a native form factor, low cost (~ $10), and long battery life. An IMU is embedded within the smart dampening device to enable a wide range of applications in sports analysis. A user experience study of the smart dampening device / system was conducted that indicated the acceptability and popularity ofAtty. Ref. No. 0073605-001025 the smart dampening device in the tennis community. To evaluate the sensing capabilities of the smart dampening device / system, an extensive study with 15 users provides an error of 3.59mph in speed, 3.03cm in impact location, and an accuracy of 96.75% in six stroke types recognition. Additionally, a number of applications in the area of sports training, mixed reality in sports, performance monitoring, injury prevention, and extensibility to other sports can be explored.

[0250] The following are exemplary compositions, devices, methods, and implementations of the embodiments disclosed herein. While the examples may focus on one implementation, it is understood that this is exemplary and the embodiments disclosed herein are not limited thereto.

[0251] References.

[0252] The following references are incorporated herein by reference in their entirety.[1] 2007. Nike+iPod, Apple, https: / / www.apple.com / ca / ipod / nike / run.html.[2] 2017. ICM20948 datasheet. https: / / invensense.tdk.com / wp-content / uploads / 2016 / 06 / DS- 000189-ICM-20948-V 1.3. pdf.[3] 2018. Tennis 101 : The 6 Basic Strokes Explained Step-by-Step - Pat Cash Tennis. https: / / www.patcash.co.uk / 2018 / 03 / the-6-basic-strokes-in-tennis-explained / .[4] 2019. Tennis racket specifications explained. https: / / tennishead.net / tennis-racket- specifications-explained / .[5] 2020. Lighter or Heavier - Which Tennis Racquet You Should Choose? https: / / www.racquets4u.com / blog / post / lighter-or-heavier-which-tennis-racquet-to- choose / https: / / www.racquets4u.com / blog / post / lighter-or-heavier-whi ch-tennis-racquet- to-choose / .[6] 2020. Multilayer piezoelectric actuators, https : / / content. kemet. com / datasheets / KEM_PO 101 _AE. pdf.[7] 2020. RTP Shocksorb Dampener, https: / / www.rtptennis.com / .[8] 2020. Vibrating Mini Motor Disc. https: / / www. adafruit. com / product / 1201 ?gad_source= 1 &gclid=Cj OKCQj wlZixBhCoARIs AIC745DCKALhxlOpW_cm2HDllNPaDuRlprupL9dGR19918N3xWbVwRpEzfoaAuO qEALw wcB.[9] 2021. AirTag Teardown: Yeah, This Tracks. https: / / www.ifixit.com / News / 50145 / airtag- teardown-part-one-yeah-this-tracks.Atty. Ref. No. 0073605-001025

[0010] 2022. NRF52832. https: / / www.nordicsemi.com / products / nrf52832.

[0011] 2022. Tennis Racquet Weight, Balance Swingweight Explained. https: / / tenniscompanion.org / tennis-racquet-weight-and-balance / .

[0012] 2023. AccuTennis. https: / / accutennis.com / .

[0013] 2023. Babolat. https: / / www.babolat.com / us.

[0014] 2023. Courtmatics. http: / / www.courtmatics.com / product.html.

[0015] 2023. Demo, https: / / streamable.com / co98oj.

[0016] 2023. Hawk eye innovations, https: / / www.hawkeyeinnovations.com / .

[0017] 2023. Head. https: / / www.head.com / en_US / sensor.

[0018] 2023. The Importance of Proper Tennis Form / Technique. https: / / www.tennismindgame.com / tennis-form.html.

[0019] 2023. Play Sight, https: / / playsight.com / .

[0020] 2023. Qlipp. https: / / www.eedesignit.com / the-tennis-sensor-thats-making-a-racket / .

[0021] 2023. Sony Smart Tennis Sensor Review, http: / / tennis-technology.com / sony-smart- tennis-sensor / .

[0022] 2023. Spivo Reviews, https: / / spivotennis.com / en-us / pages / reviews.

[0023] 2023. Swing Vision. https: / / swing. tennis / .

[0024] 2023. Wearable Devices in Sports Market Analysis. https: / / www.mordorintelligence.com / industry-reports / wearable-devices-in-sports-market.

[0025] 2023. Zepp. https: / / sensor-support.zepp.com / en / .

[0026] Adafruit. 2021. Bluefruit nRF52 Feather Learning Guide. https: / / learn.adafruit.com / bluefruit-nrf52-feather-learning-guide.

