CN113095379A - Human motion state identification method based on wearable six-axis sensing data - Google Patents
Human motion state identification method based on wearable six-axis sensing data Download PDFInfo
- Publication number
- CN113095379A CN113095379A CN202110325902.2A CN202110325902A CN113095379A CN 113095379 A CN113095379 A CN 113095379A CN 202110325902 A CN202110325902 A CN 202110325902A CN 113095379 A CN113095379 A CN 113095379A
- Authority
- CN
- China
- Prior art keywords
- data
- axis
- motion state
- model
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1123—Discriminating type of movement, e.g. walking or running
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
- A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
- A61B5/6802—Sensor mounted on worn items
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
- A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
- A61B5/6802—Sensor mounted on worn items
- A61B5/681—Wristwatch-type devices
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Veterinary Medicine (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Biomedical Technology (AREA)
- Theoretical Computer Science (AREA)
- Heart & Thoracic Surgery (AREA)
- Biophysics (AREA)
- Surgery (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Dentistry (AREA)
- Evolutionary Biology (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Bioinformatics & Computational Biology (AREA)
- Psychiatry (AREA)
- Human Computer Interaction (AREA)
- Social Psychology (AREA)
- Multimedia (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a human motion state identification and analysis method based on wearable six-axis sensing data. The method comprises the steps that firstly, six-axis sensing data are collected on the basis of hand ring equipment (which can also be configured on the waist, the chest and other parts) worn on a body, and a data set of a motion state is established; then, carrying out operations such as cleaning, slicing and intercepting on the data to obtain a standard data set corresponding to the activity state of the human body; then, enhancing and expanding data of the standard data set through offset operation; then, taking data with the length of 2 seconds as a batch, and extracting 36-dimensional robustness characteristics with direction independence; and finally, building a motion state discrimination model based on a random forest algorithm. The method has the characteristics of compactness, quickness and the like, can be loaded on embedded hardware equipment, and accurately judges the current motion state of the human body in near real time (once in 2 seconds).
Description
Technical Field
The invention belongs to the technical field of computer data mining, and particularly relates to a human motion state identification and analysis method based on wearable six-axis sensing data.
Background
The exercise safety and daily health monitoring of many people (including sub-healthy people, people with damaged exercise joints, people with empty nests and the like) become very important. People have many motion states every day, walk, run, sit down, stand etc. multiple motion modes, through monitoring human daily motion state, can instruct people to make healthy reasonable diet plan, rationally arrange the amount of exercise every day, especially to old man, senile dementia patient, feed back their daily motion data to medical personnel, improve healthy living level and motion safety and have important meaning.
Human motion state recognition is becoming a very popular research topic. Due to the complexity of the motion characteristics of human activities, there are currently many algorithmic studies in the field of motion recognition based on machine learning. Research processes typically use sensors to collect data, which is then classified using machine-learned algorithms. The most commonly used classification algorithms include Support Vector Machines (SVMs), k-nearest neighbors (KNNs), C4.5, Artificial Neural Networks (ANN), Dynamic Bayesian Networks (DBNs), Hidden Markov Models (HMMs), Gaussian Markov Models (GMMs), etc.
Human motion state identification is a typical classification problem, with the goal of detecting and identifying a person's daily activities. In order to be more suitable for wearing by the old, the designed equipment has small volume, low power consumption, low cost and no interference, and an efficient recognition algorithm needs to be designed in order to accurately recognize the behavior of the human body and realize the behavior on embedded equipment with limited computing capacity. The invention collects data in real time based on a six-axis inertial sensor LSM6DS3, and identifies the motion states of walking, running, sitting down, standing up, going upstairs, going downstairs and the like of a person in real time through big data analysis and a design algorithm.
Disclosure of Invention
The invention discloses a human motion state identification and analysis method based on wearable six-axis sensing data. The method comprises the steps that firstly, six-axis sensing data are collected on the basis of hand ring equipment (which can also be configured on the waist, the chest and other parts) worn on a body, and a data set of a motion state is established; then, carrying out operations such as cleaning, slicing and intercepting on the data to obtain a standard data set corresponding to the activity state of the human body; then, enhancing and expanding data of the standard data set through offset operation; then, taking data with the length of 2 seconds as a batch, and extracting 36-dimensional robustness characteristics with direction independence; and finally, building a motion state discrimination model based on a random forest algorithm. The method has the characteristics of compactness, quickness and the like, can be loaded on embedded hardware equipment, and accurately judges the current motion state of the human body in near real time (once in 2 seconds).
