WO2017156835A1

WO2017156835A1 - Smart method and system for body building posture identification, assessment, warning and intensity estimation

Info

Publication number: WO2017156835A1
Application number: PCT/CN2016/080735
Authority: WO
Inventors: 伍楷舜; 邹永攀; 王丹; 吴金咏
Original assignee: 深圳大学
Priority date: 2016-03-18
Filing date: 2016-04-29
Publication date: 2017-09-21
Also published as: CN105797350A

Abstract

A smart method for body building posture identification, assessment, warning and intensity estimation. The method comprises: (S1) using a thin-film pressure sensor of a pressure sensor and an IMU inertial unit on a smart glove to collect original data during strength training; (S2) transmitting the original data to a processor module for performing data pre-processing; and (S3) the processor module transmitting the pre-processed data of the pressure sensor and the IMU inertial unit to a mobile terminal APP by means of wireless communication. Further provided is a smart system for body building posture identification, assessment, warning and intensity estimation.

Description

Intelligent method and system for fitness posture recognition, assessment, early warning and strength estimation

Technical field

The invention relates to an intelligent detecting device, in particular to an intelligent method and system for fitness posture recognition, evaluation, early warning and strength estimation.

Background technique

Nowadays, with the concept of healthy living becoming more and more popular, more and more fitness enthusiasts hope to measure the calorie consumption and subsequent dietary adjustment during strength training. However, the traditional acquisition of the human body during exercise Energy consumption is not only cumbersome, but also requires additional equipment, so we urgently need to find a way to easily and effectively detect calorie consumption and motion recognition during strength training.

In the daily fitness process, especially during strength training, the human body will consume a large amount of calories. In order to help the nutritional supplement and diet after training, do not accurately estimate the calorie consumption of the human body during training; at the same time, correct The posture is a necessary condition for any training program. This is because the correct posture helps to alleviate or even eliminate any potential training injuries and helps the trainer's health.

In order to achieve the preparation for the exercise and the calorie consumption, people have proposed the use of treadmills, bicycles, running shoes to detect the calorie consumption. However, the detection system built by these methods has its own shortcomings. A specific motion calculation module is used to detect calorie consumption. When using these specific calculation modules for calorie calculations, the measurement module is not mobile and cannot be truly wearable, in current research and industry. In production, there is no device that can achieve the above functions. The existing method of the latter cannot be widely applied to this scenario due to the particularity of the module.

Summary of the invention

In order to solve the problems of the prior art, the present invention provides an intelligent method and system for fitness posture recognition, evaluation, early warning, and strength estimation, which solves the problem of cumbersome energy consumption of the human body during the exercise process and the problem of adding additional equipment. .

The present invention is achieved by the following technical solutions: an intelligent method for fitness posture recognition, evaluation, early warning, and strength estimation, including the following steps:

(S1) using the membrane pressure sensor of the pressure sensor on the smart glove and the IMU inertia unit to collect raw data during strength training;

(S2) transmitting the original data to the processor module for data preprocessing;

(S3) the processor module transmits the pre-processed pressure sensor and the IMU inertial unit data to the mobile terminal APP by means of the communication unit by wireless communication; (S4) after the mobile terminal APP receives the data, the algorithm for running the data processing is obtained. Fitness posture judgment, motion quality assessment, exercise intensity estimation results, and the results are displayed graphically in the APP; when it is determined that the fitness movement is not standard or the fitness intensity is excessive, the APP will alert the user; The APP uploads the result of the data processing to the user's personal database of the cloud server, and the cloud server promotes the corresponding fitness service for the user based on the historical data of the user.

As a further improvement of the present invention, in the step (S1), the user opens the smart glove worn on the wrist by a button or a switch, and the sensing module of the smart glove is provided with a plurality of pressure sensors.