[0027] Adafruit. 2021. ICM20X. https: / / github.com / adafruit / Adafruit_ICM20X.

[0028] Akash Anand et al. 2017. Wearable motion sensor based analysis of swing sports. In IEEE ICMLA.

[0029] Android Developer 2021. Profile battery usage with Batterystats and Battery Historian. https: / / developer.android.com / topic / performance / power / setup-battery-historian.

[0030] Jacob S Arlotti et al. 2022. Benefits of IMU-based Wearables in Sports Medicine: Narrative Review. IJKSS (2022).Atty. Ref. No. 0073605-001025

[0031] Junjie Bai et al. 2019. ONNX: Open Neural Network Exchange. https: / / github.com / onnx / onnx.

[0032] G. Bradski. 2000. The OpenCV Library. Dr. Dobb 's Journal of Software Tools (2000).

[0033] Howard Brody. 1979. Physics of the tennis racket. American Journal ' of physics (1979).

[0034] Howard Brody. 1981. Physics of the tennis racket II: The “sweet spot”. American Journal of Physics (1981).

[0035] Brzostowski et al. 2018. Data fusion in ubiquitous sports training: Methodology and application. Wireless Communications and Mobile Computing (2018).

[0036] Lars Blithe et al. 2016. A wearable sensing system for timing analysis in tennis. In IEEE BSN.

[0037] Jordan Calandre et al. 2021. Extraction and analysis of 3D kinematic parameters of Table Tennis ball from a single camera. In ICPR.

[0038] Olivia Cant et al. 2020. Validation of ball spin estimates in tennis from multi-camera tracking data. Journal of Sports Sciences (2020).

[0039] Ciaran O Conaire et al. 2009. Tennissense: A platform for extracting semantic information from multi-camera tennis data. In DSP.

[0040] Rod Cross. 1997. The dead spot of a tennis racket. American Journa ' of Physics (1997).

[0041] Yu Ding et al. 2020. Application of Internet of Things and virtual reality technology in college physical education. IEEE Access (2020).

[0042] Jose Maria Gimenez-Egido et al. 2020. Using smart sensors to monitor physical activity and technical-tactical actions in junior tennis players. International journal of environmental research and public health (2020).

[0043] Mahanth Gowda et al. 2017. Bringing {IoT} to sports analytics. In NSDI.

[0044] Yu-Chuan Huang et al. 2019. Tracknet: A deep learning network for tracking high-speed and tiny objects in sports applications. In IEEE AVSS.

[0045] Alvin Jacob et al. 2016. Implementation of IMU sensor for elbow movement measurement of badminton players. In IEEE ROMA .

[0046] C D Johnson and M P McHugh. 2006. Performance demands of professional male tennis players. British Journal of Sports Medicine 40, 8 (2006), 696-699.Atty. Ref. No. 0073605-001025 https: / / doi.org / 10. 1136 / bjsm.2OO5.021253 arXiv:https: / / bjsni.brnj.com / content / 40 / 8 / 696. full. pdf.

[0047] Aida Kamisalic et al. 2018. Sensors and functionalities of non-invasive wrist- wearable devices: A review. Sensors (2018).

[0048] Marko Kos et al. 2016. Tennis stroke detection and classification using miniature wearable IMU device. In IWSSIP.

[0049] Johannes Landlinger et al. 2011. Differences in ball speed and accuracy of tennis groundstrokes between elite and high-performance players. European Journal of Sport Science (10 2011). https: / / doi.org / 10.1080 / 17461391.2011.566363.

[0050] SuKyoung Lee et al. 2017. Motion analysis in lower extremity joints during ski carving turns using wearable inertial sensors and plantar pressure sensors. In IEEE SMC.

[0051] Huimin Liu et al. 2020. Virtual reality racket sports: Virtual drills for exercise and training. In IEEE ISMAR.

[0052] Stuart A McErlain-Naylor et al. 2020. Effect of racket-shuttlecock impact location on shot outcome for badminton smashes by elite players. Journal of sports sciences (2020).

[0053] Miha Mlakar, et al. 2017. Analyzing tennis game through sensor data with machine learning and multi -objective optimization. In UbiComp.

[0054] Vinyes Mora et al. 2017. Deep learning for domain-specific action recognition in tennis. In CVPR workshops.

