CN109805935A - A kind of intelligent waistband based on artificial intelligence hierarchical layered motion recognition method - Google Patents
A kind of intelligent waistband based on artificial intelligence hierarchical layered motion recognition method Download PDFInfo
- Publication number
- CN109805935A CN109805935A CN201711162489.2A CN201711162489A CN109805935A CN 109805935 A CN109805935 A CN 109805935A CN 201711162489 A CN201711162489 A CN 201711162489A CN 109805935 A CN109805935 A CN 109805935A
- Authority
- CN
- China
- Prior art keywords
- motion
- module
- link
- data
- acceleration
- 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
- 230000033001 locomotion Effects 0.000 title claims abstract description 133
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 16
- 230000009467 reduction Effects 0.000 claims abstract description 22
- 230000008569 process Effects 0.000 claims abstract description 18
- 230000003044 adaptive effect Effects 0.000 claims abstract description 3
- 230000001133 acceleration Effects 0.000 claims description 44
- 230000005021 gait Effects 0.000 claims description 39
- 239000011159 matrix material Substances 0.000 claims description 26
- 238000012549 training Methods 0.000 claims description 21
- 239000013598 vector Substances 0.000 claims description 21
- 238000001914 filtration Methods 0.000 claims description 15
- 239000002131 composite material Substances 0.000 claims description 12
- 230000003068 static effect Effects 0.000 claims description 11
- 238000012795 verification Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000000354 decomposition reaction Methods 0.000 claims description 4
- 230000035772 mutation Effects 0.000 claims description 4
- 230000006978 adaptation Effects 0.000 claims description 3
- 230000001575 pathological effect Effects 0.000 claims description 3
- 238000010415 tidying Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 abstract description 23
- 238000005070 sampling Methods 0.000 abstract description 8
- 238000007781 pre-processing Methods 0.000 abstract description 4
- 238000006243 chemical reaction Methods 0.000 abstract description 2
- 230000004069 differentiation Effects 0.000 abstract 1
- 230000006870 function Effects 0.000 description 10
- 230000036541 health Effects 0.000 description 9
- 230000006399 behavior Effects 0.000 description 8
- 230000009471 action Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 230000000276 sedentary effect Effects 0.000 description 6
- 208000006011 Stroke Diseases 0.000 description 5
- 238000001514 detection method Methods 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 230000002526 effect on cardiovascular system Effects 0.000 description 3
- 238000005265 energy consumption Methods 0.000 description 3
- 230000005484 gravity Effects 0.000 description 3
- 210000003141 lower extremity Anatomy 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 208000008589 Obesity Diseases 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000002612 cardiopulmonary effect Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 210000000629 knee joint Anatomy 0.000 description 2
- 210000002414 leg Anatomy 0.000 description 2
- 235000020824 obesity Nutrition 0.000 description 2
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 206010008190 Cerebrovascular accident Diseases 0.000 description 1
- 208000004067 Flatfoot Diseases 0.000 description 1
- 206010061159 Foot deformity Diseases 0.000 description 1
- 206010028347 Muscle twitching Diseases 0.000 description 1
- 206010044565 Tremor Diseases 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002490 cerebral effect Effects 0.000 description 1
- 208000026106 cerebrovascular disease Diseases 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 210000004072 lung Anatomy 0.000 description 1
- 208000030159 metabolic disease Diseases 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 235000020030 perry Nutrition 0.000 description 1
- 210000000707 wrist Anatomy 0.000 description 1
Landscapes
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The present invention discloses a kind of intelligent waistband based on artificial intelligence hierarchical layered motion recognition method, comprising: motion state solves module, movement link judgment module, dimensionality reduction and solves module, physiological parameter setting module, physiological parameter acquisition module；The present invention analyzes the comprehensive different characteristic dimensionality reduction sampling of link, on the one hand subtle motion feature is both acquired comprehensively in preprocessing process, on the other hand further through adaptive algorithmic match, limited different sensors function operation is only called in different specific differentiation processes, extract different specific feature calculations, the comprehensive different characteristic dimensionality reduction sampling of link is analyzed, calculation amount is greatly reduced；Emphasis verifying link is placed on cloud simultaneously, remaining calculating is placed on wearable device end, effectively balances the contradiction of fast reaction and calculation amount, improves the medical reference value of user's physiological parameter obtained, and power-consumption standby ability is extracted under square one.
Description
Technical Field
The invention relates to the field of intelligent wearing, in particular to an intelligent waistband based on an artificial intelligence layered and graded motion recognition method.
Background
Regular movement can improve the functions of heart and lung, and reduce the incidence of cardiovascular and cerebrovascular diseases, fat metabolic disorder and other diseases; for some special cases, the physiological parameters may be acquired during a particular gait cycle. For example, the prediction and diagnosis of stroke (stroke) and parkinson are related to the behavior of the action state in different steps of the gait cycle. Therefore, the motion state, period and motion fine characteristics need to be acquired finely.
More and more wearable equipment possesses the meter step function at present, through step rough estimation energy consumption, distance, can't further discern user's activity, and this kind of limitation causes the problem in two aspects:
1) effective exercises cannot be distinguished: effective exercise is an important way for improving the cardio-pulmonary function, enhancing strength and flexibility and improving the health level of people. For example, in the case of a user performing a strenuous activity such as tennis or basketball, the statistics of the existing devices are intermittent steps, and the calculated energy consumption is far below the actual level.
2) Sedentary immobility recognition error: sedentary immobility generally refers to standing still for more than 1 hour, which can cause cardiovascular, cervical and lumbar diseases; studies have shown that the longer the sitting time, the higher the risk of obesity and death; the conventional pedometer may erroneously recognize a non-counting situation such as a long-time riding or riding as sedentary.