Compared with the prior art, the method overcomes the defects of the traditional motion state identification method based on time difference and threshold, utilizes big data analysis and artificial intelligence technology, automatically classifies and identifies, and further improves the accuracy and the efficiency; particularly, operations such as cleaning, slicing, offsetting, enhancing and expanding of data are performed, so that the data are more balanced and effective, and the accuracy of judgment is further improved. The method has wide application prospect.
The method comprises the following steps:
(1) establishing a six-axis active state acquisition data set;
(2) constructing a standard data set corresponding to the six-axis motion state of the human body;
(3) performing enhancement expansion on the standard data set;
(4) extracting data characteristics of the motion state of the human body;
(5) constructing an algorithm model for judging the six-axis motion state based on the characteristics;
(6) training a random forest algorithm model based on a training data set;
(7) and carrying out motion state identification and identification result accuracy evaluation based on the test data set.
The step (1) of establishing the acquisition data set of the six-axis activity state specifically comprises the steps of adopting six-axis sensor equipment of an LSM6DS model to acquire data, reading 3D digital accelerometers (X-axis acceleration, Y-axis acceleration and Z-axis acceleration) and 3D digital gyroscope data (X-axis angular velocity, Y-axis angular velocity and Z-axis angular velocity) with sampling frequency of 10 times per second. The motion states to be detected include: sit down, stand up, walk, run, go upstairs, go downstairs, sleep, define as the label respectively: 0,1,2,3,4,5,6. Data obtained to "acquireThe device type-tag-timestamp-six-axis information "form is saved as a set of file lists. And constructing an initial six-axis motion state data set on the basis of the file list set. And cleaning the data, and screening out some dirty data according to whether the dirty data is smaller than a preset combined acceleration threshold T1 and a preset combined angular velocity threshold T2. Wherein, the calculation formula of the resultant acceleration is as follows:
wherein
Represents the acceleration corresponding to the X-axis,
represents the acceleration corresponding to the Y-axis,
representing the acceleration corresponding to the Z-axis. And the calculation formula of the angular velocity is:
wherein
Representing the angular velocity corresponding to the X-axis,
representing the angular velocity corresponding to the Y-axis,
representing the angular velocity corresponding to the Z-axis.
And (3) constructing a standard data set corresponding to the six-axis motion state of the human body in the step (2), and particularly dividing a data interval of effective activity time based on the acquired data set. The effective activity time refers to a time interval in a motion state in the data recording process. Slicing the data of the whole data file into a section of data set; then, calculating the sum of the combined accelerations of the data in each section of data, solving the section with the maximum sum of the combined accelerations, selecting the section with the maximum sum of the combined accelerations as the central position of effective data, and taking the position as the center; and finally, traversing the adjacent data segments from near to far in sequence, and if the sum of the combined accelerations of the data segments is less than a predefined threshold value T, terminating the traversal, thereby obtaining the data segments of the effective activity time in the data file.
The standard data set is enhanced and expanded in the step (3), and two data enhancement schemes are provided, wherein the first data enhancement scheme is as follows: and integrally translating the data part corresponding to the effective activity time in the data section where the effective activity time is located, and obtaining a new data sample every time moving one unit of data. The range interval of the movement is +/-16% L, wherein L is the data length corresponding to the effective activity time. The first part of the extended data set is obtained through the steps, and the part of the data set is only used for training the model and is not used for testing. The second data enhancement scheme is: by finding the central point of the effective activity time, intercepting data with the length of 2 seconds as data of one batch by taking the point as the center, and offsetting one data unit to two sides each time, and obtaining one data sample by offsetting one data unit each time. The shift range interval is 2 seconds ± 16% of the data interval, resulting in the second portion of the expanded data set. The partial data set is only used for testing, so that the method is more suitable for the actual application environment and better tests the generalization of the model.