[0055] Borja Muniz-Pardos et al. 2018. Integration of wearable sensors into the evaluation of running economy and foot mechanics in elite runners. Current sports medicine reports (2018).

[0056] Ellen O’Reilly et al. 2001. ‘They Ought to Enjoy Physical Activity, You Know?’: Struggling with Fun in Physical Education. Sport, education and society (2001).

[0057] NEIL Owens et al. 2003. Hawk-eye tennis system. In 2003 international conference on visual information engineering VIE 2003.

[0058] Adam Paszke et al. 2019. PyTorch: An Imperative Style, High-Performance Deep Learning Library. In NeurlPS.

[0059] F. Pedregosa et al. 2011. Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research 12 (2011), 2825-2830.Atty. Ref. No. 0073605-001025

[0060] Weiping Pei et al. 2017. An embedded 6-axis sensor based recognition for tennis stroke. In IEEE ICCE.

[0061] Qazi et al. 2015. Automated ball tracking in tennis videos. In ICIIP.

[0062] Reno et al. 2018. Convolutional neural networks based ball detection in tennis games. In CVPR workshops.

[0063] Manish Sharma et al. 2017. Wearable motion sensor based phasic analysis of tennis serve for performance feedback. In IEEE ICASSP.

[0064] Yu-Tza Tsai et al. 2021. Unity game engine: Interactive software design using digital glove for virtual reality baseball pitch training. Microsystem Technologies (2021).

[0065] Guido Van Rossum and Fred L. Drake. 2009. Python 3 Reference Manual. CreateSpace, Scotts Valley, CA.

[0066] Zhelong Wang, Ming Guo, and Cong Zhao. 2016. Badminton stroke recognition based on body sensor networks. IEEE Transactions on Human-Machine Systems (2016).

[0067] Xinyu Wei et al. 2013. Predicting shot locations in tennis using spatiotemporal data. In IEEE DICTA.

[0068] Xinyu Wei et al. 2013. Sweet-spot: Using spatiotemporal data to discover and predict shots in tennis. In 7th Annual MIT Sloan Sports Analytics Conference, Boston, MA.

[0069] Graham J. Weir and Peter Norman McGavin. 2008. The coefficient of restitution for the idealized impact of a spherical, nano-scale particle on a rigid plane. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 464 (2008), 1295 - 1307. https: / / api.semanticscholar.Org / CorpusID: 122562612.

[0070] David Whiteside et al. 2017. Monitoring hitting load in tennis using inertial sensors and machine learning. IJSPP (2017).

[0071] Wikipedia contributors. 2024. Momentum — Wikipedia, The Free Encyclopedia. https: / / en. wikipedia. org / w / index.php?title=Momentum&oldid=l 197280395.

[0072] Fei Yan et al. 2005. A tennis ball tracking algorithm for automatic annotation of tennis match. In British machine vision conference.

[0073] Disheng Yang et al. 2017. TennisMaster: An EMU-based online serve performance evaluation system. In ACM AH.Atty. Ref. No. 0073605-001025

[0074] Hongyang Zhao et al. 2019. TennisEye: tennis ball speed estimation using a racketmounted motion sensor. In IPSN.

[0075] Hao Zhou et al. 2023. One Ring to Rule Them All: An Open Source Smartring Platform for Finger Motion Analytics and Healthcare Applications. IEEE / ACM loTDI (2023).

[0253] It should be understood that the disclosure of a range of values is a disclosure of every numerical value within that range, including the end points. It should also be appreciated that some components, features, and / or configurations may be described in connection with only one particular embodiment, but these same components, features, and / or configurations can be applied or used with many other embodiments and should be considered applicable to the other embodiments, unless stated otherwise or unless such a component, feature, and / or configuration is technically impossible to use with the other embodiment. Thus, the components, features, and / or configurations of the various embodiments can be combined together in any manner and such combinations are expressly contemplated and disclosed by this statement.

[0254] It will be apparent to those skilled in the art that numerous modifications and variations of the described examples and embodiments are possible considering the above teachings of the disclosure. The disclosed examples and embodiments are presented for purposes of illustration only. Other alternate embodiments may include some or all of the features disclosed herein. Therefore, it is the intent to cover all such modifications and alternate embodiments as may come within the true scope of this invention, which is to be given the full breadth thereof.

[0255] It should be understood that modifications to the embodiments disclosed herein can be made to meet a particular set of design criteria. Therefore, while certain exemplary embodiments of the compositions, materials, apparatuses, and methods of using and making the same disclosed herein have been discussed and illustrated, it is to be distinctly understood that the invention is not limited thereto but may be otherwise variously embodied and practiced within the scope of the following claims.

Claims

Atty. Ref. No. 0073605-001025WHAT IS CLAIMED IS:

1. A racket-vibration-dampening device configured to generate analytical data, the racket- vibration-dampening device comprising: a housing configured for removable attachment to one or more strings of a racket, such that the racket-vibration-dampening device is configured to dampen vibrations transmitted via the one or more strings of the ra ket; one or more sensors disposed within the housing and onfigured to generate sensor data; one or more processors disposed within the housing and in communicative connection with the one or more sensors, the one or more processors configured to receive the sensor data from the one or more sensors; at least one transceiver disposed within the housing and in communicative connection with the one or more processors, the at least one transceiver configured to transmit the sensor data to at least one remote processor in wireless communication with the at least one transceiver; and at least one power supply disposed within the housing, the at least one power supply in electrical connection with, and configured to transmit electrical energy to, the one or more sensors, the one or more processors, and the at least one transceiver.

2. The racket-vibration-dampening device of claim 1 further comprising a flexible printed circuit board fixedly disposed within the housing, wherein the one or more sensors, the one or more processors, and the at least one transceiver are disposed on the flexible printed circuit board.

3. The racket-vibration-dampening device of claim 1, wherein the housing comprises thermoplastic polyurethane.

4. The racket-vibration-dampening device of claim 1, wherein the one or more sensors comprises at least one inertial measurement unit sensor.Atty. Ref. No. 0073605-0010255. The racket- vibration-dampening device of claim 1, wherein the at least one power supply comprises at least one rechargeable battery.

6. The racket-vibration-dampening device of claim 1, wherein the racket- vibrationdampening device is further configured to transmit, via the at least one transceiver, the sensor data in real-time or near real-time.

7. A system for generating activity data, the system comprising: a racket-vibration-dampening device comprising: a housing configured for removable attachment to one or more strings of a racket, such that the racket-vibration-dampening device is configured to dampen vibrations transmitted via the one or more strings of the ra ket; one or more sensors disposed within the housing and configured to generate sensor data; at least one first processor disposed within the housing and in communicative connection with the one or more sensors, the at least one first processor configured to receive the sensor data from the one or more sensors; at least one first transceiver disposed within the housing and in communicative connection with the at least one first processor, the at least one transceiver configured to transmit the sensor data; and at least one power supply disposed within the housing, the at least one power supply in electrical connection with, and configured to transmit electrical energy to, the one or more sensors, the at least one first processor, and the at least one transceiver; at least one computing device in wireless communication with the racket-vibrationdampening device via the at least one first transceiver, wherein the at least one computing device comprises: at least one second transceiver configured to receive the sensor data from the racket -vibration-dampening devi e via the at least one first trans eiver; and at least one second processor in communicative connection with the at least one second transceiver and configured to:Atty. Ref. No. 0073605-001025 re eive the sensor data from the at least one se ond transceiver; access one or more machine learning models stored in a non-transitory memory, the one or more machine learning models Configured to analyze the sensor data; input, to the one or more machine learning models, the sensor data; and generate, via the one or more machine learning models, activity data based on the sensor data.

8. The system of claim 7, wherein at least one of the one or more machine learning models comprises a support vector regressor.

9. The system of claim 7, wherein the at least one computing device comprises a mobile user device having at least one display.

10. The system of claim 9, wherein the at least one second processor is further configured to display, via the at least one display, the activity data via one or more graphic user interfaces.

11. The system of claim 10, wherein the at least one second processor is further configured to generate and / or display the activity data in real-time.

12. The system of claim 7, wherein the activity data comprises at least one of ball speed data, impact location data, or stroke classification data.

13. The system of claim 12, wherein the stroke classification data comprises a plurality of stroke type classifications.

14. The system of claim 13, wherein the plurality of stroke type classifications comprises a serve stroke classification, a forehand groundstroke classification, a backhand groundstroke classification, a forehand volley classification, a backhand volley classification, and an overhead classification.Atty. Ref. No. 0073605-00102515. The system of claim 7, wherein the at least one second processor is further configured to pre-process the sensor data input to the at least one machine learning model.

16. The system of claim 15, wherein pre-processing the sensor data comprises data segmentation and cubic spine interpolation.

Images