Some equipment can carry out the division of motion state, through the operation of detailed user activity recognition model, carries out motion state's discernment, but this kind of motion state's discernment must bring the increase of the technical data volume of operation, if accomplish the calculation at wearable intelligent terminal (like bracelet, wrist-watch, waistband), must increase the consumption of electric quantity. Due to constraints on volume, weight, and system performance, power supply capacity is one of the most important factors that limit widespread use. At present, wearable power consumption standby time is very limited originally, the detailed characteristics of the motion state are comprehensively collected in the prior art, the calculation amount and power consumption are large, and the cruising ability of the wearable equipment is reduced. If the calculation process is placed in the cloud, the increase of data transmission bandwidth and judgment time delay is inevitably brought, so that the timeliness of some movement responses needing to be judged quickly is influenced.
In conclusion, when the physiological parameters are collected by the existing wearable intelligent equipment, the motion state of the user is not considered, the motion state collection such as simple step counting is only carried out, and the difference of motion links and motion subtle characteristics is not collected. Or, the comprehensive collection aggravates the power consumption of the wearable terminal or increases the data transmission bandwidth and the judgment delay.
Disclosure of Invention
In view of the above defects in the prior art, the technical problem to be solved by the present invention is to provide an intelligent belt based on an artificial intelligence hierarchical motion recognition method, aiming to solve the problem that the prior art does not consider the motion state of the user when acquiring physiological parameters; the invention identifies the movement links and movement subtle characteristics and acquires corresponding physiological parameter information, thereby facilitating the analysis of the health state of the user by doctors or other personnel. Meanwhile, the motion state is acquired in a grading mode, the data processing operation amount is reduced, the power consumption of the wearable equipment is reduced, and the cruising ability is improved.
In order to achieve the above object, the present invention provides an intelligent waistband based on an artificial intelligence hierarchical motion recognition method, comprising:
the physiological parameter setting module is used for setting physiological parameters to be acquired under different motion states and gaits;
and the physiological parameter acquisition module is used for acquiring corresponding physiological parameters under the appointed motion state and gait according to the physiological parameters required to be acquired.
The motion state solving module is used for acquiring acceleration data output by the acceleration sensor at regular time, filtering high-frequency noise through wavelet transformation and dividing the motion state of a user;
the motion link judgment module is used for subdividing the motion state by combining the angular speed to obtain a motion subdivision link;
the dimensionality reduction solving module is used for further finely distinguishing partial pathological features by using the optimized SVM hierarchical level and combining different hierarchical data;
in the above modules, the physiological parameter setting module is completed through interaction of a user smart phone and a cloud end, the motion state solving module and the motion link judging module are completed at the wearable device end, and the dimensionality reduction solving module is completed at the cloud end.
Further, the motion segment determination module is configured to:
acquiring acceleration values a of the acceleration sensor in the directions of the x, y and z axesx、ay、azAnd solving the acceleration signal vector modulus SVMA(ii) a Collecting angular velocity values w of the angular velocity sensor in the directions of three axes x, y and zx、wy、wzAnd solving the vector modulus SVM of the angular velocity signalW(ii) a The above-mentioned
Vector modulo SVM by acceleration signalAVector mode SVM for angular velocity signalWAnd establishing and self-adaptively completing an identification model, and subdividing the motion state to obtain a motion subdivision link.
Furthermore, the motion link judgment module further comprises a filtering unit, wherein the filtering unit performs wavelet transformation operation of three steps of wavelet decomposition, high-frequency wavelet coefficient processing and wavelet reconstruction on the acquired data of the acceleration sensor and the angular velocity sensor, discretizes time domain signals in each direction, decomposes mixed signals of various frequency components into different frequency bands, and then processes the mixed signals according to different characteristics of each seed signal in the frequency domain according to frequency bands; acquiring gait data with high signal to noise ratio;
the wavelet transform adopts a hard threshold method, and the wavelet coefficient is Cj,kThe threshold is lambda;
the above-mentioned
Further, the dimension reduction solving module is configured to:
acquiring physiological parameter data of a user, and reducing the dimension of the physiological parameter data of the user;
taking a standard deviation of the whole sample population, taking N as a sample size, training a classifier, and identifying a step sample by using the classifier;
comprehensively calculating the degree x of deviation of certain gait from the tidying crowd, wherein
Wherein, ai、bi、ciAcceleration in the x, y, z axis directions respectively,the acceleration of the whole human group in the directions of the x axis, the y axis and the z axis of a certain gait link is respectively.
Inputting the personal gait samples of N types registered in the database into a classifier for training, judging which type is (1, N) according to an input value, if the input value exceeds the range of (1, N), newly registering the type of N +1, and then updating the classifier again;
and on the basis of the different motion division, the same motion is subdivided again, and a classification result is determined by adopting a voting mode.
Further, the dimension reduction solving module is further configured to:
the raw data is normalized by subtracting the mean of the column from each element in the proof and dividing by the standard deviation of the column, such that each variable is normalized to a matrix X with a mean of 0 and a variance of 1, i.e., X ═ X1，X2，......Xn]T＝[Xij](n×p)；
Wherein,i＝1，2…n，j＝1，2…p；
to obtain
Solving a correlation coefficient matrix:
where R is a real symmetric matrix (i.e., R)ij＝rji) Wherein r is a correlation coefficient;
solving a correlation coefficient matrix:
if the accumulated contribution rate reaches more than 50%, reserving the characteristic value vector of the highest contribution rate as a fixed working sample set by adopting a ratio-height method, and discarding the rest;
calculating a score matrix, leavingThe characteristic value of the lower matrix is used as a new variable principal component, and a partial matrix F is obtained by calculation according to the following formula(n×m)＝X(n×p)·U(p×m)；
Wherein X is an original data matrix, U is a principal component load, and a score matrix F is a result obtained after dimensionality reduction;
continuous training subdivision is performed in combination with big data: fitness function f (x) of SVM classifieri)＝min(1-g(xi))，And dividing the sample into correct rates for the SVM classifier.
Further, the dimension reduction solving module is further configured to:
calculating a first condition, a second condition and a third condition by using the output data of the acceleration sensor, and judging the gait of the human motion by using median filtering;
the first condition is: the accelerometer outputs a synthesized amplitude value, and if the synthesized amplitude value is between the given upper threshold value and the given lower threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
the accelerometer output composite amplitude is:
the upper and lower thresholds are respectively: th (h)amin＝8m/s，thamax＝11m/s；
The first condition is expressed as:
the second condition is: if the local variance output by the accelerometer is lower than a given threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
the above-mentionedThe local variance of the accelerometer output is:
whereinFor the interval, the accelerometer synthesizes an output average value of the amplitude, and the expression is as follows:
s is the number of half-window samples, and a given threshold is defined as: th (h)σa＝0.5m/s2The second condition is represented as:
the third condition is: the angular velocity sensor outputs an angular velocity composite amplitude, and if the angular velocity composite amplitude is lower than a given threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
define the composite amplitude of the gyroscope output as:
the given thresholds are: th (h)wmaxThe third condition is expressed as:
further, the intelligent waistband based on the artificial intelligence layered and graded motion recognition method further comprises a model self-adaptive perfecting module; the model adaptation refinement module is configured to:
reading a new input sample, and calculating the recognition rate of the SVM classifier according to a cross verification method;
if the current recognition rate of the training is higher than or equal to the original recognition rate, setting the parameters of the training as the optimal parameters; otherwise, a selection operation, a crossover operation and/or a mutation operation are executed, and the training parameters are further optimized.
The invention has the beneficial effects that: the scheme overcomes the two contradictions, on one hand, fine motion characteristics are comprehensively collected in the preprocessing process, on the other hand, only limited different sensor functions are called in different specific judging processes through self-adaptive algorithm matching, different specific characteristic calculation is extracted, different characteristic dimension reduction sampling is integrated in an analyzing link, the amount of exercise is greatly reduced, and the accuracy is guaranteed through a key calculation verification link. Therefore, the detailed characteristics of the motion state are comprehensively collected, the calculated amount and the power consumption are greatly reduced, the medical reference value is improved, and the power consumption standby capability is extracted under the same condition.
Drawings
FIG. 1 is a block diagram of the architecture of an embodiment of the present invention;
fig. 2 is a schematic diagram of human motion hierarchy division in the present embodiment;
FIG. 3 is a schematic view of the user walking geometry in the present embodiment;
FIG. 4 is a schematic diagram of three condition gait detection in the present embodiment;
fig. 5 is a schematic flow chart of adaptive refinement of the model in this embodiment.
Detailed Description
The invention is further illustrated by the following examples in conjunction with the accompanying drawings:
when current wearable smart machine gathered physiological parameters, or do not consider user's motion state, only carry out motion state collection such as simple meter step, do not gather to the difference of motion link, motion subtle characteristics, this kind of limitation causes the problem in two aspects:
1) effective exercises cannot be distinguished: effective exercise is an important way for improving the cardio-pulmonary function, enhancing strength and flexibility and improving the health level of people. For example, in the case of a user performing a strenuous activity such as tennis or basketball, the statistics of the existing devices are intermittent steps, and the calculated energy consumption is far below the actual level.
2) Sedentary immobility recognition error: sedentary immobility generally refers to standing still for more than 1 hour, which can cause cardiovascular, cervical and lumbar diseases; studies have shown that the longer the sitting time, the higher the risk of obesity and death; the conventional pedometer may erroneously recognize a non-counting situation such as a long-time riding or riding as sedentary.
Or, the motion state data is comprehensively collected, the motion state can be divided, the motion state is recognized through the detailed operation of the user activity recognition model, the technical data amount of the operation is inevitably increased due to the motion state recognition, and if the wearable intelligent terminal (such as a bracelet, a watch and a belt) completes the calculation, the electric quantity consumption is inevitably increased. Due to constraints on volume, weight, and system performance, power supply capacity is one of the most important factors that limit widespread use. At present, wearable power consumption standby time is very limited originally, the detailed characteristics of the motion state are comprehensively collected in the prior art, the calculation amount and power consumption are large, and the cruising ability of the wearable equipment is reduced. If the calculation process is placed in the cloud, the increase of data transmission bandwidth and judgment time delay is inevitably brought, so that the timeliness of some movement responses needing to be judged quickly is influenced.
The scheme overcomes the two contradictions, on one hand, fine motion characteristics are comprehensively collected in the preprocessing process, on the other hand, only limited different sensor functions are called in different specific judging processes through self-adaptive algorithm matching, different specific characteristic calculation is extracted, different characteristic dimension reduction sampling is integrated in an analyzing link, the calculated amount is greatly reduced, and the accuracy is ensured through a key calculation verification link; meanwhile, the key verification link is placed at the cloud end, and the rest of calculation is placed at the wearable device end, so that the contradiction between quick response and calculation amount is effectively balanced, the accuracy is ensured, and the calculation amount of the wearable device end is not too large.
Thereby, the following steps are achieved: the detailed characteristics of the motion state are comprehensively collected, the calculated amount and the power consumption are greatly reduced, the medical reference value is improved, and the power consumption standby capability is extracted under the same condition.
As shown in fig. 1, in a first embodiment of the present invention, an intelligent waistband based on an artificial intelligence hierarchical motion recognition method is provided, which includes:
the motion state solving module is used for acquiring acceleration data output by the acceleration sensor at regular time, filtering high-frequency noise through wavelet transformation and dividing the motion state of a user, wherein the motion state comprises a first cyclic action and a second non-cyclic action; the first cyclical action comprises walking and running; the second non-cyclic action comprises jumping, standing up, sitting down and squatting down; the walking comprises going upstairs, going downstairs and walking on flat ground;
the motion link judgment module is used for subdividing the motion state by combining the angular speed to obtain a motion subdivision link;
the dimensionality reduction solving module is used for further finely distinguishing partial pathological features by using the optimized SVM hierarchical level and combining different hierarchical data;
the physiological parameter setting module is used for setting physiological parameters to be acquired under different motion states and gaits;
and the physiological parameter acquisition module is used for acquiring corresponding physiological parameters under the appointed motion state and gait according to the physiological parameters required to be acquired.
For the application of fall detection and common human behavior analysis, the acceleration sensor can well distinguish the motion and static states of human behaviors, similar motion behaviors are difficult to distinguish, and the characteristics are that the distinguishing degree is large and the confusion degree is also large. The wrist angular velocity is combined to further distinguish, but the acceleration and the angular velocity are calculated simultaneously, the calculation amount is large, the timeliness and the electric quantity power consumption are affected, and therefore the calculation amount and the accuracy are considered by adopting a mode of hierarchical classification, dimension reduction classification and key verification. As shown in fig. 2, the human motion may be divided into different levels.
In this embodiment, the motion state solving module firstly uses the acceleration sensor data to perform a first-level motion judgment on the intelligent belt triaxial acceleration sensor, and collects the acting forces in the x, y and z directions according to a sampling frequency of 100Hz (the walking frequency of a person is generally 110 steps/minute (1.8Hz), the running frequency does not exceed 5Hz, and the sampling frequency of 100Hz can accurately reflect the acceleration change.
In this embodiment, the motion state solving module counts the frequency of occurrence of the peak through the pair of trajectories secondly. In horizontal movement of the user, the vertical and forward accelerations may exhibit periodic variations. In the walking and foot-receiving action, the gravity center is upward, and only one foot touches the ground, the vertical acceleration tends to increase in a positive direction, then the gravity center is moved downwards, and the two feet touch the bottom, and the acceleration is opposite. The horizontal acceleration decreases when the foot is retracted and increases when the stride is taken, as shown in fig. 3.
It is worth mentioning that in walking exercise, the acceleration generated by vertical and forward motion is approximately sinusoidal with time and has a peak at a certain point where the acceleration in the vertical direction changes most.
And finally, the motion state solving module is used for filtering the data. Because the electromagnetic interference in the circuit is a main interference source in the acquisition process, the electromagnetic interference is high-frequency noise; the human motion is mainly low-frequency signals within 50Hz, and the wavelet transform threshold method is selected. For such interference, we add a threshold and a step frequency decision to the detection to filter, that is, the time interval between two adjacent steps is at least greater than 0.2 seconds, and filter high frequency noise.
In this embodiment, the motion segment determining module is configured to:
acquiring acceleration values a of the acceleration sensor in the directions of the x, y and z axesx、ay、azAnd solving the acceleration signal vector modulus SVMA(ii) a Collecting angular velocity values w of the angular velocity sensor in the directions of three axes x, y and zx、wy、wzAnd solving the vector modulus SVM of the angular velocity signalW(ii) a The above-mentioned
Vector modulo SVM by acceleration signalAVector mode SVM for angular velocity signalWAnd establishing and self-adaptively completing an identification model, and subdividing the motion state to obtain a motion subdivision link.
Specifically, the motion link judgment module is used for subdividing motion links and detecting falling by combining angular speed.
Since the acceleration is suitable for the judgment of the motion with definite direction, the judgment of the incapability of falling detection, motion cycle links, splayfoot and the like needs to be carried out by using the angular velocity.
Based on the principle of a kinematic algorithm, four gait event time phases are detected: a gait cycle is divided into two phases, a "support phase" and a "swing phase".
Perry doctors at the national rehabilitation center for RLA, California, USA put forward the RLA staging method according to the occurrence sequence of the walking cycle; the support period is divided into 5 stages; the stride period is broken down into 3 epochs.
1. First touchdown, initialcontact: as the starting point of the walking cycle and the support period; the moment when the heel or other parts of the sole of the foot first contact the ground. The first landing mode for normal people walking is heel landing.
2. Load bearing reaction period, loadingresponse: the foot bottom is in full contact with the ground for a period of time after the heel is grounded; namely, when the heel at one side is grounded and the toe of the lower limb at the opposite side is lifted off; is the process of transferring the center from the heel to the sole. This phase is 0-15% of the gait cycle.
3. Mid-stance, mid-stance: when the finger is lifted from the lower limb at the opposite side to the trunk right above the leg at the side; the center of gravity is now directly above the support surface. This period is 15% -40% of the gait cycle:
4. late stance, terminalstance: the fingers are from the time the support heel lifts off to the time the contralateral lower limb heel lands. This period is 40% -50% of the gait cycle.
5. In the early stage of stepping, pre-swing: the fingers are held for a period of time from heel-strike of the contralateral lower limb until toe-off support. This period is 50% -60% of the gait cycle.
6. Initial step, initial stroke: from the point at which the supporting leg lifts to the point at which the knee joint reaches maximum flexion. This period is 60% -70% of the gait cycle.
7. Mid-swing: from the maximum flexion swing of the knee joint to when the lower leg is perpendicular to the ground. This period is 70% -85% of the gait cycle.
8. At the end of the step, terminating: the lower leg, which is perpendicular to the ground, swings forward until the heel lands again. This period is 85% -100% of the gait cycle.
In this embodiment, the links may be divided by time in the exercise cycle, and certainly, each exercise link may be determined by artificially setting the time of each link.
In the present embodiment, an acceleration signal vector mode and an angular velocity signal vector mode are used as input features of the model.
The analysis module builds a recognition model with perfect self-adaptation, mainly carries out modeling through an acceleration sensor rule, gives a motion state recognition result through the operation of a genetic operator, and is used for analyzing and managing remote health big data.
The handheld device has a low amplitude and a quick twitching state, or what is commonly called hand trembling, or a mischief user wants to simulate walking by quickly and repeatedly shaking the device for a short time, and the accurate value of step counting can be influenced if the interference data are not eliminated.
Carrying out wavelet transformation operations of three steps of wavelet decomposition, high-frequency wavelet coefficient processing and wavelet reconstruction on the collected data in each direction, discretizing time domain signals in each direction, decomposing mixed signals of various frequency components into different frequency bands, and then processing according to different characteristics of each seed signal on a frequency domain and frequency bands; and acquiring gait data with high signal to noise ratio.
Falls are characterized by large acceleration and peak angular velocity because the SVM peak due to collisions with low-lying objects during falls is larger than most common procedures of walking, ascending stairs, etc. in daily activities. However, the process of the human body movement behavior has complexity and randomness, and great misjudgment can be brought by using single acceleration related information to judge the occurrence of the human body falling behavior. Information thresholding using a combination of SVMA and SVMW can distinguish between falls and low intensity motions that produce smaller SVM peaks. Through analyzing the experimental result data SVMA and SVMW in the human body falling process and other daily life behavior processes, the acceleration signal vector modulus threshold value for identifying falling is SVMAT =20m/s2, and the angular velocity signal vector modulus threshold value is SVMWT =4 rad/s.
In this embodiment, the motion link determination module further includes a filtering unit, where the filtering unit performs wavelet transform operations of three steps of wavelet decomposition, high-frequency wavelet coefficient processing, and wavelet reconstruction on the acquired data of the acceleration sensor and the angular velocity sensor, discretizes time domain signals in each direction, decomposes mixed signals of multiple frequency components into different frequency bands, and then processes the mixed signals according to different characteristics of each seed signal in the frequency domain and according to frequency bands; acquiring gait data with high signal to noise ratio;
the wavelet transform adopts a hard threshold method, and the wavelet coefficient is Cj,kThe threshold is lambda;
the above-mentioned
In this embodiment, the dimension reduction solving module is configured to:
acquiring physiological parameter data of a user, and reducing the dimension of the physiological parameter data of the user;
taking a standard deviation of the whole sample population, taking N as a sample size, training a classifier, and identifying a step sample by using the classifier;
comprehensively calculating the degree x of deviation of certain gait from the tidying crowd, wherein
Wherein, ai、bi、ciAcceleration in the x, y, z axis directions respectively,the acceleration of the whole human group in the directions of the x axis, the y axis and the z axis of a certain gait link is respectively.
Inputting the personal gait samples of N types registered in the database into a classifier for training, judging which type is (1, N) according to an input value, if the input value exceeds the range of (1, N), newly registering the type of N +1, and then updating the classifier again;
and on the basis of the different motion division, the same motion is subdivided again, and a classification result is determined by adopting a voting mode.
Such as: and the center-to-center distance of the crowd boundary is characterized in that the definition of the center-to-center distance of the boundary is the distance from the boundary point to the centroid.
In this embodiment, in order to eliminate the influence of different dimensions and different orders of magnitude between data, the multidimensional signal needs to normalize the raw data to make it comparable, and each variable is normalized to a matrix X with a mean value of 0 and a variance of 1 by subtracting the mean value of the column from each element in the proof and then dividing by the standard deviation of the column.
In this embodiment, the dimension reduction solving module is further configured to:
the raw data is normalized by subtracting the mean of the column from each element in the proof and dividing by the standard deviation of the column, such that each variable is normalized to a matrix X with a mean of 0 and a variance of 1, i.e., X ═ X1，X2，......Xn]T＝[Xij](n×p)；
Wherein,i＝1，2…n，j＝1，2…p；
to obtain
Solving a correlation coefficient matrix:
where R is a real symmetric matrix (i.e., R)ij＝rji) Wherein r is a correlation coefficient;
solving a correlation coefficient matrix:
if the accumulated contribution rate reaches more than 50%, reserving the characteristic value vector of the highest contribution rate as a fixed working sample set by adopting a ratio-height method, and discarding the rest;
calculating a partial matrix, using the remained characteristic value as a new variable principal component, and calculating the partial matrix F by using the following formula(n×m)＝X(n×p)·U(p×m)；
Wherein X is an original data matrix, U is a principal component load, and a score matrix F is a result obtained after dimensionality reduction;
the method not only uses the characteristic of high calculation speed of the fixed working sample set method, but also avoids the problems that the number of the vectors exceeds the scale of the working sample set, and the algorithm only optimizes one part of the support vectors and has range limitation. The abnormal people are selected by the method.
Continuous training subdivision is performed in combination with big data: fitness function f (x) of SVM classifieri)＝min(1-g(xi))，And dividing the sample into correct rates for the SVM classifier.
As shown in fig. 4, in the present embodiment, the gait of the movement of the human body can be effectively determined by using the output data of the accelerometer, using a three-condition (C1, C2 and C3) determination algorithm, and using a median filtering method, where the state "0" represents movement and the state "1" represents rest.
In this embodiment, the dimension reduction solving module is further configured to:
calculating a first condition, a second condition and a third condition by using the output data of the acceleration sensor, and judging the gait of the human motion by using median filtering;
the first condition is: the accelerometer outputs a synthesized amplitude value, and if the synthesized amplitude value is between the given upper threshold value and the given lower threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
the accelerometer output composite amplitude is:
the upper and lower thresholds are respectively: th (h)amin＝8m/s，thamax＝11m/s；
The first condition is expressed as:
the second condition is: if the local variance output by the accelerometer is lower than a given threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
the local variance of the accelerometer output is:
whereinFor the interval, the accelerometer synthesizes an output average value of the amplitude, and the expression is as follows:
s is the number of half-window samples, which is typically defined to have a value of 15. The given threshold is defined as: th (h)σa＝0.5m/s2The second condition is represented as:
the third condition is: the angular velocity sensor outputs an angular velocity composite amplitude, and if the angular velocity composite amplitude is lower than a given threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
define the composite amplitude of the gyroscope output as:
the given thresholds are: th (h)wmaxThe third condition is expressed as:
and logic is adopted among the 3 conditions, namely the gait is considered to be in an absolute static state only when the judgment results of the 3 conditions are all 1. And then, by a median filtering method, the continuously output motion result can be effectively judged, noise points are eliminated, and effective and reasonable gait detection data are obtained.
The intelligent waistband based on the artificial intelligence hierarchical motion recognition method further comprises a model self-adaptive perfecting module; the model adaptation refinement module is configured to:
reading a new input sample, and calculating the recognition rate of the SVM classifier according to a cross verification method;
if the current recognition rate of the training is higher than or equal to the original recognition rate, setting the parameters of the training as the optimal parameters; otherwise, a selection operation, a crossover operation and/or a mutation operation are executed, and the training parameters are further optimized.
Specifically, with the increase of the sample size, the SVM classifier can be adaptively and continuously optimized and perfected:
(1) sampling calculation of motion state
During the judgment, the standard deviation of one state is larger, and the standard deviation of the other state is smaller and exactly balanced, so that no abnormality is found, and random sampling verification is performed again.
And (4) inputting a new sample every time, and calculating the recognition rate of the SVM classifier according to the cross verification method principle.
(2) And for the characteristic values of the samples which are not found to be abnormal, using an SVM classifier fitness function to divide the accuracy of the samples for the SVM classifier. The parallel execution process is simulated by maintaining a plurality of groups and appropriately controlling the interaction between the groups, thereby improving the execution efficiency of the algorithm even without using a parallel computer.
As shown in fig. 5, each time a new sample is input, the recognition rate of the SVM classifier is calculated according to the principle of the cross validation method, fitness evaluation is performed, the termination value of the genetic algorithm is not set, the termination condition adopts a proportion method, if the recognition rate of training is higher than the existing one, the training parameter is set as the optimal parameter, otherwise, the training parameter is further optimized by performing operations such as selection, crossing and mutation.
In sum, wearable health intelligenceWaist beltThe health sign data of collection only refer to as health analysis in the industry at present, do not have the extensive domestic medical instrument of traditional non-wearing formula of replacing and popularize fast, have two important reasons:
1. the accuracy problem is as follows: the acquisition of many health signs has higher requirements on the state environment of an acquirer, for example, the acquisition of an electrocardiogram requires the acquisition of a human body in a calm state, and if the human body is required to rest for 10 minutes after exercise, the acquired heart rate data has medical reference value; the exercise heart rate data measured 3 minutes after the exercise starts have medical reference value; the prediction and diagnosis of stroke (cerebral apoplexy), Parkinson and the like are related to the behavior and other conditions of the action states of different links of the gait cycle. Therefore, the motion state, period and motion fine characteristics need to be acquired finely.
The wearable bracelet of industry can only carry out motion state collection such as simple meter step at present, does not gather to the difference of motion link, the slight characteristic of motion. At present, wearable health acquisition equipment authenticated by CFDA medical instruments is not available, and can only be used as reference.
2. The problem of electric quantity: due to constraints on volume, weight, and system performance, power supply capacity is one of the most important factors that limit widespread use. The watch bracelet power consumption standby time of taking green light collection at present is very limited originally, if the detailed characteristic of the motion state is gathered in a comprehensive way, the operand, the consumption of bringing will be huge more.
The invention overcomes the two contradictions, on one hand, the fine motion characteristics are comprehensively collected in the preprocessing process, on the other hand, only limited different sensor functions are called in different specific judging processes through self-adaptive algorithm matching, different specific characteristic calculation is extracted, the analysis link integrates different characteristics to reduce the dimension and sample, the amount of exercise is greatly reduced, and the accuracy is ensured through the key calculation verification link. Thereby, the following steps are achieved: the detailed characteristics of the motion state are comprehensively collected, the calculated amount and the power consumption are greatly reduced, the medical reference value is improved, and the power consumption standby capability is extracted under the same condition.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (7)
1. An intelligent waistband based on an artificial intelligence hierarchical motion recognition method is characterized by comprising the following steps:
the motion state solving module is used for acquiring acceleration data output by the acceleration sensor at regular time, filtering high-frequency noise through wavelet transformation and dividing the motion state of a user;
the motion link judgment module is used for subdividing the motion state by combining the angular speed to obtain a motion subdivision link;
the dimensionality reduction solving module is used for further finely distinguishing partial pathological features by using the optimized SVM hierarchical level and combining different hierarchical data;
the physiological parameter setting module is used for setting physiological parameters to be acquired under different motion states and gaits;
and the physiological parameter acquisition module is used for acquiring corresponding physiological parameters under the appointed motion state and gait according to the physiological parameters required to be acquired.
2. The intelligent belt based on artificial intelligence hierarchical motion recognition method according to claim 1, characterized in that: the motion link judging module is configured to:
acquiring acceleration values a of the acceleration sensor in the directions of the x, y and z axesx、ay、azAnd solving the acceleration signal vector modulus SVMA(ii) a Collecting angular velocity values w of the angular velocity sensor in the directions of three axes x, y and zx、wy、wzAnd solving the vector modulus SVM of the angular velocity signalW(ii) a The above-mentioned
Vector modulo SVM by acceleration signalAVector mode SVM for angular velocity signalWAnd establishing an identification model which is perfect in self-adaptation, and subdividing the motion state to obtain a motion subdivision link.
3. The intelligent belt based on artificial intelligence hierarchical motion recognition method according to claim 2, characterized in that: the motion link judging module also comprises a filtering unit, wherein the filtering unit carries out wavelet transformation operation of three steps of wavelet decomposition, high-frequency wavelet coefficient processing and wavelet reconstruction on the collected data of the acceleration sensor and the angular velocity sensor, discretizes time domain signals in all directions, decomposes mixed signals of various frequency components into different frequency bands, and then processes the mixed signals according to different characteristics of various seed signals on the frequency domain and frequency bands; acquiring gait data with high signal to noise ratio;
the wavelet transform adopts a hard threshold method, and the wavelet coefficient is Cj,kThe threshold is lambda;
the above-mentioned
4. The intelligent belt based on artificial intelligence hierarchical motion recognition method according to claim 1, characterized in that: the dimension reduction solving module is configured to:
acquiring physiological parameter data of a user, and reducing the dimension of the physiological parameter data of the user;
taking the standard deviation of the whole sample population, taking N as the sample size, training a classifier, and identifying gait samples by using the classifier;
comprehensively calculating the degree x of deviation of certain gait from the tidying crowd, wherein
Wherein, ai、bi、ciAcceleration in the x, y, z axis directions respectively,the acceleration of the whole population in the directions of the x axis, the y axis and the z axis of a certain gait link is respectively.
Inputting the personal gait samples of N types registered in the database into a classifier for training, judging which type is (1, N) according to an input value, if the input value exceeds the range of (1, N), newly registering the type of N +1, and then updating the classifier again;
and on the basis of the different motion division, the same motion is subdivided again, and a classification result is determined by adopting a voting mode.
5. The intelligent waistband based on an artificial intelligence hierarchical motion recognition method of claim 4 wherein the dimension reduction solution module is further configured to:
the raw data is normalized by subtracting the mean of the column from each element in the proof and dividing by the standard deviation of the column, such that each variable is normalized to a matrix X with a mean of 0 and a variance of 1, i.e., X ═ X1，X2，......Xn]T＝[Xij](n×p)；
Wherein,
to obtain
Solving a correlation coefficient matrix:
where R is a real symmetric matrix (i.e., R)ij＝rji) Wherein r is a correlation coefficient;
solving a correlation coefficient matrix:
if the accumulated contribution rate reaches more than 50%, a ratio height method is adopted, the characteristic value vector of the highest contribution rate is left as a fixed working sample set, and the rest is discarded;
calculating a partial matrix, using the remained characteristic value as a new variable principal component, and calculating a partial matrix F by using the following formula(n×m)＝X(n×p)·U(p×m)；
Wherein X is an original data matrix, U is a principal component load, and a score matrix F is a result obtained after dimensionality reduction;
continuous training subdivision is performed in combination with big data: fitness function f (x) of SVM classifieri)＝min(1-g(xi))，And dividing the sample into correct rates for the SVM classifier.
6. The intelligent waistband based on artificial intelligence hierarchical motion recognition method as claimed in claim 4, wherein: the dimension reduction solving module is further configured to:
calculating a first condition, a second condition and a third condition by using the output data of the acceleration sensor, and judging the gait of the human motion by using median filtering;
the first condition is: the accelerometer outputs a synthesized amplitude value, and if the synthesized amplitude value is between the given upper threshold value and the given lower threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
the accelerometer output composite amplitude is:
the upper and lower thresholds are respectively: th (h)amin＝8m/s，thamax＝11m/s；
The first condition is expressed as:
the second condition is: if the local variance output by the accelerometer is lower than a given threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
the local variance of the accelerometer output is:
whereinFor the interval, the accelerometer synthesizes an output average value of the amplitude, and the expression is as follows:
s is the number of half-window samples, and a given threshold is defined as: th (h)σa＝0.5m/s2The second condition is represented as:
the third condition is: the angular velocity sensor outputs an angular velocity composite amplitude, and if the angular velocity composite amplitude is lower than a given threshold value, the human body is judged to be static; otherwise, judging the motion of the person;
define the composite amplitude of the gyroscope output as:
the given thresholds are: th (h)wmaxThe third condition is expressed as:
7. the intelligent belt based on the artificial intelligence hierarchical motion recognition method as claimed in claim 1, further comprising a model adaptive perfecting module; the model adaptation refinement module is configured to:
reading a new input sample, and calculating the recognition rate of the SVM classifier according to a cross verification method;
if the current recognition rate of the training is higher than or equal to the original recognition rate, setting the parameters of the training as the optimal parameters; otherwise, a selection operation, a crossover operation and/or a mutation operation are executed, and the training parameters are further optimized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711162489.2A CN109805935A (en) | 2017-11-21 | 2017-11-21 | A kind of intelligent waistband based on artificial intelligence hierarchical layered motion recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711162489.2A CN109805935A (en) | 2017-11-21 | 2017-11-21 | A kind of intelligent waistband based on artificial intelligence hierarchical layered motion recognition method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109805935A true CN109805935A (en) | 2019-05-28 |
Family
ID=66599195
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711162489.2A Pending CN109805935A (en) | 2017-11-21 | 2017-11-21 | A kind of intelligent waistband based on artificial intelligence hierarchical layered motion recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109805935A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390565A (en) * | 2019-07-23 | 2019-10-29 | 厦门市佳音在线股份有限公司 | The method and system of intelligent gateway adaptive management are realized by AI edge calculations |
| CN111221419A (en) * | 2020-01-13 | 2020-06-02 | 武汉大学 | Array type flexible capacitor electronic skin for sensing human motion intention |
| CN111544005A (en) * | 2020-05-15 | 2020-08-18 | 中国科学院自动化研究所 | Parkinson's disease dyskinesia quantification and identification method based on support vector machine |
| CN111990967A (en) * | 2020-07-02 | 2020-11-27 | 北京理工大学 | Gait-based Parkinson disease recognition system |
| CN116746910A (en) * | 2023-06-15 | 2023-09-15 | 广州医科大学附属脑科医院 | Gait monitoring method and device based on wearable equipment and wearable equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101558996A (en) * | 2009-05-15 | 2009-10-21 | 天津大学 | Gait recognition method based on orthogonal projection three-dimensional reconstruction of human motion structure |
| CN106971059A (en) * | 2017-03-01 | 2017-07-21 | 福州云开智能科技有限公司 | A kind of wearable device based on the adaptive health monitoring of neutral net |
-
2017
- 2017-11-21 CN CN201711162489.2A patent/CN109805935A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101558996A (en) * | 2009-05-15 | 2009-10-21 | 天津大学 | Gait recognition method based on orthogonal projection three-dimensional reconstruction of human motion structure |
| CN106971059A (en) * | 2017-03-01 | 2017-07-21 | 福州云开智能科技有限公司 | A kind of wearable device based on the adaptive health monitoring of neutral net |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390565A (en) * | 2019-07-23 | 2019-10-29 | 厦门市佳音在线股份有限公司 | The method and system of intelligent gateway adaptive management are realized by AI edge calculations |
| CN111221419A (en) * | 2020-01-13 | 2020-06-02 | 武汉大学 | Array type flexible capacitor electronic skin for sensing human motion intention |
| CN111544005A (en) * | 2020-05-15 | 2020-08-18 | 中国科学院自动化研究所 | Parkinson's disease dyskinesia quantification and identification method based on support vector machine |
| CN111544005B (en) * | 2020-05-15 | 2022-03-08 | 中国科学院自动化研究所 | Parkinson's disease dyskinesia quantification and identification method based on support vector machine |
| CN111990967A (en) * | 2020-07-02 | 2020-11-27 | 北京理工大学 | Gait-based Parkinson disease recognition system |
| CN116746910A (en) * | 2023-06-15 | 2023-09-15 | 广州医科大学附属脑科医院 | Gait monitoring method and device based on wearable equipment and wearable equipment |
| CN116746910B (en) * | 2023-06-15 | 2024-05-28 | 广州医科大学附属脑科医院 | Gait monitoring method and device based on wearable equipment and wearable equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106971059B (en) | Wearable equipment based on neural network self-adaptation health monitoring | |
| CN108244744B (en) | Motion state identification method, sole and shoe | |
| CN109805935A (en) | A kind of intelligent waistband based on artificial intelligence hierarchical layered motion recognition method | |
| Huang et al. | TSE-CNN: A two-stage end-to-end CNN for human activity recognition | |
| CN108831527B (en) | User motion state detection method and device and wearable device | |
| CN104729507B (en) | A kind of gait recognition method based on inertial sensor | |
| Meyer et al. | Wearables and deep learning classify fall risk from gait in multiple sclerosis | |
| CN108514421A (en) | The method for promoting mixed reality and routine health monitoring | |
| CN112754468B (en) | Human body lower limb movement detection and identification method based on multi-source signals | |
| Yoneyama et al. | Accelerometry-based gait analysis and its application to Parkinson's disease assessment—part 1: detection of stride event | |
| CN107753026A (en) | For the intelligent shoe self-adaptive monitoring method of backbone leg health | |
| CN109069066A (en) | Wearable and connection gait analysis system | |
| Jiang et al. | The elderly fall risk assessment and prediction based on gait analysis | |
| Muscillo et al. | An adaptive Kalman-based Bayes estimation technique to classify locomotor activities in young and elderly adults through accelerometers | |
| Dasgupta et al. | Acceleration gait measures as proxies for motor skill of walking: A narrative review | |
| CN110390565A (en) | The method and system of intelligent gateway adaptive management are realized by AI edge calculations | |
| CN108958482B (en) | Similarity action recognition device and method based on convolutional neural network | |
| CN108836337A (en) | A method of personalized sufficient type health detection is carried out by foot movement state | |
| CN110169774B (en) | Motion state identification system and method based on block chain | |
| CN108021888A (en) | A kind of fall detection method | |
| CN104586402B (en) | A kind of feature extracting method of physical activity | |
| Chuang et al. | A wearable activity sensor system and its physical activity classification scheme | |
| Pan et al. | Evaluation of hemiplegic gait based on plantar pressure and inertial sensors | |
| CN109045682A (en) | A method of it reducing projection mobile phone and interacts body-building game propagation delay time with intelligent shoe | |
| Datta et al. | Upper limb movement profiles during spontaneous motion in acute stroke |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190528 |
|
| WD01 | Invention patent application deemed withdrawn after publication |