And (4) performing data feature extraction on the human motion state, specifically, taking data with the length of 2 seconds as data of one batch every time to perform feature extraction during data feature extraction. The mean, standard deviation, maximum, minimum, square and/or length of the data over 2 seconds, and pearson's coefficient were calculated, respectively. The statistics of each axis in the six-axis data with the length of 2 seconds are respectively extracted, and the statistics comprises the average value, the standard deviation, the maximum value, the minimum value, the square and/or the length corresponding to each axis, so that 30-dimensional data statistical characteristic information is obtained. Then calculating the correlation of different directions of the acceleration of the shaft for 2 seconds X, Y, Z based on the Pearson coefficient; and the correlations in different directions of the angular velocity corresponding to the axis of the gyroscope X, Y, Z, to obtain 6-dimensional data correlation characteristic information. And counting the 30-dimensional data and the 6-dimensional data correlation characteristic information to obtain 36-dimensional human motion state data characteristics in total. And finally, obtaining a data set form of 36-dimensional data feature-label, and dividing the data set into a training set and a testing set.
The method comprises the following steps of (5) constructing an algorithm model for six-axis motion state discrimination based on characteristics, wherein the six-axis motion state discrimination algorithm is realized by building a Random Forest (RF) algorithm model. The random forest algorithm mainly combines a plurality of weak classifiers, and each classifier votes to obtain a final result. The random forest is a classifier with a plurality of decision trees, each decision tree is not related, when data to be tested enters the random forest, each tree is classified, the output class of the tree is determined by the mode of the output results of some trees, and finally, the result with the most classification in all the decision trees is the final classification. Training the sample data through a random forest algorithm to obtain a motion state classifier. The specific model parameters are as follows.
Table model parameters of random forest algorithm
The method comprises the following steps of (6) training a random forest algorithm model based on a training data set, wherein the model training and identifying process for judging the six-axis motion state comprises 4 stages: preparing a data set, designing a learning model, training a model and verifying the model. The data set preparation stage comprises four steps, namely firstly, sampling six-axis motion data to generate a six-axis data file; secondly, data cleaning, data enhancement and data feature extraction are carried out on the data; and finally, repeatedly training and testing the six-axis motion state type. The design process of the model comprises the following steps: evaluating the machine learning model, selecting a basic model frame, setting hyper-parameters and the like. And then, carrying out a training process of the model, and continuously learning and optimizing parameters of the model by capturing characteristic information of the data. And (5) using the test data to verify the effect of the model and test the generalization of the model after training. And after carrying out repeated iterative training on the sample data through a random forest algorithm, finally obtaining the expected motion state classifier model.
And (7) identifying the motion state and evaluating the accuracy of the identification result based on the test data set, wherein the data set is divided into a training set and a test set, the training set is firstly input into a random forest classifier for training, and then the test set is utilized for simulation test. The recognition accuracy of the established random forest-based motion state discrimination model on the self-established six-axis motion state data set is 93.9%, and the recognition accuracy of each motion state classification is as follows.
TABLE estimation of accuracy of motion states
Drawings
Fig. 1 is a flowchart of the six-axis motion state identification process.
FIG. 2 is a block diagram of a process for training and recognition of a six-axis motion state recognition model.
FIG. 3 illustrates a Bagging structure of a random forest algorithm.
Fig. 4 is a confusion matrix of six-axis motion state data set motion state recognition results.
Detailed Description
The invention is described in detail below with reference to the figures and examples.
The motion state determination is a determination of various motion states for the motion behavior of the user. The exercise states included in the invention include sitting down, standing up, walking, running, going up stairs, going down stairs, sleeping. The six-axis sensing motion data of the user are read through hardware equipment worn by the user, and then the motion state of the user is judged.
When the motion state is judged, the motion postures are diversified due to different motion duration, and the invention discloses a wearable six-axis sensing data-based human motion state identification and analysis method. The method identifies the motion state of the user through the self-established six-axis motion state data set, is applied to analyzing the motion state of the user all day, tracks the motion characteristic change of the user, and further analyzes the physical function state of the user. A flowchart of the six-axis motion state identification process is shown in fig. 1.
The method comprises the steps that firstly, six-axis sensing data are collected on the basis of hand ring equipment (which can also be configured on the waist, the chest and other parts) worn on a body, and a data set of a motion state is established; then, carrying out operations such as cleaning, slicing and intercepting on the data to obtain a standard data set corresponding to the activity state of the human body; then, enhancing and expanding data of the standard data set through offset operation; then, taking data with the length of 2 seconds as a batch, and extracting 36-dimensional robustness characteristics with direction independence; and finally, building a motion state discrimination model based on a random forest algorithm. The method has the characteristics of compactness, quickness and the like, can be loaded on embedded hardware equipment, and accurately judges the current motion state of the human body in near real time (once in 2 seconds). The flow chart of the six-axis motion state recognition model training and recognition is shown in fig. 2. The method comprises the following specific implementation steps.