As a further improvement of the present invention, the step (S2) further includes: (S21), collecting pressure data and motion posture data by a pressure sensor and an IMU inertia unit, and performing data of the obtained acceleration sensor, magnetic sensor, and gyroscope Sliding mean filtering to eliminate background noise; (S22), according to the motion characteristics of a complete fitness movement, combined with 9-axis IMU data to cut the fitness exercise, get the starting point and end point of each complete movement, thus the continuous fitness process According to each motion movement, the segmentation is performed; (S23), the segmented data is classified by the machine learning method to realize the gesture recognition function; (S24), after the posture determination is performed, the user's motion data and the standard motion are implemented. The database is matched to evaluate the overall quality of the user's actions. At the same time, the stability and smoothness of the user action data are analyzed, and various evaluation indicators of the user's fitness action quality are obtained; (S25), after the posture determination is performed, for each action process Further segmentation, split into the forward and return, and two sub-processes In the analysis of the time domain and the amplitude domain, when the indicators on the time domain and the amplitude domain exceed the set threshold, the action is judged as non-standard, and the user is given a corresponding non-standard prompt; (S26), the acceleration sensor and the gyro The reading of the instrument uses an complementary filter to accurately estimate the attitude angle of the glove. The complementary filter is used to combine the estimated value of the attitude angle of the acceleration sensor with the attitude of the glove and the attitude of the glove to the attitude angle of the glove; in step S27, a complementary filter is used. The way to estimate the attitude angle of the glove during the movement, according to the mechanical principle, eliminate the projection component of gravity in the direction of the wisdom glove, and extract the wisdom glove itself. The acceleration value generated by the motion realizes the acceleration calibration; (S28), according to the kinesiology principle of the human body, the trajectory of the fitness movement is reconstructed in combination with the nine-axis IMU data, and the reading of the pressure sensor is combined with the processing of the step S3 to calculate the glove movement. The work, according to the relationship between the established work and calorie consumption model to calculate calorie consumption; (S29), record the calorie consumption, and remind the meal mix and nutritional balance.

The invention also provides an intelligent system for fitness posture recognition, assessment, early warning and strength estimation, including:

a signal acquisition and calculation module for collecting signals of the movement state of the smart glove and evaluating the motion state information for preliminary calculation;

An abnormality detecting module, configured to identify, by using an anomaly detection algorithm, whether the signal is abnormal;

The action judging module is used for classifying the target action class and other action classes to distinguish the abnormal mode caused by the strength training threshold as the target action class, and determining whether the posture of the smart glove is deviated;

And an alarm module for issuing an alert signal when it is determined that an attitude deviation occurs.

As a further improvement of the present invention, the signal acquisition and calculation module includes:

An inductive acquisition unit for turning on the entire system and collecting motion data, and the collected motion data includes motion data in the direction of strength training and motion data in all directions of three-dimensional coordinates;

a data processing unit, configured to average the force and motion information in each direction, and use the average value as the motion state information;

The smoothing unit performs ETL analysis on the data obtained by the data processing unit, and smoothes the motion state information by a moving average method.

As a further improvement of the present invention, the abnormality detecting module includes:

An abnormality calculating unit, configured to perform data segmentation on the time series of the motion state information to obtain a subsequence, and calculate local abnormal data of the subsequence;

The abnormal output unit is configured to output the sub-sequence as an abnormal mode when the abnormal data is greater than or equal to a preset threshold.

As a further improvement of the present invention, the action judging module includes:

Establishing a model unit for establishing a high-dimensional feature model based on a statistical learning theory preset, wherein the high-dimensional feature model uses an abnormal pattern in a set space due to changes in motion state information due to various human actions as a training sample;

The motion recognition unit is configured to map the abnormal pattern output by the abnormal output unit to a high-dimensional feature model of a type of support vector machine, and separate the target action class.

As a further improvement of the present invention, a feedback module is further included for feeding back response information for the hand posture warning signal, and adjusting a high-dimensional feature model of a type of support vector machine.

The invention has the beneficial effects that the invention performs calorie measurement and hand gesture recognition on the basis of the existing sensor technology, and adopts an effective data processing method and an algorithm model, and can be widely applied in strength training. Provide a good and useful reference for fitness enthusiasts, and provide a recommendation for post-meal mix and nutritional balance.

DRAWINGS

1 is a schematic diagram of an intelligent method for transmitting and receiving data of fitness posture recognition, evaluation, early warning, and strength estimation according to the present invention.

detailed description

The invention will now be further described with reference to the drawings and embodiments.

An intelligent method for fitness posture recognition, assessment, early warning, and strength estimation includes the following steps:

(S3) the processor module transmits the pre-processed pressure sensor and the IMU inertial unit data to the mobile terminal APP by means of the communication unit by wireless communication; (S4) after the mobile terminal APP receives the data, the algorithm for running the data processing is obtained. Fitness posture judgment, motion quality assessment, exercise intensity estimation results, and the results are displayed graphically in the APP; when determining that the fitness posture is not standard, the APP will alert the user; meanwhile, the APP will process the data. The result is uploaded to the user's personal database of the cloud server, and the cloud server promotes the corresponding fitness service for the user based on the historical data of the user.

