IL291425A - A wearable device for measuring accuracy in physical activity and a method therefor - Google Patents

Description

39991/IL/19-ORP- - A WEARABLE DEVICE FOR EXERCISE ACCURACY MEASUREMENT AND A METHOD THEREFOR Field of the InventionThe present invention relates to the field of training accessories. More particularly, the invention relates to a smart wearable device for exercise accuracy measurement and a method therefor.

Background of the InventionPhysical training is important for health, and it is important that the trainee receives feedback during the training on the exercises he performs, for example, how many calories he burns, what heart rate he reaches. While walking and running, it is important that the user knows his physiological parameters, such as heart rate, blood pressure, and more. Nowadays, there are conventional wearable devices that the user wears on the hands, on the legs, and the whole body (for example, a suit, a vest, a watch, elastic straps, a bracelet, etc.) and then the trainee (the user) can get indications of certain physiological parameters during training, for example, the body condition and the body response during training. This is important feedback because it enables the user to know if the heart rate is fast enough for achieving efficiency in the exercise but not too fast. Furthermore, it is important to get feedback to improve the training and prevent damage and overexertion of the trainee.

There are different types of exercises that need to be performed accurately. Inaccurate performance can cause inefficiency in performing the exercise and even damage or health problems, such as over-stretching of muscles, over-contraction of muscles, over-stretching of tendons, inaccurate or even incorrect weight lifting, etc. The conventional devices are unable to address this problem because they are not designed to measure and track movement paths and parameters of organ motion in the body, for example, speed, trajectories, forces, acceleration, etc., so the trainee cannot get accurate feedback on his performance and whether his organ movements while performing 39991/IL/19-ORP- - exercises and cannot know if the training is correct (correct hand movements, legs movements and more).

When performing training exercises, the correctness and precision of the movements that the trainee performs while doing a specific exercise are very important for preventing damage that can be caused by making incorrect movements. Furthermore, performing an exercise incorrectly, may result in the exercise not achieving the purpose for what it was intended and the trainee not benefiting from performing the exercise, which leads to unnecessary waste of the trainee's time.

It is, therefore, an object of the present invention to provide a system and method for measuring the accuracy of movement while performing exercises to guide the user to perform exercises correctly and precisely.

It is another object of the present invention to provide a system and method that provides the user feedback on inaccurate movements while performing exercises.

It is a further object of the present invention to provide a system and method for guiding the user on how to improve the accuracy of performed exercises.

Other objects and advantages of the invention will become apparent as the description proceeds.

Summary of the Invention A system for motion tracking and analysis during sports activities, comprising: a) a wearable device comprising a plurality of deployed motion sensors for generating transmitting motion raw data parameters, upon activation by an exercising user; b) a processing unit with memory, for storing and processing the raw data and process data according to instructions of an operating software; c) a battery for providing power to the sensors and the processing unit; d) a display device for displaying processing results to the exercising user; 39991/IL/19-ORP- - e) a data communication transceiver, for exchanging data with at least one remote database of the performance of the exercises that stores data collected from regular users, reference data recorded from professional athletes and common user mistakes, wherein the processor and the operating software are adapted to: f) generate an input vector from collected samples by processing and arranging the collected data in a structured manner by selecting only values of data points that describe the location and movement events features of each sensor and concatenating a series of movement events, to thereby capturing a movement that corresponds to an exercise; g) feed the input vector to a predictive model and apply Machine Learning models to classify the execution of an exercise as accurate or inaccurate, while using a different model for each kind of exercise; and h) display feedback to the user regarding the accuracy of exercises performed by the exercising user.

In some embodiments of the present invention, the motion sensors are selected from the group of: accelerometers, gyroscopes, and magnetometers.

In some embodiments of the present invention, a machine learning model is used to identify and analyze the source of the error in the execution of an exercise, while using techniques of Explainable Machine Learning to identify the source of the errors.

In some embodiments of the present invention, the wearable device further comprises Electromyography (EMG) sensors and heart rate sensors.

In some embodiments of the present invention, the wearable device further comprises a Heart Rate Monitor (HRM).

In some embodiments of the present invention, the wearable device further comprises multiple Inertial Measurement Units (IMUs) to collect data regarding movements in axes. 39991/IL/19-ORP- - In some embodiments of the present invention, a Binary Classification model is used to classify the exercise execution into correct or incorrect and an Explainable Machine Learning model, to indicate the reason that the exercise is incorrect.

A method for motion tracking and analysis during sports activities, comprising: a) deploying motion sensors at predetermined locations over the body of an exercising user; b) receiving real-time data in three axes in space from the motion sensors, including at least the angular position of each sensor in each axis relative to a rest pose; c) processing the received data and converting the received data to structured representation in the form of a vector containing a list of points in time to an argument that expresses a motion segment; d) comparing the features of the motion segment to features stored in a statistical database; e) determining by an AI model that receives the comparison results if the motion segment has been performed in a correct or incorrect exercise execution; f) using Machine Learning to analyze the motion segment and pointing on a specific location along the motion segment at which a mistake in the motion path occurred, while specifying the deviation from correct motion characteristics along the motion path; g) displaying to the user on a GUI, an indication if the exercise execution was correct or incorrect; and h) if the exercise execution was found incorrect, providing indication on the GUI what was the mistake and its location along the motion path. 39991/IL/19-ORP- - In some embodiments of the present invention, the deployment of the motion sensors is performed on a wearable mocap suit.

In some embodiments of the present invention, the deployment of the motion sensors is performed on a wearable device, selected from a group consisting of: a vest, a watch, elastic straps, a bracelet.

In some embodiments of the present invention, additional sensors are used in combination with the motion sensors for collecting physiological data from the body of the exercising user and processing the physiological data using Machine Learning to thereby improve the accuracy of the indications provided to the exercising user.

In some embodiments of the present invention, the indication if the exercise execution is correct or incorrect in real-time is provided in a visual form, audible form, tactile form, or any combination thereof.

In some embodiments of the present invention, one or more of the indications are provided on a relevant region on the wearable device, in particular in accordance with the location of incorrect activity.

Brief Description of the Drawings The above and other characteristics and advantages of the invention will be better understood through the following illustrative and non-limitative detailed description of preferred embodiments thereof, with reference to the appended drawings, wherein: - Fig. 1 is a flowchart of the process of generating the input vector from samples collected from the vest of an exercising person; - Fig. 2 is a flowchart of the execution using the Exercise Classification Binary Model and the Exercise Explanation Model; - Fig. 3 is a flow chart that describes the data collecting and analysis, according to an embodiment of the present invention; - Figs. 4-5 illustrate two possible deployments of sensors on a Shadow® motion-capture (mocap) suit; 39991/IL/19-ORP- - - Fig. 6 schematically illustrates an architecture diagram of a system for motion tracking and analysis during sports activities, according to an embodiment of the invention; and - Fig. 7 schematically illustrates an Entity Relationship Diagram (ERD) of the system of Fig. 6, according to an embodiment of the invention.

Detailed Description of the Invention The present invention provides a system in the form of a smart wearable device, for motion tracking analysis during sports activities. This smart wearable device may be any one of the following forms: a suit, a vest, a watch, elastic straps, or a bracelet. According to an embodiment of the invention, the system of the present invention combines various metrics such as electrical activity inside the muscle and physiological measurements such as heartbeat. This combination produces a system that completely solves the need for physical activity analysis. In other words, a system that "knows" how to tailor a training program to the user, hermetically, in real-time, and accordingly provides performance feedback to the user, as will be described in further details hereinbelow.

The main goal of the smart wearable device of the present invention is to be a personal trainer, i.e., to teach the user how to perform training exercises correctly. The method of the present invention applies Machine Learning models (Deep Learning) to the data generated by the smart wearable device to classify the execution of the fitness exercise. The system collects a large amount of data from many users, some of them performing the exercise correctly and others not. The system uses the collected data to train deep learning models that perform a binary classification, by using a different model for each exercise.

The present invention also provides a machine learning model to identify and analyze the source of the error in the execution of the fitness exercise, while using techniques of Explainable Machine Learning (a set of tools and frameworks to help to understand and 39991/IL/19-ORP- - interpret predictions made by the machine learning models), to identify the source of the errors.

According to the method proposed by the present invention, the collected data is arranged and stored in a structured manner, such as a vector of values. The vector containing the data generated by the smart wearable device has several values (value points) for each sample of the sensors. These values are used to train the machine learning models.

According to an embodiment of the invention, the smart wearable device comprises three types of sensors: motion sensors, Electromyography (EMG) sensors, and heart rate sensors.

Motion Sensors: The motion sensors provide the following data about the position and the orientation of different segments in the trainee's body. These motion sensors include: ● Accelerometers are a type of sensors that provide information about the linear acceleration of the sensor in 3 axes in space, mainly to know the pace of the movement and are used to determine the position in space.

● Gyroscopes are a type of sensor that provides data about the angular acceleration of the sensor. The data provided by gyroscopes can be derived to obtain angular velocity and angular position (space orientation) to calculate the moment applied on a joint or a segment.

● Magnetometers are a type of sensors that stabilize the system and assist in processing the raw data collected from the previous components. For example, a magnetometer that can measure compass direction allows the device to know its orientation relative to the magnetic north, similar to how a hand-held compass works. The integrated 16-bit ADCs simultaneously sample the three-axis of movements (X, Y, Z). The system gets a common reference point for all the sensors by measuring the magnetic compass. 39991/IL/19-ORP- - Electromyography (EMG) Sensors: The EMG sensors indicate the strain over time in a specific muscle EMG Sensor, by measuring electrical signals generated by the muscles when moving them. Electromyography detects the electric potential generated by muscle cells when these cells are electrically or neurologically activated. The signals produced by the EMG sensors are analyzed to detect abnormalities, activation level, or recruitment order, or to analyze the biomechanics of humans or animals movement. The data provided by the EMG sensors is related to the EMG frequency, which may be characterized by slow twitches in the range of 75 -125 Hz (twitches/sec) or fast twitches in the range of 125 -250 Hz (twitches/sec).

Heart Rate Sensors: The Heart Rate sensors measure electrical signals from the heart. These signals are transmitted to a wristwatch or a data center. The collected data is analyzed to interpret the workout performed by a user and better understand the benefits from the executed exercises and measure the energy efficiency of the exercising user.

The following Table 1 specifies all the sensors mounted on the wearable device, with their respective locations on the human body: 39991/IL/19-ORP- - Table 1 Fig. 3 illustrates a Shadow ® motion-capture (mocap) suit (manufactured by Motion Workshop, Seattle, WA, U.S.A.) that consists of 17 motion sensors and two pressure sensors. The motion sensors are marked by red dots. The pressure sensors are marked by blue dots. 39991/IL/19-ORP- - The data provided by the motion sensors is specified below. Each motion sensor produces the current X, Y and Z coordinates, in Radians: Motion sensor located on the hip : 'Hips.rx', 'Hips.ry', 'Hips.rz', Motion sensor located on the right thigh :'RightThigh.rx', 'RightThigh.ry','RightThigh.rz', Motion sensor located on the right leg: 'RightLeg.rx', 'RightLeg.ry', 'RightLeg.rz', Motion sensor located on the right thigh: 'RightFoot.rx', 'RightFoot.ry', 'RightFoot.rz', Motion sensor located on the left thigh: 'LeftThigh.rx', 'LeftThigh.ry', 'LeftThigh.rz', Motion sensor located on the left leg: 'LeftLeg.rx', 'LeftLeg.ry', 'LeftLeg.rz', Motion sensor located on the right thigh: 'LeftFoot.rx', 'LeftFoot.ry', 'LeftFoot.rz', Motion sensor located on the low spine: 'SpineLow.rx', 'SpineLow.ry', 'SpineLow.rz', Motion sensor located on the middle spine: 'SpineMid.rx', 'SpineMid.ry', 'SpineMid.rz', Motion sensor located on the chest: 'Chest.rx', 'Chest.ry', 'Chest.rz', Motion sensor located on the right shoulder: 'RightShoulder.rx', 'RightShoulder.ry', 'RightShoulder.rz', Motion sensor located on the right arm: 'RightArm.rx', 'RightArm.ry', 'RightArm.rz', Motion sensor located on the right forearm: 'RightForearm.rx', 'RightForearm.ry', 'RightForearm.rz', Motion sensor located on the right hand: 'RightHand.rx', 'RightHand.ry', 'RightHand.rz', Motion sensor located on the left shoulder: 'LeftShoulder.rx', 'LeftShoulder.ry', 'LeftShoulder.rz', Motion sensor located on the left arm: 'LeftArm.rx', 'LeftArm.ry', 'LeftArm.rz', Motion sensor located on the left forearm: 'LeftForearm.rx', 'LeftForearm.ry', 'LeftForearm.rz', Motion sensor located on the left hand: 'LeftHand.rx', 'LeftHand.ry', 'LeftHand.rz', Motion sensor located on the neck: 'Neck.rx', 'Neck.ry', 'Neck.rz', Motion sensor located on the head: 'Head.rx', 'Head.ry', 'Head.rz'] Figs. 4-5 illustrates a Shadow® motion-capture (mocap) suit (manufactured by Motion Workshop, Seattle, WA, U.S.A.) that consists of 17 motion sensors and two pressure 39991/IL/19-ORP- - sensors. The motion sensors are marked by red dots. The pressure sensors are marked by blue dots. The EMG sensors are marked by green dots.

The system provided by the present invention may also include: a Heart Rate Monitor (HRM), which is a personal monitoring device that allows one to measure/display heart rate in real-time, or record the heart rate for later analysis. Heart rate monitors commonly use electrical or optical methods to record heart signals. Both types of signals can provide the same basic heart rate data, using fully automated algorithms to measure heart rate, such as the Pan-Tompkins algorithm (this algorithm is used to detect QRS complexes in electrocardiographic signals. The QRS complex represents the ventricular depolarization and the main spike visible in an ECG signal. This feature makes it particularly suitable for measuring heart rate, in order to assess the heart health state); ECG (Electrocardiography) sensors that measure the bio-potential generated by electrical signals that control the expansion and contraction of heart chambers, typically implemented in medical devices and PPG (Photoplethysmography) sensors that use light-based technology, in order to measure the blood volume that is controlled by the heart's pumping action.

According to one aspect, the sampling rate should be 100 events per second. Each event should contain all the data from all the sensors. For example, the sampling rates of the sensors installed in a Shadow® motion-capture (mocap) suit are 100, 200, or 400 Hz.with a 1KHz internal update rate. The orientation of the sensors is drift-free 3D rotation (in which position drift resulting from the integration of acceleration or velocity is removed, so as to obtain accurate position estimation) with 0.5 ° static accuracy (at rest) and 2 ° dynamic accuracy (during rapid motion).

In a preferred embodiment, all sensors should are water-proof and are adapted to be easily removable and re-inserted (in case the shirt needs washing, for example). Maximum sensor size: 33 x 18 x 6 mm for miniature sensors. The device comprises a battery that has a duration of at least 5 hours of uninterrupted usage. 39991/IL/19-ORP- - The suit should comprise a data transmission device (such as Bluetooth), a controller, and a memory, in order to allow a connection to a computer or smartphone to connect to the suit in order to read the data collected by the sensors.

The system includes a database of the performance of the exercises that stores the historical data collected from regular users, additional reference data recorded from professional athletes, and common user mistakes. The stored data will be used to compare the collected data during the exercise of each user and make decisions regarding the accuracy of his performance. The system also comprises a Software Development Kit (SDK) that will enable easy access to the information received from the various sensors. All access to the sensor data being generated by the suit is done through the interface defined in the SDK. This provides encapsulation and decoupling between the operational software and the hardware components in the suit.

The machine learning models of the system of the present invention work with local Euler angles, but can also work with global quaternion, local quaternion, and global acceleration. For example, the preview service provides access to the current orientation output as a quaternion, a set of Euler angles, or a 4-by-4 rotation matrix. The orientation output can be accessed in the global or local coordinate frame. The preview service also provides a current estimate of linear acceleration in the global coordinate frame. ]global quaternion, local quaternion, local euler, global acceleration[ {Gqw Gqx Gqy Gqz Lqw Lqx Lqy Lqz Rx Ry Rz lax lay laz } Figure 1: Preview service data format. The global coordinate frame is defined by gravity and the geomagnetic field. The global identity orientation is Y pointed up in the direction of gravity and X pointed towards the geographic pole. The global quaternion is the primary output of the filtering pipeline, and it is used to compute the other Preview elements. Where applicable, orientations are specified in a right-handed coordinate system. 39991/IL/19-ORP- - The local coordinate frame is expressed relative to: • an arbitrary start orientation • a user-defined rest pose orientation, and with respect to parent orientation.

By default, the local orientation is a rotation of the global axes. There are two additional local rotation modes for convenience. Use the Lua command node.system.set local mode() to switch between the rotation modes.

• Sensor, rotate about the local axes of the sensor defined at the start time.

• World + Heading, rotate about the global axes with the X-axis pointing forward. The forward direction is defined by the rotation of the vertical axis at the start time.

Example 1: Technique Improvement The user will be able to choose from several different options of exercises. The data collected during the performance of the exercise will be stored in a dedicated database.

When working on the technique of a specific exercise, the system will analyze the performance of the user (trainee) and correct the performance using a Machine Learning algorithm. After completing the exercise, the system will provide the user with a score and, in addition, will provide him real-time feedback for the immediate correction of the technique.

Example 2: Real-Time Training Program Adjustment (Personal Trainer) The user will choose his goal (for example, build muscles or gain power), and according to the selected goal, his profile, and his history (that is based on data previously recorded), the system will build a dynamic training program that will be adjusted continuously, according to the trainee's progress toward the selected goal. 39991/IL/19-ORP- - Example 3: Weights and Repetitions The system will be adapted to determine what would be the right weight for the user to lift and the right amount of repetitions for improving muscle strength/hypertrophy. The system will also determine the user's abilities for each muscle group, followed by recommendations of weights and the number of repetitions for that muscle group.

The user will be invited to try lifting different weights (sorted from low to high), and simultaneously, the system will also be adapted to check if the technique is within a range of good practice and will correct the user if not. The system will check with the EMG sensors if the strain in the muscle is in the correct range for the exercise and the goal. If the strain is found to be below the recommended range, the system will suggest increasing the weight, and if it is above this range, the system will suggest decreasing the weight.

Example 4: "Be like" The user can choose to practice on a specific movement technique he wishes to perform like his favorite athlete (for example, a soccer, football, basketball, or baseball player). The system will compare the user’s technique to the recorded data that corresponds to the athlete and then will guide the trainee, step by step, using continuous real-time feedback and correction to perform the movement exactly like the athlete.

Example 5: Adjustment by Sport Adjustment of the athlete's training program according to the sport type he carries out. Since each sport type requires different training on the human body, the system will adapt the training program to the type of sport being performed by the athlete.

Example 6: Injury Prevention The EMG and the motion sensors can identify stress applied on the muscles and warn before an impending injury. In such a case, the system gives the user a warning to stop exercising. 39991/IL/19-ORP- - The wearable device comprises multiple Inertial Measurement Units (IMUs – an IMU is an electronic device that measures and reports a body's specific force, angular rate, and the orientation of the body, using a combination of accelerometers, gyroscopes, and optionally, magnetometers. The IMU sensors are fixed to the wearable device, which is worn on different organs of a user (hands, legs, etc.) and collects data regarding the movement in 3 axes (radial- r, ϕ, θ or x, y, z). The information obtained from all the sensors is being encoded into one input vector for performing analysis, which is compatible with ML models.

The data is processed using Artificial intelligence (AI), by using Machine Learning models. The machine learning models automatically extract movement features. The movement accuracy is then classified into two categories: accurate or inaccurate. An indication (feedback) is provided to the user regarding the accuracy of the movements by using a Predictive Model (for example, Random Forest or Recurrent Neural Network). This model checks for every motion segment if it is within a specific predefined range of correct movement. The feedback is displayed to the user in real-time by a dedicated GUI that reflects his motion with an avatar (a graphical display in the shape of a doll that mimics the user). An indication can be provided by a simple sign if it is erroneous and shows the user what corrective action is required in order to fix the error and get into a specific predefined range of correct exercise movements.

In order to prepare the input for the predictive model, the method of the present invention applies feature processing, in order to convert the signals from the sensors into an input vector that can be used predictive, to predict if the user is performing the exercise correctly.

Fig. 1 is a flowchart of the process of generating the input vector from samples collected from the vest of an exercising person. At the first step 101, the wearable device generates 100 samples (events) per second, where each sample contains approximately 1000 data points which are the diverse measurements captured by each one of the sensors in the vest. At the next step 102, the data points that are more relevant (among 39991/IL/19-ORP- - the 1000 possible values that are captured by the sensors) are selected using only the values rx, ry and rz that describe, for each sensor, its location in Radians. At the next step 103 a series of events are concatenated to capture a movement. Experiments have shown that the minimum amount of time to capture a predictable movement is one second, for example. Therefore, a series of 100 consecutive events that were captured in the period of one second are concatenated together to create the vector that will be used as input for the predictive model. In other words, each vector will be the result of the concatenation of 100 consecutive events (samples) and represents a user movement with the duration of exactly one second.

Fig. 2 is a flowchart of the execution using the Exercise Classification Binary Model and the Exercise Explanation Model. At the first step 201, feature processing is performed, where the input is a raw data vector and the output is movement vectors. At the next step 202, classification is performed by the classification model, where the input is the movement vectors, and the output is the binary classification result: performing a correct or incorrect exercise. At the next step 203, feedback is provided to the user by the explanation model, where the input is the binary classification, and the output is the source of the mistake in the exercise (the most relevant part of the body that caused the mistake).

Fig. 3 is a flow chart that describes the data collecting and analysis, according to an embodiment of the present invention. At the first step 301, real-time data is received from the plurality of sensors in the wearable device in 3 axes in space: x,y,z. The angular position of each sensor (31 in total) is used on the suit in each axis relative to the rest pose. At the next step 302, data obtained in the structure of a vector containing a list of points in time to one argument that expresses motion segment is transferred to a processor. At the next step 303, the motion segment is compared to a statistical database. At the next step 304, the system determines by comparison with the database and using the AI model, if the motion segment was in a right or wrong technique. At the next step 305, a Machine Learning algorithm is used to point to a specific place where a 39991/IL/19-ORP- - mistake in the motion occurred and what the deviation was. At the next step 306, displaying to the user in dedicated GUI if the technique was right or wrong and if it was wrong what and where was wrong.

Model Prediction The system provided by the present invention uses two Machine Learning models: The first is the Binary Classification model, to classify the exercise execution into correct or incorrect. The second is a model that uses techniques of Explainable Machine Learning to indicate the reason that the exercise is incorrect.

The Exercise Classification Model (Binary Classification) receives as an input the data vector that is the result of step 103. The Exercise Classification Model then produces as an output a binary value that may indicate that this particular vector is part of a correct exercise or an incorrect exercise. Two different versions of this Classification Model were implemented, one using Random Forest and the other using Recurrent Neural Networks (Deep Learning).

The Exercise Explanation Model receives the output from the previous model and defines which were the data points that had the most weight in the classification. This model is able to identify which are the sensors that had the most impact on the classification result, and thus determine in which body part has located the source of the mistake in the execution of the exercise.

Fig. 4 illustrates the deployment of sensors on a wearable device in the form of a suit, according to an embodiment of the present invention. In this example, there are seven motion sensors, two EMG sensors, and one heart rate sensor on the front side of the wearable device (left view). There are five motion sensors and two EMG sensors on the rear side of the wearable device (right view).

Fig. 5 illustrates the deployment of sensors on the wearable device, according to another embodiment of the present invention. In this example, there are ten motion 39991/IL/19-ORP- - sensors on the front side of the wearable device (left view). There are seven motion sensors and two EMG sensors on the rear side of the wearable device (right view).

Fig. 6 schematically illustrates an architecture diagram of a system 600 for motion tracking and analysis during sports activities, according to an embodiment of the invention. System 600 comprises the following components: Vest Signal Preprocessing : This component receives the input from the sensors that are located on a vest of a user (as indicated by numeral 601), and then selects and prepares the relevant values for the system.

Feature Engineering : This component is responsible for building the feature vectors that serve as an input for the Machine Learning models.

Exercise Performance Verification : This component is responsible for checking if the performance of the exercise is correct or incorrect. If the performance is incorrect, this component also identifies the reason for the error.

User Session Management : This component manages all the input/output between the system and the user doing the exercise.

User Interface (602) : This component enables the user to enter commands into the system and is also responsible for displaying all the information and feedback about the exercise being performed.

User Dashboard Management : This component is responsible for the different information pieces displayed in the user dashboard, including graphs and alerts.

Population Statistics Management : This component is responsible for aggregating, computing, and displaying population statistics.

Exercise Execution Model : For each exercise, there is a Binary Classification model that can detect if the exercise is being performed correctly or incorrectly, and an Explainable ML model that can detect the source of the error. 39991/IL/19-ORP- - Exercise Specification : For each exercise, the exercise specification object contains information and necessary parameters to analyze the exercise performance.

System 600 further comprises the following databases: User Sessions Log : This database stores all the data generated during the session. Later these data may be analyzed or used to train new models.

User Profile : This database contains all the personal information about the user.

Exercise Specifications : This database contains the specifications for each exercise.

Exercise Execution Models : This database contains the models for each exercise.

Population Statistics : This database contains aggregated data for several user segments.

Fig. 7 schematically illustrates the Entity Relationship Diagram (ERD) of system 600, according to an embodiment of the invention. Entity Relationship Diagram (ERD) may include the followings: User : each User has a Profile and executes Exercises; Profile : the Profile contains the User’s attributes; Exercise : each Exercise has a Specification and an Execution Model; Exercise Specification : this entity contains information and parameters that are necessary to analyze the exercise performance; Exercise Execution Model : this model is adapted to detect if the Exercise is being performed correctly or incorrectly; Exercise Execution : the Execution contains all the data that is being generated during the session; Execution Performance : the Performance is the result of the analysis of the Execution; Population Statistics : this entity aggregates data for different segments of the population; and 39991/IL/19-ORP- - User Dashboard : this entity is responsible for displaying Performance and Statistics results.

The wearable device of the present invention creates the optimal fitness suite incorporating real-time feedback and a complex machine-learning algorithm. The wearable device is comfortable, enables the user to perform training exercises more precisely, and by that generates a new level of personal training. The wearable device is fitted to each client uniquely monitors essential parameters during a physical activity using sensors and algorithms. As a result, the wearable device provides real-time status on the user's muscles, the efficiency of the user's training, and recommendations for smarter training in the future.

In other embodiment of the present invention, the device may be used for tracking additional activities, like rehabilitation.

The method of the present invention can be applied to any purpose that requires analysis of physical activity, mutatis mutandis. The purpose of the system can be tailored to each specific purpose, whether it is for the needs of rehabilitation, military training, or even movement analysis for computer games.

Feature engineering: for each sensor in the wearable device, the system may use three signals, the spatial coordinates rx, ry and rz, which enable to determine of the accurate position of each member of the body. Each event (i.e., sample) generated by the wearable device has a value for each one of these coordinates. The goal is to capture a movement, and not a single event. Therefore, the system concatenates together a sequence of consecutive events (consecutive samples). For example, the wearable device may generate one hundred events per second. The experiments have indicated that the time-lapse of one second is appropriate to capture a movement. Therefore, one hundred consecutive samples are concatenated together for creating a vector of values that are used as the input of the Binary Classification model. For example, every ten milliseconds, a new vector is created and sent to be classified by the model, i.e., a new 39991/IL/19-ORP- - vector is created for each new event (sample) that is generated by the wearable device. This guarantees that the system is able to provide immediate feedback to the user.

The system may involve one or more of the following: - User Sessions Logs - Information about the execution of exercises and other parameters performed or analyzed on the current user in the active or past sessions; - User Profile - Information about users, including weight, age, height, gender, etc. and other personal details; - Exercise Specifications - Biomechanical and complementary information about the correct and incorrect executions of the different exercises; - Exercise Execution Models - Trained Machine Learning models that will analyze the execution of the user’s performance on the different exercises; and Population Statistics - Statistics about the user population segmented according to different demographic attributes.

As will be appreciated by the skilled person, the arrangement described hereinabove results in a system that is capable of providing technical accuracy of exercise while taking into account electrical activity inside the muscle (e.g., how much effort the muscle activates, whether it is about to be injured and more, etc.) and physiological measurement such as heartbeat. Thus, the combination of such various metrics produces a system that solves the need for physical activity analysis completely. In other words, the system of the present invention "knows" how to tailor a training program to the user, hermetically, in real-time, and also provides performance feedback to the user. For example, what to perform, how much to perform, with what resistance, and what the user needs to improve to perform the exercise correctly. Combining these metrics with machine learning provides the right tools to perform the calculations and personalize the training to the user given physiological conditions and goals.

While some embodiments of the invention have been described by way of illustration, it will be apparent that the invention can be carried out with many modifications,

Claims (9)

39991/IL/19-ORP- - Claims

1. A system for motion tracking and analysis during sports activities, comprising: a) a wearable device comprising a plurality of deployed motion sensors for generating transmitting motion raw data parameters, upon activation by an exercising user; b) a processing unit with memory, for storing and processing said raw data and processed data according to instructions of an operating software; c) a battery for providing power to said sensors and said processing unit; d) a display device for displaying processing results to said exercising user; e) a data communication transceiver, for exchanging data with at least one remote database of the performance of the exercises that stores data collected from regular users, reference data recorded from professional athletes and common user mistakes, wherein said processor and said operating software are adapted to: f) generate an input vector from collected samples by processing and arranging the collected data in a structured manner by selecting only values of data points that describe the location and movement events features of each sensor and concatenating a series of movement events, to thereby capturing a movement that corresponds to an exercise; g) feed said input vector to a predictive model and apply Machine Learning models to classify the execution of an exercise as accurate or inaccurate while using a different model for each kind of exercise; and h) display feedback to said user regarding the accuracy of exercises performed by said exercising user.

2. A system according to claim 1, in which the motion sensors are selected from the group of: accelerometers, gyroscopes, magnetometer. 39991/IL/19-ORP- -

3. A system according to claim 1, in which a machine learning model is used to identify and analyze the source of the error in the execution of an exercise, while using techniques of Explainable Machine Learning, to identify the source of the errors.

4. A system according to claim 1, in which the wearable device further comprises Electromyography (EMG) sensors and heart rate sensors.

5. A system according to claim 1, in which the wearable device further comprises a Heart Rate Monitor (HRM).

6. A system according to claim 1, in which the wearable device further comprises multiple Inertial Measurement Units (IMUs) to collect data regarding movements in 3 axes.

7. A system according to claim 1, in which a Binary Classification model is used to classify the exercise execution into correct or incorrect and an Explainable Machine Learning model, to indicate the reason that the exercise is incorrect.

8. A method for motion tracking and analysis during sports activities, comprising: a) deploying motion sensors at predetermined locations over the body of an exercising user; b) receiving real-time data in three axes in space from said motion sensors, including at least the angular position of each sensor in each axis relative to a rest pose; c) processing the received data and converting said received data to structured representation in the form of a vector containing a list of points in time to an argument that expresses a motion segment; 39991/IL/19-ORP- - d) comparing the features of said motion segment to features stored in a statistical database; e) determining by an AI model that receives the comparison results if the motion segment has been performed in a correct or incorrect exercise execution; f) using Machine Learning to analyze said motion segment and pointing on a specific location along said motion segment at which a mistake in the motion path occurred while specifying the deviation from correct motion characteristics along said motion path; g) displaying to the user on a GUI, an indication if the exercise execution was correct or incorrect; and h) if said exercise execution was found incorrect, providing an indication on said GUI what the mistake was and its location along said motion path.

9. A method according to claim 8, wherein the deployment of the motion sensors is performed on a wearable mocap suit. 10.A method according to claim 8, wherein the deployment of the motion sensors is performed on a wearable device, selected from a group consisting of: a vest, a watch, elastic straps, a bracelet. 11.A method according to claim 8, wherein additional sensors are used in combination with the motion sensors, for collecting physiological data from the body of the exercising user and processing said physiological data using Machine Learning to thereby improve the accuracy of the indications provided to exercising user. 39991/IL/19-ORP- - 12.A method according to claim 8, wherein the indication if the exercise execution is correct or incorrect in real-time is provided in a visual form, audible form, tactile form, or any combination thereof. 13.A method according to claim 12, wherein one or more of the indications are provided on a relevant region on the wearable device, in particular in accordance with the location of incorrect activity.