Firstly, establishing a six-axis active state acquisition data set.
The motion states to be detected include: sitting down, standing up, walking, running, going up stairs, going down stairs, sleeping, for which we define the labels as: 0,1,2,3,4,5,6. Data acquisition is performed by using six-axis sensor equipment of an LSM6DS model, sampling frequency is 10 times per second, and 3D digital accelerometers (X-axis acceleration, Y-axis acceleration and Z-axis acceleration) and 3D digital gyroscope data (X-axis angular velocity, Y-axis angular velocity and Z-axis angular velocity) are read. The obtained data is saved as a file list set in the form of acquisition device type, tag, timestamp and six-axis information. And constructing a six-axis motion state data set on the basis of the file list set.
In order to ensure the validity of the data, the data needs to be cleaned. In the process of data cleaning, dirty data in the data file needs to be filtered out, so that the effect of improving the accuracy of the model is achieved. Through statistical observation, when the userWhen the device is in a motion state, the values of the resultant acceleration and the resultant angular velocity of the data are increased sharply. Therefore, some dirty data are screened out according to whether the dirty data are smaller than the threshold value or not through the preset combined acceleration threshold value T1 and the combined angular velocity threshold value T2. The calculation formula of the resultant acceleration is as follows:
wherein
Represents the acceleration corresponding to the X-axis,
represents the acceleration corresponding to the Y-axis,
representing the acceleration corresponding to the Z-axis. And the calculation formula of the angular velocity is:
wherein
Representing the angular velocity corresponding to the X-axis,
representing the angular velocity corresponding to the Y-axis,
representing the angular velocity corresponding to the Z-axis.
And secondly, constructing a standard data set corresponding to the six-axis motion state of the human body.
And on the basis of acquiring the data set, constructing a standard data set corresponding to the six-axis motion state of the human body. The method first divides a data interval of effective activity time based on an acquired data set. The effective activity time refers to a time interval in a motion state in the data recording process. Firstly, slicing the data of the whole data file into a section of data set; then, calculating the sum of the combined accelerations of the data in each section of data, solving the section with the maximum sum of the combined accelerations, selecting the section with the maximum sum of the combined accelerations as the central position of effective data, and taking the position as the center; and finally, traversing the adjacent data segments from near to far in sequence, and if the sum of the combined accelerations of the data segments is less than a predefined threshold value T, terminating the traversal, thereby obtaining the data segments of the effective activity time in the data file.
And thirdly, performing enhancement expansion on the standard data set.
Two data enhancement schemes are employed in the data enhancement section.
The first data enhancement scheme is: and integrally translating the data part corresponding to the effective activity time in the data section where the effective activity time is located, and obtaining a new data sample every time moving one unit of data. The range interval of the movement is +/-16% L, wherein L is the data length corresponding to the effective activity time. The first part of the extended data set is obtained through the steps, and the part of the data set is only used for training the model and is not used for testing.
The second data enhancement scheme is: by finding the central point of the effective activity time, intercepting data with the length of 2 seconds as data of one batch by taking the point as the center, and offsetting one data unit to two sides each time, and obtaining one data sample by offsetting one data unit each time. The shift range interval is 2 seconds ± 16% of the data interval, resulting in the second portion of the expanded data set. The partial data set is only used for testing, so that the method is more suitable for the actual application environment and better tests the generalization of the model.
And fourthly, extracting the data characteristics of the motion state of the human body.
In the data feature extraction, data of 2 seconds length is taken as data of one batch for feature extraction each time. The mean, standard deviation, maximum, minimum, square and/or length of the data over 2 seconds, and pearson's coefficient were calculated, respectively. The calculation formula is shown in table 1 below.
Table 1 data feature extraction formula table.
Wherein
Represents a data set of one dimension of six-axis data corresponding to a length of 2 seconds.
Compressing and converting data within six-axis 2 seconds into a matrix with the dimension of 36 x 1, and specifically comprising the following steps of: the statistics of each axis in the six-axis data with the length of 2 seconds are respectively extracted, and the statistics comprises the average value, the standard deviation, the maximum value, the minimum value, the square and/or the length corresponding to each axis, so that 30-dimensional data statistical characteristic information is obtained. Then calculating the correlation of different directions of the acceleration of the shaft for 2 seconds X, Y, Z based on the Pearson coefficient; and the correlations in different directions of the angular velocity corresponding to the axis of the gyroscope X, Y, Z, to obtain 6-dimensional data correlation characteristic information. And counting the 30-dimensional data and the 6-dimensional data correlation characteristic information to obtain 36-dimensional human motion state data characteristics in total. And finally, obtaining a data set form of 36-dimensional data feature-label, and dividing the data set into a training set and a testing set.