In the step (S1), the user opens the smart glove worn on the wrist through a button or a switch, and the sensing module of the smart glove is provided with a plurality of pressure sensors.

The step (S2) further includes: (S21), collecting pressure data and motion posture data through the pressure sensor and the IMU inertia unit, and performing sliding average filtering on the obtained acceleration sensor, the magnetic sensor and the gyroscope data to eliminate the background Noise; (S22), according to the motion characteristics of a complete fitness movement, combined with the 9-axis IMU data to cut the fitness exercise, get the starting point of each complete movement And the end point, thereby dividing the continuous exercise process according to each exercise action; (S23), applying the machine learning method to the segmented data (ie, data corresponding to each action) (eg, SVM, KNN) ()) to perform classification and realize the function of gesture recognition; (S24), after the posture determination is performed, the user's motion data is matched with the standard action database (for example, the fitness coach's fitness data), and the overall quality of the user motion is evaluated. Perform stability and smoothness analysis on the user action data, and obtain various evaluation indicators of the user's fitness action quality; (S25), after the posture determination is implemented, further segmentation is performed for each action process to be divided into a forward process and a return process. And the two sub-processes are analyzed in the time domain and the amplitude domain. When the indicators in the domain and the amplitude domain exceed the set threshold, the action is judged as non-standard, and the user is given a corresponding non-standard prompt; (S26 ), the accelerometer and the gyroscope are read by a complementary filter to achieve an accurate estimation of the glove attitude angle. The complementary filter is used to combine The speed sensor estimates the glove attitude angle and the gyroscope's estimated attitude angle of the glove; in step S27, the complementary filter is used to estimate the attitude angle of the glove during the movement, and the gravity is removed in the wisdom glove according to the mechanical principle. The projection component in the direction of motion extracts the acceleration value generated by the movement of the smart glove itself to achieve acceleration calibration; (S28), according to the kinesiology of the human body, combined with the nine-axis IMU data to reconstruct the trajectory of the fitness movement, combined with the reading of the pressure sensor After the processing of step S3, the work done by the glove movement is calculated, and the calorie consumption is calculated according to the relationship model between the established work and the calorie consumption; (S29), the calorie consumption is recorded, and the meal matching and nutritional balance are reminded.

The signal acquisition and calculation module includes:

The abnormality detecting module includes:

The action judging module includes:

A feedback module is further included for feeding back response information for the hand posture warning signal, and adjusting a high-dimensional feature model of a type of support vector machine.

In an embodiment, the smart glove is a smart glove for recognizing calories and hand gestures, including: a processor module, a storage module, a communication module, a sensing module, an early warning prompt module, a display module, a power module, and a switch module. And the cloud server, the storage module, the communication module, the sensing module, the warning prompt module, the display module, the power module, the switch module, and the cloud server are respectively connected to the processor module, wherein the sensing module includes a pressure sensor and an IMU inertia The unit, the pressure sensor and the IMU inertia unit are respectively responsible for sensing the amount of pressure of the hand and the IMU data during the fitness process of the user, and collecting the source data for the user strength training, and transmitting the source data to the processor module. The processor module is responsible for preprocessing the source data; the storage module is responsible for storing the result of the processor preprocessing and the processing result returned from the cloud server; the communication module is responsible for the source data, the preprocessed Data is passed to the cloud server for further data processing and analysis by the cloud server And is responsible for transmitting the result of the cloud server processing to the smart glove; the warning prompting module is responsible for issuing a warning to the user when the user action is not standard, the exercise is excessive, or the force is not in any situation; the display module is responsible for Will use The results of the type of action, calories burned, and posture standard during the fitness process are displayed.

One or more pressure sensors are disposed on the sensing module of the glove, the pressure sensor is disposed at the hand grasping force, and the IMU inertia unit is disposed at the wrist for convenient use by the customer; the storage module includes a built-in large capacity The memory and the external memory interface facilitate the storage and invocation of the user's motion data; the source data is respectively used to sense the amount of pressure of the hand and the data collected by the IMU data during the fitness process of the user through the pressure sensor and the IMU inertia unit. .

In one embodiment, an intelligent method of fitness posture recognition, assessment, early warning, and intensity estimation includes the following steps:

The pressure sensor's membrane pressure sensor and the IMU inertia unit are used to collect source data during strength training; the source data is transmitted to the processor module for processing;

Performing an ETL process by performing the ETL process on the data obtained by the received pressure sensor and the IMU inertia unit, which is a process of extracting, transposing, loading, and delivering the data obtained by the step S2 by the ETL technology; The process refers to the Extract, Transform and Load of the data, that is, the process of extracting, transposing, loading and delivering the data, which refers to the important steps of preprocessing the data before the data processing; in step S4, according to the set calculation model, the human body The mechanical work outputted during the movement can be linked to the calories burned by the human body in the process by a conversion factor, combined with the readings of the load cell and the reconstructed motion trajectory to calculate the work, thereby calculating the calories burned and A type of support vector machine that distinguishes the target action class from other action classes; takes the strength training as the target action class, and determines whether the threshold of the strength training is exceeded, and if so, sends a reminder signal; this step S4 transmits the data to the Mobile devices, record, recommend dietary recipes, nutritional mix and training programs; Threshold may be based on user demand to customize settings, such as in the use of barbells, preset by the size of the force, called the threshold, and if this exceeds a preset magnitude, a signal is issued;

And calculating the calories burned, establishing a classification model through the support vector machine, using the source data beyond the strength training as the target action class, and the other as the non-target action class; then determining whether the threshold of the strength training is exceeded, and if so, issuing a reminder signal If otherwise, transfer the data to a mobile device for recording, and recommend dietary recipes, nutritional mixes, and training programs.

In practical applications, the processor module receives the pressure sensor and the IMU inertia unit to measure the required motion data, and establishes the relationship between the motion signal and the strength training action, and only needs to use a simple sensor to pass the calorie of the detected person. Consumption and recognition of hand posture, judging that the subject is No deviation of the hand posture occurs and an alarm is issued, which reduces the dependence on the cumbersome measuring device, and greatly improves the correct rate of hand posture recognition; in a specific movement, the calorie consumption can be displayed through the mobile device terminal Recommended with meals.

The user preferably opens the entire system by means of a button or switch, which may be a button and a switch implemented by a touch sensor, which may be provided at the IMU inertial unit accessory for user operation.

The abnormality detection algorithm is used to identify the abnormality of the acquired motion information is a time series anomaly detection algorithm based on the local anomaly factor, and includes the following steps:

The pressure data and the motion attitude data are collected by the pressure sensor and the IMU inertia unit, and the obtained acceleration sensor, the magnetic sensor and the gyroscope data are subjected to sliding mean filtering to eliminate background noise based on the wearable computing technology, and the sliding mean filter is eliminated. The sliding window width is 7;

An accurate estimation of the glove attitude is achieved by using a complementary filter for reading the acceleration sensor and the gyroscope, the complementary filter being used to combine the estimated value of the attitude angle of the acceleration sensor with the attitude of the glove and the estimated attitude of the gyroscope to the attitude of the glove;

After using the complementary filter to estimate the attitude angle of the glove during the movement, the projection component of the gravity in the direction of the wisdom glove is removed according to the mechanical principle, and the acceleration value generated by the movement of the wisdom glove is extracted to realize the acceleration calibration. ;

Obtain the calibrated acceleration data, obtain the speed value of the wisdom glove movement by one-time integration; after the integral gets the speed value of the wisdom glove movement, according to the movement characteristics of the wisdom glove starting from the static state and ending at the static state, the wisdom The speed of the glove movement is calibrated;

The resulting velocity value is integrated once to obtain the displacement of the wisdom glove movement.

The analysis and processing of the data is realized by the complementary filter, and the attitude of the object in space can be combined by the acceleration and the angular velocity to achieve accurate estimation. Therefore, the estimated value of the glove attitude angle measured by the complementary filter combined with the acceleration sensor is The estimated value of the glove attitude angle measured by the gyroscope, which in turn gives an accurate estimate of the glove posture. The acceleration value generated by the movement of the smart glove itself is extracted, that is, the acceleration component of the wisdom glove in three directions of the three-dimensional coordinates is filtered out, which can be realized by complementary filtering or Kalman filtering. In order to reduce the speed drift after the integration, the integral is only for the acceleration value corresponding to the movement of the glove. For this reason, the exponential mean (EMA) is used as the judgment whether the glove acceleration value corresponds to the stationary state or the motion state of the glove; When the exponential mean is greater than the threshold, The acceleration in the window is determined to correspond to the motion state of the glove; otherwise, it is determined to correspond to the stationary state of the glove; when the judgment process is completed, the integral is only applied to the acceleration value corresponding to the glove motion state, thereby obtaining Corresponding speed values, which correspond to the acceleration values of the glove in a static state, are regarded as zero values; the threshold values are comprehensively determined in the data processing process in combination with sensors, algorithms, and user requirements. In different cases, the threshold values are It will be different. The index state average (EMA) described in step S24 is used to determine the motion state of the glove. When it is determined to be stationary, the corresponding speed value is forcibly set to zero.