And fifthly, constructing an algorithm model for judging the six-axis motion state based on the characteristics.
The six-axis motion state discrimination algorithm is realized by building a Random Forest (RF) algorithm model. The random forest is an Ensemble Learning algorithm (Ensemble Learning), and belongs to the Bagging type, and the Bagging structure of the random forest algorithm is shown in fig. 3. The random forest algorithm mainly combines a plurality of weak classifiers, and each classifier votes to obtain a final result. The random forest is a classifier with a plurality of decision trees, each decision tree is not related, when data to be tested enters the random forest, each tree is classified, the output class of the random forest is determined by the mode of output results of some trees, and finally, the result with the most classification in all the decision trees is the final classification. The random can make it have anti-overfitting ability, while the forest makes it more accurate, making it able to obtain good classification effect. Training the sample data through a random forest algorithm to obtain a motion state classifier. Specific model parameters are shown in table 2 below.
TABLE 2 model parameters of random forest Algorithm
And sixthly, training a random forest algorithm model based on the training data set.
And on the basis of building a six-axis motion algorithm, combining the built training data set to start model training. The model training and recognition process for judging the six-axis motion state is shown in fig. 2. The method comprises 4 stages: preparing a data set, designing a learning model, training a model and verifying the model. The data set preparation stage comprises four steps, namely firstly, sampling six-axis motion data to generate a six-axis data file; secondly, data cleaning, data enhancement and data feature extraction are carried out on the data; and finally, repeatedly training and testing the six-axis motion state type. The design process of the model comprises the following steps: evaluating the machine learning model, selecting a basic model frame, setting hyper-parameters and the like. And then, carrying out a training process of the model, and continuously learning and optimizing parameters of the model by capturing characteristic information of the data. And (5) using the test data to verify the effect of the model and test the generalization of the model after training. And carrying out repeated iterative training on the sample data through a random forest algorithm to finally obtain the expected motion state classifier model.
And seventhly, carrying out motion state identification and identification result accuracy evaluation based on the test data set.
The data set is divided into a training set and a testing set, the training set is firstly input into a random forest classifier for training, and then the testing set is used for carrying out simulation testing. The recognition accuracy of the motion state discrimination model based on the random forest on the self-built six-axis motion state data set is 93.9%, and a confusion matrix of state recognition is shown in FIG. 4. The recognition accuracy of each motion state classification is shown in table 3 below.
TABLE 3 estimation of accuracy of motion states
In summary, the invention provides a human motion state identification and analysis method based on wearable six-axis sensing data. The method is based on a six-axis motion state data set, and after cleaning, slicing interception and offset enhancement expansion are carried out on collected data, a random forest machine learning method is adopted for model network training; and then, the trained model is applied, and the motion state of the user is recognized and recorded in quasi-real time (once in 2 seconds) based on the real-time collected six-axis motion data of the bracelet. The movement state before backtracking and correction can be carried out according to the daily work and rest arrangement of the user, so that the complete and accurate movement condition of the user in one day can be obtained.
Method of the invention although specific embodiments of the invention have been disclosed for illustrative purposes and in the accompanying drawings for purposes of promoting an understanding of the principles of the invention and of its implementation, those skilled in the art will recognize that: no alterations, changes, and modifications are possible without departing from the spirit and scope of the invention, as defined in the appended claims. Therefore, the present invention should not be limited to the disclosure of the preferred embodiments and the accompanying drawings. The presently disclosed embodiments are to be considered in all respects as illustrative and not restrictive on the scope of the appended claims.
Claims (3)
1. A human motion state identification method based on wearable six-axis sensing data is characterized by comprising the following steps:
(1) establishing a six-axis active state acquisition data set; data acquisition is carried out by adopting six-axis sensor equipment of an LSM6DS model, the sampling frequency is 10 times per second, and data (X-axis acceleration, Y-axis acceleration and Z-axis acceleration) of a 3D digital accelerometer and data (X-axis angular velocity, Y-axis angular velocity and Z-axis angular velocity) of a 3D digital gyroscope are read; the motion states to be detected include: sit down, stand up, walk, run, go upstairs, go downstairs, sleep, define as the label respectively: 0,1,2,3,4,5, 6; the obtained data is stored as a file list set in the form of 'acquisition equipment type-tag-timestamp-six-axis information'; cleaning the data, and screening out some dirty data according to whether the dirty data is smaller than a preset combined acceleration threshold T1 and a preset combined angular velocity threshold T2;
(2) constructing a standard data set corresponding to the six-axis motion state of the human body; dividing a data interval of effective activity time based on the acquired data set; the effective activity time refers to a time interval in a motion state in the data recording process; firstly, slicing the data of the whole data file into a section of data set; then, calculating the sum of the combined accelerations of the data in each section of data, solving the section with the maximum sum of the combined accelerations, selecting the section with the maximum sum of the combined accelerations as the central position of effective data, and taking the position as the center; finally, traversing the adjacent data segments from near to far in sequence, and if the sum of the combined accelerations of the data segments is less than a predefined threshold value T, terminating the traversal, thereby obtaining the data segments of the effective activity time in the data file;
(3) performing enhancement expansion on the standard data set; two data enhancement schemes are respectively used for enhancing training data and testing data; the enhancement expansion of the training data is that the data part corresponding to the effective activity time is integrally translated in the data segment where the data part is located, a new data sample is obtained when a unit of data is moved, and the moving range interval is +/-16% L; the enhancement of the test data is to intercept data with the length of 2 seconds as data of a batch by finding a central point of effective activity time and taking the central point as a center, and offset one data unit to two sides every time, and obtain one data sample every time the data unit is offset; the shift range interval is a data interval of 2 seconds ± 16%;
(4) extracting data characteristics of the motion state of the human body; taking data with the length of 2 seconds as data of one batch each time for feature extraction; respectively extracting the statistics of each axis in the six-axis data with the length of 2 seconds, wherein the statistics comprises the average value, the standard deviation, the maximum value, the minimum value, the square and/or the length corresponding to each axis; obtaining 30-dimensional data statistical characteristic information; then calculating the correlation of different directions of the acceleration of the shaft for 2 seconds X, Y, Z based on the Pearson coefficient; and the correlations of different directions of the angular velocity corresponding to the axis of the gyroscope X, Y, Z, so as to obtain 6-dimensional data correlation characteristic information; finally, synthesizing to obtain 36-dimensional human motion state data characteristics;
(5) constructing an algorithm model for judging the six-axis motion state based on the characteristics; the six-axis motion state discrimination algorithm is realized by building a Random Forest (RF) algorithm model; mainly combining a plurality of weak classifiers, and voting by each classifier to obtain a final result; training sample data through a random forest algorithm to obtain a motion state classifier;
(6) training a random forest algorithm model based on a training data set; the model training and recognition process includes 4 stages: preparing a data set, designing a learning model, training a model and verifying the model; wherein the data set preparation phase comprises four steps: sampling six-axis motion data, cleaning data, enhancing data and extracting data characteristics; the design process of the model comprises the following steps: evaluating a machine learning model, selecting a basic model frame, setting hyper-parameters and the like; then, performing a training process of the model, continuously learning by capturing characteristic information of data and optimizing parameters of the model;
(7) performing motion state identification and identification result accuracy evaluation based on the test data set; firstly, inputting a training set into a random forest classifier for training, and then carrying out simulation test by using a test set; the accuracy of recognition of the established random forest-based motion state discrimination model on the self-established six-axis motion state data set is 93.9%.
2. The human motion state identification method based on the wearable six-axis sensing data according to claim 1, characterized in that an enhanced extension scheme is performed on a standard data set;
the first data enhancement scheme is: integrally translating the data part corresponding to the effective activity time in the data section where the data part is located, and obtaining a new data sample when moving one unit of data; the range interval of the movement is +/-16% L, wherein L is the data length corresponding to the effective activity time; obtaining a first part of extended data set through the steps, wherein the part of data set is only used for training the model and is not used for testing;
the second data enhancement scheme is: intercepting data with the length of 2 seconds as data of a batch by taking a central point of effective activity time as the center, and offsetting one data unit to two sides every time to obtain one data sample by offsetting one data unit; the deviation range interval is 2 seconds +/-16% of the data interval, so that the second part of the expanded data set is obtained; the partial data set is only used for testing, so that the method is more suitable for the actual application environment and better tests the generalization of the model.