This example also includes a high-dimensional feature model for the adjustment of a type of support vector machine for calorie consumption and hand gesture recognition.

Combining the obtained displacement value, combined with the reading of the film pressure sensor, after processing, calculating the work done by the glove movement, and calculating the calorie consumption according to the relationship model between the established work and the calorie consumption;

Record calorie consumption and remind you of dietary mix and nutritional balance;

The high-dimensional feature model is pre-established based on the statistical learning theory, and the high-dimensional feature model is used as a training sample by setting an abnormal pattern of information changes due to various posture deviation actions in the space;

The abnormal pattern is mapped to a high-dimensional feature model of a type of support vector machine, and the target action class is separated, and whether the motion posture is deviated is determined, and if so, a warning signal is issued.

The relationship model is that the mechanical work outputted by the human body during the movement is linked to the calories consumed by the human body in the process by the conversion factor, and the work is calculated by combining the readings of the load cell and the reconstructed motion trajectory, thereby calculating The calories burned out then distinguish the target action class from the other action classes.

In this example, the model is constructed based on the relationship between work and calories in the glove posture, so that the calculation and analysis can be performed. The work state can be expressed by the motion state, which shows the start time of the motion, the end time of the motion, and the instantaneous speed of the motion; Not all internal energy can be used for work, and part of the internal energy can be lost due to heat consumption and blood flow. The internal energy of the human body can be successfully converted according to a certain ratio. The measurement of energy loss is calculated according to the following formula: the mechanical output of the external output during the movement of the human body = the efficiency of calorie consumption to mechanical work * the calories burned during the movement of the human body. Based on the relevant knowledge of machine learning, the model relationship between work and calorie consumption is constructed. Mechanical work is measured in wearable computing, usually by measuring the motion trajectory to calculate the work done by the human body. Further, measuring motion trajectories is accomplished using some common sensors such as acceleration sensors and gravity sensors.

The data obtained in this example is generated over time, that is, the time series model. Now data analysis and processing, this example divides the acquired data, and the subsequence refers to intercepting a piece of data acquired by the sensing module and passing through the ETL process. As for the length and data amount of the subsequence, according to the processing capability of the system, Defining settings; the local anomaly data refers to data calculated by a support vector machine method in a subsequence, and part of the data is significantly deviated from other data of the subsequence; the preset threshold is a preset threshold In the data processing process, combined with sensors, algorithms and user requirements for comprehensive determination, in different cases, this threshold will be different.

This example also preferably includes a high-dimensional feature model for adjusting and improving a class of support vector machines, and provides system feedback that can optimize detection and decision algorithms. If the alert is not turned off in time, the system will send help information to others through other devices associated with the signal, such as sending instant messages or text messages through third-party applications.

The above content is further described in connection with the specific implementation manner, and it should not be construed that the specific implementation of the present invention is limited to the above description. It will be apparent to those skilled in the art that the present invention may be practiced without departing from the spirit and scope of the invention.

Claims

An intelligent method for fitness posture recognition, evaluation, early warning and strength estimation, which comprises the following steps:

(S1) using the membrane pressure sensor of the pressure sensor on the smart glove and the IMU inertia unit to collect raw data during strength training;

(S2) transmitting the original data to the processor module for data preprocessing;

(S3) the processor module transmits the pre-processed pressure sensor and the IMU inertial unit data to the mobile terminal APP by means of the communication unit in a wireless communication manner;

(S4) After receiving the data, the mobile terminal APP runs an algorithm of data processing to obtain a result of fitness posture determination, motion quality evaluation, and exercise intensity estimation, and displays the result in a graphical manner in the APP; When the standard is not standard or the fitness intensity is excessive, the APP will prompt the user for warning; at the same time, the APP uploads the data processing result to the user personal database of the cloud server, and the cloud server introduces the corresponding fitness service for the user based on the historical data of the user. .
The intelligent method for fitness posture recognition, evaluation, early warning and strength estimation according to claim 1, wherein in the step (S1), the user opens the smart glove worn on the wrist by a button or a switch, There are several pressure sensors on the sensor module of the wisdom glove.
The intelligent method for fitness posture recognition, evaluation, early warning and strength estimation according to claim 1, wherein the step (S2) further comprises: (S21), collecting pressure data by using a pressure sensor and an IMU inertia unit; Motion posture data, the obtained acceleration sensor, magnetic sensor and gyroscope data are subjected to sliding mean filtering to eliminate background noise; (S22), according to the motion characteristics of a complete fitness movement, combined with 9-axis IMU data to cut the exercise The starting point and the end point of each complete movement are obtained, so that the continuous exercise process is divided according to each movement action; (S23), the segmented data is classified by the machine learning method, and the posture recognition function is realized. (S24), after the posture determination is implemented, the user's motion data is matched with the standard action database, and the overall quality of the user motion is evaluated, and at the same time, the user motion data is analyzed for stability and smoothness, and the user's fitness motion quality is obtained. Each evaluation indicator; (S25), after the posture determination, for each action process Further segmentation is performed to divide into the forward and return paths, and the two sub-processes are analyzed in the time domain and the amplitude domain. When the indicators in the domain and the amplitude domain exceed the set threshold, the action is judged as not Standard and give the user a corresponding non-standard prompt; (S26), the complementary sensor is used for the reading of the accelerometer and the gyroscope The method achieves accurate estimation of the attitude angle of the glove, and the complementary filter is used to combine the estimated value of the attitude angle of the acceleration sensor with the attitude of the glove and the attitude of the glove to the attitude angle of the glove; and in step S27, the complementary filter is used to estimate the movement of the glove during the movement. After the attitude angle, the projection component of gravity in the direction of the wisdom glove is removed according to the mechanics principle, and the acceleration value generated by the movement of the wisdom glove is extracted to realize the acceleration calibration; (S28), according to the kinematics principle of the human body The axis IMU data reconstructs the trajectory of the fitness movement, combined with the reading of the pressure sensor, through the processing of step S3, calculates the work done by the glove movement, and calculates the calorie consumption according to the relationship model between the established work and the calorie consumption; (S29), recording Calorie consumption, and reminders of dietary mix and nutritional balance.
An intelligent system for fitness posture recognition, assessment, early warning and strength estimation, comprising:

a signal acquisition and calculation module for collecting signals of the movement state of the smart glove and evaluating the motion state information for preliminary calculation;

An abnormality detecting module, configured to identify, by using an anomaly detection algorithm, whether the signal is abnormal;

The action judging module is used for classifying the target action class and other action classes to distinguish the abnormal mode caused by the strength training threshold as the target action class, and determining whether the posture of the smart glove is deviated;

And an alarm module for issuing an alert signal when it is determined that an attitude deviation occurs.
The intelligent system for fitness posture recognition, evaluation, early warning and strength estimation according to claim 4, wherein the signal acquisition and calculation module comprises:

An inductive acquisition unit for turning on the entire system and collecting motion data, and the collected motion data includes motion data in the direction of strength training and motion data in all directions of three-dimensional coordinates;

a data processing unit, configured to average the force and motion information in each direction, and use the average value as the motion state information;

The smoothing unit performs ETL analysis on the data obtained by the data processing unit, and smoothes the motion state information by a moving average method.
The intelligent system for fitness posture recognition, evaluation, early warning, and strength estimation according to claim 4, wherein the abnormality detecting module comprises:

An abnormality calculating unit, configured to perform data segmentation on the time series of the motion state information to obtain a subsequence, and calculate local abnormal data of the subsequence;

An abnormal output unit, configured to use the subsequence as different when the abnormal data is greater than or equal to a preset threshold Normal mode output.
The intelligent system for fitness posture recognition, evaluation, early warning and strength estimation according to claim 4, wherein the action judgment module comprises:

Establishing a model unit for establishing a high-dimensional feature model based on a statistical learning theory preset, wherein the high-dimensional feature model uses an abnormal pattern in a set space due to changes in motion state information due to various human actions as a training sample;

The motion recognition unit is configured to map the abnormal pattern output by the abnormal output unit to a high-dimensional feature model of a type of support vector machine, and separate the target action class.
The intelligent system for fitness posture recognition, evaluation, early warning and strength estimation according to claim 4, further comprising a feedback module for feeding back response information for the hand posture warning signal, and adjusting a type of support vector machine High dimensional feature model.