3. The human motion state identification method based on the wearable six-axis sensing data according to claim 1, characterized in that data feature extraction is performed on the human motion state; during data feature extraction, data with the length of 2 seconds are taken as data of one batch every time for feature extraction; calculating the average value, standard deviation, maximum value, minimum value, square and/or length of the data in 2 seconds and the Pearson coefficient respectively; respectively extracting the statistics of each axis in the six-axis data with the length of 2 seconds, wherein the statistics comprises the average value, the standard deviation, the maximum value, the minimum value, the square and/or the length corresponding to each axis, and therefore 30-dimensional data statistical characteristic information is obtained; then calculating the correlation of different directions of the acceleration of the shaft for 2 seconds X, Y, Z based on the Pearson coefficient; and the correlations of different directions of the angular velocity corresponding to the axis of the gyroscope X, Y, Z, so as to obtain 6-dimensional data correlation characteristic information; counting the 30-dimensional data according to the characteristic information and obtaining 6-dimensional data correlation characteristic information to obtain 36-dimensional human motion state data characteristics in total; and finally, obtaining a data set form of 36-dimensional data feature-label, and dividing the data set into a training set and a testing set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110325902.2A CN113095379A (en) | 2021-03-26 | 2021-03-26 | Human motion state identification method based on wearable six-axis sensing data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110325902.2A CN113095379A (en) | 2021-03-26 | 2021-03-26 | Human motion state identification method based on wearable six-axis sensing data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113095379A true CN113095379A (en) | 2021-07-09 |
Family
ID=76670092
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110325902.2A Pending CN113095379A (en) | 2021-03-26 | 2021-03-26 | Human motion state identification method based on wearable six-axis sensing data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113095379A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114440884A (en) * | 2022-04-11 | 2022-05-06 | 天津果实科技有限公司 | Intelligent analysis method for human body posture for intelligent posture correction equipment |
-
2021
- 2021-03-26 CN CN202110325902.2A patent/CN113095379A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114440884A (en) * | 2022-04-11 | 2022-05-06 | 天津果实科技有限公司 | Intelligent analysis method for human body posture for intelligent posture correction equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8935195B2 (en) | Method of identification and devices thereof | |
| Khan et al. | Accelerometer signal-based human activity recognition using augmented autoregressive model coefficients and artificial neural nets | |
| CN104268577B (en) | Human body behavior identification method based on inertial sensor | |
| CN105678222B (en) | A kind of mobile device-based Human bodys' response method | |
| CN110245718A (en) | A kind of Human bodys' response method based on joint time-domain and frequency-domain feature | |
| Uddin et al. | Random forests based recognition of human activities and postural transitions on smartphone | |
| CN106971059A (en) | A kind of wearable device based on the adaptive health monitoring of neutral net | |
| CN108446733A (en) | A kind of human body behavior monitoring and intelligent identification Method based on multi-sensor data | |
| Tian et al. | MEMS-based human activity recognition using smartphone | |
| CN103970271A (en) | Daily activity identifying method with exercising and physiology sensing data fused | |
| CN106228200A (en) | A kind of action identification method not relying on action message collecting device | |
| Kawsar et al. | A novel activity detection system using plantar pressure sensors and smartphone | |
| Jalloul et al. | Activity recognition using complex network analysis | |
| EP3459453B1 (en) | Information processing device, information processing method, and information processing program | |
| CN111401435B (en) | Human body motion mode identification method based on motion bracelet | |
| CN108958482A (en) | A kind of similitude action recognition device and method based on convolutional neural networks | |
| CN114881079A (en) | Human body movement intention abnormity detection method and system for wearable sensor | |
| Mekruksavanich et al. | Detection of freezing of gait in parkinson's disease by squeeze-and-excitation convolutional neural network with wearable sensors | |
| CN113095379A (en) | Human motion state identification method based on wearable six-axis sensing data | |
| CN112487902B (en) | Exoskeleton-oriented gait phase classification method based on TCN-HMM | |
| CN109271889A (en) | A kind of action identification method based on the double-deck LSTM neural network | |
| KR20210046121A (en) | Apparatus and method for identify patients with parkinson's disease and patients with podarthritis by performing neural network analysis by various detection information | |
| KR101483218B1 (en) | Activity diagnosis apparatus | |
| Eskaf et al. | Aggregated activity recognition using smart devices | |
| CN115336979A (en) | Multitask tremor automatic detection method based on wearable device and detection device thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |