CN103892840B - The feature extracting method of a kind of Intelligent worn device and human motion - Google Patents

Abstract

The present invention relates to the measurement field of the regular action of human body, especially the feature extracting method of a kind of Intelligent worn device and human motion is related to, this Intelligent worn device comprises characteristic extracting module, described characteristic extracting module comprises: pretreatment unit, in order to carry out the three-dimensional gyroscope component in the data queue collected to convert the first data queue to and three-dimensional acceleration component is converted to the process of the second data queue; First searches unit, in order to carry out finding out the beginning of special exercise section and the process of end in the first described data queue; Second searches unit, in order to carry out the process finding out the beat chain in described special exercise section in the second described data queue, provides beat chain on average to describe; Feature extraction unit, in order to carry out the extraction process of basic acts feature; And sports immunology generation unit.The present invention can require and power consumption requirements to reduce calculation resources by shortcut calculation effectively.

Description

The feature extracting method of a kind of Intelligent worn device and human motion

Technical field

The present invention relates to the equipment and method that measure the regular action of human body, referring more particularly to equipment and method that a kind of sensor by being worn on human body carries out the feature extraction of regular action.

Background technology

At society, the quickening of rhythm of life and the increasing of operating pressure make increasing people be in sub-health state.Such people also just more and more pay close attention to oneself health status, take various measures to improve the health status of oneself, such as start to adjust oneself work and rest rhythm, rational diet, moderately do various motion.Improve in the measure of health status various, motion is a very important measure.Suitable motion can strengthen the metabolism of human body, moulds perfect figure, helps people to get rid of unhealthy emotion.Along with the development of science and technology, society there is the electronic product of a series of monitoring moving.Such as: Chinese patent CN200710097593.8 discloses a kind of Wrist watch type acceleration sensing module of measuring amount of exercise, a microprocessor, an acceleration transducer, the data base of the corresponding step-length of an intervalometer, on the other hand oscillating acceleration and a display is comprised.The hand number of oscillations that acceleration transducer is divided a word with a hyphen at the end of a line in order to sense movement person and hand oscillating acceleration, intervalometer is in order to calculate the transient time of sporter.Received hand oscillating acceleration can the curve of the corresponding step-length of hands oscillating acceleration stored by data base of step-length corresponding to hands oscillating acceleration compare by microprocessor, and obtain corresponding step-length, then the hand number of oscillations of step-length, sporter and the transient time of sporter are obtained via formulae discovery Distance geometry speed of dividing a word with a hyphen at the end of a line.These products generally can calculate user more accurately and walk, and run, swimming, the time experienced during the motions such as mountain-climbing, distance and the energy consumed.But before use equipment, the monitoring content often needing user oneself to remove to arrange equipment can make equipment can measure accurately the motion that user will do.Like this, be easy to occur that user is forgotten the monitoring content of switching device and makes the inaccurate situation of exercise data.Further, easily making user produce a kind of monitoring content manually arranging these equipment that goes frequently is the sensation of a pretty troublesome thing, causes Consumer's Experience not good enough.Along with the development of sensing technology, there is the business application of the nine axle sensing modules integrating 3-axis acceleration, three-axis gyroscope and three axis magnetometer, such as: disclose in US Patent No. 2012/0323520 and in Intelligent wearable equipment, adopt machine learning and automatic identification technology to catch, analyze the regular action of human body, with further to user report quantity of motion.The employing of these intellectual technologies, requires also day by day to promote to the computing capability of equipment, correspondingly also can cause the lifting of power consumption requirements.

In view of Intelligent worn device, such as: movement spire lamella, be limited to the less device space, there is the design constraint that computing capability is limited and need battery to power as far as possible for a long time, under the prerequisite promoting Consumer's Experience as much as possible, how shortcut calculation requires and power consumption requirements to reduce calculation resources, is the direction that people make great efforts always.

Summary of the invention

The technical problem to be solved in the present invention is, for the above-mentioned defect of prior art, provides the feature extracting method of a kind of Intelligent worn device and human motion, can effectively require and power consumption requirements to reduce calculation resources by shortcut calculation.

The technical solution adopted for the present invention to solve the technical problems comprises: the feature extracting method providing a kind of human motion, comprises successively:

Carry out pretreatment, it comprises the process carrying out one first data queue in order to the to weigh physical activity amplitude three-dimensional gyroscope component in the data queue collected being converted to an one dimension, and carries out the process of one second data queue in order to weigh the change beat that human cyclin the moves three-dimensional acceleration component in the data queue collected being converted to an one dimension;

Carry out finding out the beginning of special exercise section and the process of end in the first described data queue;

Carry out the process finding out the beat chain in described special exercise section in the second described data queue, provide beat chain on average to describe;

Carry out the extraction process of basic acts feature;

To extract and whether the quantity of the basic acts feature of preserving reaches the judgement process of setting requested number, if so, generated the description of basic acts of motion; Otherwise, carry out slip process, and return and above-mentioned in the first data queue, find out the beginning of special exercise section and the process of end, carry out circular treatment.

The technical solution adopted for the present invention to solve the technical problems also comprises: provide a kind of Intelligent worn device, comprises a module, and in order to complete the feature extraction of human motion, described module comprises:

Pretreatment unit, in order to carry out the process of one first data queue in order to the to weigh physical activity amplitude three-dimensional gyroscope component in the data queue collected being converted to an one dimension, and carry out the process of one second data queue in order to weigh the change beat that human cyclin the moves three-dimensional acceleration component in the data queue collected being converted to an one dimension;

First searches unit, in order to carry out finding out the beginning of special exercise section and the process of end in the first described data queue;

Second searches unit, in order to carry out the process finding out the beat chain in described special exercise section in the second described data queue, provides beat chain on average to describe;

Feature extraction unit, in order to carry out the extraction process of basic acts feature; And

Sports immunology generation unit, in order to extract and whether the quantity of the basic acts feature of preserving reaches the judgement process of setting requested number, is generate the description of basic acts of motion; Otherwise, carry out slip process, and turn back to above-mentioned first and search unit and carry out circular treatment.

Beneficial effect of the present invention is, is converted to first data queue in order to weigh physical activity amplitude of an one dimension, and finds out special exercise section accordingly by the data queue of the gyroscope component by three-dimensional; The second data queue of the change beat in order to weigh human cyclin motion of an one dimension is converted to by the data queue of the component of acceleration by three-dimensional, and find out the beat chain of special exercise section accordingly, carry out the extraction of basic acts feature more on this basis, until complete the generation of the description of the basic acts of motion, and then can identify according to the description of the basic acts of the motion obtained, can effectively require and power consumption requirements to reduce calculation resources by shortcut calculation.

Accompanying drawing explanation

Below in conjunction with drawings and Examples, the invention will be further described, in accompanying drawing:

Fig. 1 is the schematic diagram of feature extracting method of the present invention.

Fig. 2 is the flow chart of feature extracting method embodiment of the present invention.

Fig. 3 is the structured flowchart of the present invention's intelligence object wearing device.

Fig. 4 is the structured flowchart of characteristic extracting module embodiment of the present invention.

Detailed description of the invention

Now by reference to the accompanying drawings, preferred embodiment of the present invention is elaborated.

Fig. 1 is the schematic diagram of feature extracting method of the present invention.The present invention proposes a kind of feature extracting method of human motion, comprises the following steps successively:

S101: carry out pretreatment, it comprises the process carrying out one first data queue in order to the to weigh physical activity amplitude three-dimensional gyroscope component (three-axis gyroscope signal) in the data queue collected being converted to an one dimension, and carries out the process of one second data queue in order to weigh the change beat that human cyclin the moves three-dimensional acceleration component (three axis accelerometer signal) in the data queue collected being converted to an one dimension;

S102: carry out finding out the beginning of special exercise section and the process of end in the first described data queue;

S103: carry out the process finding out the beat chain in described special exercise section in the second described data queue, provide beat chain on average to describe;

S104: the extraction process of carrying out basic acts feature;

S105: to have extracted and whether the quantity of the basic acts feature of preserving reaches the judgement process of setting requested number, if so, generates the description of basic acts of motion; Otherwise, carry out slip process, and return and above-mentioned in the first data queue, find out the beginning of special exercise section and the process of end, carry out circular treatment.

In the present invention, described special exercise section refers to the longest one of span in the multiple motor segments found in the first described data queue.

In the present invention, the description of the basic acts of described motion comprises average and the mean-square value of the every one dimension component to numerical quantity group of each basic acts feature.

In the present invention, the process in order to one first data queue weighing physical activity amplitude that three-dimensional gyroscope component in the data queue collected is converted to an one dimension by described carrying out refers to: for each sequence of points, and after averaging respectively to the three-dimensional gyroscope component of each history point, the value of square root is first sued for peace average process again.

In the present invention, the process that three-dimensional acceleration component in the data queue collected is converted to one second data queue of the change beat in order to weigh human cyclin motion of an one dimension by described carrying out refers to: for each sequence of points, to the process that three-dimensional acceleration component is sued for peace.

Particularly, step S101 comprises further:

Processor constantly gathers synchrodata from gyroscope and accelerometer, and by the deposit data that collects in data fifo queue windowData, like this until windowData is filled up by data; It should be noted that, choosing of the length of data fifo queue, relevant to the sample frequency of sensor, such as, for the sample frequency of 25Hz, the length of data fifo queue should be not less than 200, for the sample frequency of 50Hz, the length of data fifo queue should be not less than 400, and in other words, a data fifo queue approximately can preserve the sampled data in 8 seconds.Design like this, for cycle of the regular action of general human body at about 1 second, maximumly generally more than the situation of 1.5 seconds, can not can capture by a data fifo queue motor segment Wave data that includes multiple beat (wave period).

With the data queue motionLevel of the reflection physical activity amplitude of 3 dimension gyroscope component construction, one 1 dimension in the data fifo queue windowData of 6 DOF, and specify that human motion and static marginal value are motionLevelThreshold=1 for the data in motionLevel.If when certain one piece of data in motionLevel is higher than motionLevelThreshold, just think that during this period of time, people is in motion; Otherwise, just think that during this period of time, people is static;

With one the 1 dimension beat data queue motionClock of 3 dimension accelerometer component construction in the data fifo queue windowData of 6 DOF, the change beat of motionClock embodies the change beat of human cyclin motion.

Step S102 comprises further: contrast motion active index threshold values motionLevelThreshold, the queue of current kinetic active index is searched, record active index is more than the often a bit of starting position of motionLevelThreshold and end position (serial number of namely corresponding data queue), and from these segments, find out maximum a bit of of sequence span, as pending special exercise section.

Step S103 comprises further: the beat data queue motionClock process that the special exercise section starting position obtained according to step S102 and end position and step S101 obtain.The result of beat chain on average describes except above-mentioned beat chain, also provides starting position and the end position of the number of beat and multiple beat.Wherein, beat chain on average describes the average description of the multiple beats referred in described beat chain.More specifically, the acquisition of the described average description to multiple beat have employed the feature extraction based on wavelet analysis.In the present embodiment, the described feature extraction based on wavelet analysis comprises: for each sequence of points in each beat, carry out the normalized of deviation relative variance.In the present embodiment, the process of beat chain have employed secondary clustering method.Particularly, the first order cluster of described secondary clustering method have employed C-means algorithm idea, main employing difference comparsion technology in categorizing process.The second level cluster of described secondary clustering method have employed C-means algorithm idea, main employing similarity system design technology in categorizing process.

In step S104, the every one dimension component that have employed subtend numerical quantity group carries out wavelet analysis.So, by reducing the calculating dimension of wavelet analysis, greatly amount of calculation can be reduced.Similarity system design between have employed based on basic acts feature, and basic acts characteristic standard is filled along with the queue of basic acts feature and dynamically upgrades.Described basic acts feature and the similarity system design of basic acts characteristic standard perform under a suspection mechanism, when suspection reaches the degree of agreement, will empty the queue of basic acts feature and basic acts characteristic standard.Described basic acts feature have employed multilamellar with the similarity system design of basic acts characteristic standard and compares.Described multilamellar compares the similarity system design between the average description comprising beat chain.

In step S105, exemplify as one, this setting requested number is 20.

Further preferred version of the present invention is: described carry out in the first data queue the process that motor segment and static segment search and specifically comprise:

First, arrange motor segment and be numbered 0, static segment is numbered 0;

Then, according to vertical order, the data of the first data queue are compared with human motion and static marginal value successively:

For the 1st element of the first data queue, when the value of discovery the 1st element is more than or equal to human motion and static marginal value, by motor segment numbering from adding 1, and numbering 1 is stored as the starting position of the motor segment of motor segment numbering indication, and then the value of the 2nd element is searched, if find that the value of the 2nd element is more than or equal to human motion and static marginal value, so exit; If find that the value of the 2nd element is less than human motion and static marginal value, so numbering 1 can be stored as the end position of the motor segment of motor segment numbering indication, calculate simultaneously and store the length of this motor segment, static segment numbering is added 1 certainly simultaneously, numbering 2 is stored as the starting position of the static segment of static segment numbering indication; When the value of discovery the 1st element is less than human motion and static marginal value, can by static segment numbering from adding 1, and numbering 1 is stored as the starting position of the static segment of static segment numbering indication, and then the value of the 2nd element is searched, if find that the value of the 2nd element is less than human motion and static marginal value, exit; If find that the value of the 2nd element is more than or equal to human motion and static marginal value, so numbering 1 can be stored as the end position of the static segment of static segment numbering indication, calculate simultaneously and store the length of this static segment, motor segment numbering is added 1 certainly simultaneously, numbering 2 is stored as the starting position of the motor segment of motor segment numbering indication;

For the n-th element of the first data queue, this n-th element is the 1st element between element and last element, when the value of discovery n-th element be more than or equal to human motion be less than human motion and static marginal value with static marginal value and the value of (n+1)th element time, numbering n is stored as the end position of the motor segment of motor segment numbering indication, calculate and store the length of this motor segment, by static segment numbering from adding 1, numbering n+1 is stored as the starting position of the static segment of static segment numbering indication simultaneously; When the value of discovery n-th element be less than human motion be more than or equal to human motion and static marginal value with static marginal value and the value of (n+1)th element time, then numbering n is stored as the end position of the static segment of static segment numbering indication, calculate and store the length of this static segment, by motor segment numbering from adding 1, numbering n+1 is stored as the starting position of the motor segment of motor segment numbering indication simultaneously;

For last element of the first data queue, when finding that the value of last element is more than or equal to human motion and static marginal value, then the numbering of last element is stored as the end position of the motor segment of motor segment numbering indication, calculates and store the length of this motor segment; When finding that the value of last element is less than human motion and static marginal value, then the numbering of last element being stored as the end position of the static segment of static segment numbering indication, calculating and storing the length of this static segment.

Further preferred version of the present invention is: the process of the information of the beat chain in described special exercise section of searching in the second described data queue have employed secondary clustering method, the first order cluster of described secondary clustering method have employed carries out the C-means algorithm of classifying based on difference comparsion, and the second level cluster of described secondary clustering method have employed carries out the C-means algorithm of classifying based on similarity system design.

Further preferred version of the present invention is: the calculation procedure (namely the implementation procedure of second level cluster) of the feature extraction of the cadence signal that human body repeatable motion produces is as follows:

A beat { a of human body repeatable motion is provided _i} _1≤i≤n;

Calculate expectation and the variance of this beat: $E=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{a}_i,V=\sqrt[2]{\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(a_i-E)}^2};$

Structure { b _i} _1≤i≤n, $b_i=\frac{a_i-E}{V};$

Segmentation hop count is set: sectionNum, and segmentation yardstick:

And do following calculating:

Work as i=1 ..., during sectionNum-1, have

$s_i=\frac{1}{\sec \mathrm{tionMeasure}}\underset{j=(i-1)\×\sec \mathrm{tionMeasure}+1}{\overset{i\×\sec \mathrm{tionMeasure}}{\mathrm{\Σ}}}{b}_i$

As i=sectionNum, have

$s_i=\frac{1}{n-(i-1)\×\sec \mathrm{tionMeasure}}\×\underset{j=(\sec \mathrm{tionNum}-1)\×\sec \mathrm{tionMeasure}+1}{\overset{n}{\mathrm{\Σ}}}{b}_j;$

Therefore claim { E, V, { s _i} _{1≤i≤sectionNum}be { a _i} _1≤i≤nfeature, wherein { s _i} _{1≤i≤sectionNum}for { a _i} _1≤i≤nshape facility.

Further preferred version of the present invention is: the calculation procedure of the similarity system design between body weight for humans renaturation tact of motion signal is as follows:

Arranging beat A is its shape facility is beat B is its shape facility is

Definition { d _i} _{1≤i≤sectionNum}, wherein

Similar bottom valve limit similarityThreshold is set, does following calculating:

$E^d=\frac{1}{\sec \mathrm{tionNum}}{\mathrm{\Σ}}_1^{\sec \mathrm{tionNum}}{d}_i,$

$V^d={\left(\frac{1}{\sec \mathrm{tionNum}}{\mathrm{\Σ}}_{i=1}^{\sec \mathrm{tionNum}}{(d_i-E^d)}^2\right)}^{\frac{1}{2}},$

Get 0.1≤similarityThreshold≤0.3,

Work as V ^dduring≤similarityThreshold, { a _i} _1≤i≤nwith { b _i} _1≤i≤nsimilar;

Work as V ^dduring > similarityThreshold, { a _i} _1≤i≤nwith { b _i} _1≤i≤ndissimilar.

Fig. 2 is the flow chart of feature extracting method embodiment of the present invention.It roughly comprises the following steps:

S201: pretreatment, three-dimensional vector is formed for each three-dimensional gyroscope component, by the length of itself and in buffered data queue the length come before it corresponding to all history vectors be added together and average again, result of calculation, as the value on relevant position in the first data queue, has so just constructed the first data queue of the reflection physical activity amplitude of one 1 dimension; Three-dimensional vector is formed for each three-dimensional acceleration component, three-dimensional acceleration component is carried out the process of suing for peace, result of calculation, as the value on relevant position in the second data queue, so just constructs the second data queue of the reflection physical activity beat of one 1 dimension.

S202: the starting position in queue and the end position that find out special exercise section in the first data queue, its concrete steps are as follows:

First, arrange motor segment and be numbered 0, static segment is numbered 0;

Then, according to vertical order, the data of the first data queue are compared with human motion and static marginal value successively:

For the 1st element of the first data queue, when the value of discovery the 1st element is more than or equal to human motion and static marginal value, by motor segment numbering from adding 1, and numbering 1 is stored as the starting position of the motor segment of motor segment numbering indication, and then the value of the 2nd element is searched, if find that the value of the 2nd element is more than or equal to human motion and static marginal value, so exit; If find that the value of the 2nd element is less than human motion and static marginal value, so numbering 1 can be stored as the end position of the motor segment of motor segment numbering indication, calculate simultaneously and store the length of this motor segment, static segment numbering is added 1 certainly simultaneously, numbering 2 is stored as the starting position of the static segment of static segment numbering indication; When the value of discovery the 1st element is less than human motion and static marginal value, can by static segment numbering from adding 1, and numbering 1 is stored as the starting position of the static segment of static segment numbering indication, and then the value of the 2nd element is searched, if find that the value of the 2nd element is less than human motion and static marginal value, exit; If find that the value of the 2nd element is more than or equal to human motion and static marginal value, so numbering 1 can be stored as the end position of the static segment of static segment numbering indication, calculate simultaneously and store the length of this static segment, motor segment numbering is added 1 certainly simultaneously, numbering 2 is stored as the starting position of the motor segment of motor segment numbering indication;

For the n-th element of the first data queue, this n-th element is the 1st element between element and last element, when the value of discovery n-th element be more than or equal to human motion be less than human motion and static marginal value with static marginal value and the value of (n+1)th element time, numbering n is stored as the end position of the motor segment of motor segment numbering indication, calculate and store the length of this motor segment, by static segment numbering from adding 1, numbering n+1 is stored as the starting position of the static segment of static segment numbering indication simultaneously; When the value of discovery n-th element be less than human motion be more than or equal to human motion and static marginal value with static marginal value and the value of (n+1)th element time, then numbering n is stored as the end position of the static segment of static segment numbering indication, calculate and store the length of this static segment, by motor segment numbering from adding 1, numbering n+1 is stored as the starting position of the motor segment of motor segment numbering indication simultaneously;

For last element of the first data queue, when finding that the value of last element is more than or equal to human motion and static marginal value, then the numbering of last element is stored as the end position of the motor segment of motor segment numbering indication, calculates and store the length of this motor segment; When finding that the value of last element is less than human motion and static marginal value, then the numbering of last element being stored as the end position of the static segment of static segment numbering indication, calculating and storing the length of this static segment.

S203: according to the starting position found in the first data queue and end position, search the information of the beat chain in the special exercise section in the second data queue, comprise the characteristic information of beat waveform, the number of beat, the starting position of each beat and final position.

Wherein, the process of the information of the beat chain in described special exercise section of searching in the second described data queue have employed secondary clustering method, the first order cluster of described secondary clustering method have employed carries out the C-means algorithm of classifying based on difference comparsion, and the second level cluster of described secondary clustering method have employed carries out the C-means algorithm of classifying based on similarity system design.

S204: extract the wave character of the three axis accelerometer signal segment synchronous with beat and three-axis gyroscope signal segment as basic acts feature from sensing data buffer queue according to the starting position of each beat of the information of beat chain and final position, if first time extracts basic acts feature, first of feature queue element is then it can be used as to store, and with its characteristic standard as basic acts.

The extracting method of basic acts feature, the feature extraction of both synchronous with the cadence signal sextuple data sequence fragment be made up of 3-axis acceleration signal and three-axis gyroscope signal, specifically comprises:

Sextuple data sequence fragment is respectively tieed up and calculates, obtain the expectation and variance of each dimension; Re-construct sextuple data sequence fragment according to above-mentioned expectation and variance and obtain data float scaling matrices; According to setup parameter, comprise segmentation hop count and segmentation yardstick, data sliding scales matrix is carried out longitudinal direction segmentation and calculated, obtains data float ratio and describe sequence; The expectation respectively tieed up according to sextuple data sequence fragment and variance and data float ratio describe sequence, obtain the basic acts feature that corresponding sports beat indicates; Wherein, the hop count that this segmentation hop count is longitudinally split for arranging described data float scaling matrices, this segmentation yardstick is used for the data length that setting data sliding scales matrix respectively ties up compartmented, and described segmentation hop count is set to 3 ~ 10 sections.

Specific implementation comprises:

Sextuple data sequence fragment: unitData ∈ R ^{6 × l}; Sextuple data sequence fragment length: l;

The expectation that sextuple data sequence fragment is respectively tieed up and variance:

$\mathrm{accXE}=\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}\mathrm{unitData}(1,j),\mathrm{accYE}=\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}\mathrm{unitData}(2,j),\mathrm{accZE}=\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}\mathrm{unitData}(3,j),$

$\mathrm{accXV}={\left(\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{(\mathrm{unitData}(1,j)-\mathrm{accXE})}^2\right)}^{\frac{1}{2}},\mathrm{accYV}={\left(\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{(\mathrm{unitData}(2,j)-\mathrm{accYE})}^2\right)}^{\frac{1}{2}},$

$\mathrm{accZV}={\left(\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{(\mathrm{unitData}(3,j)-\mathrm{accZE})}^2\right)}^{\frac{1}{2}},$

$\mathrm{gyroXE}=\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}\mathrm{unitData}(4,j),\mathrm{gyroYE}=\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}\mathrm{unitData}(5,j),\mathrm{gyroZE}=\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}\mathrm{unitData}(6,j),$

$\mathrm{gyroXV}={\left(\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{(\mathrm{unitData}(4,j)-\mathrm{gyroXE})}^2\right)}^{\frac{1}{2}},\mathrm{gyroYV}={\left(\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{(\mathrm{unitData}(5,j)-\mathrm{gyroYE})}^2\right)}^{\frac{1}{2}},$

$\mathrm{gyroZV}={\left(\frac{1}{l}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{(\mathrm{unitData}(6,j)-\mathrm{gyroZE})}^2\right)}^{\frac{1}{2}},$

Wherein:

The amount that accX represents its place is relevant to the X-axis component of three-dimensional acceleration data;

The amount that accY represents its place is relevant to the Y-axis component of three-dimensional acceleration data;

The amount that accZ represents its place is relevant to the Z axis component of three-dimensional acceleration data;

The amount that gyroX represents its place is relevant to the X-axis component of three-dimensional gyro data;

The amount that gyroY represents its place is relevant to the Y-axis component of three-dimensional gyro data;

The amount that gyroZ represents its place is relevant to the Z axis component of three-dimensional gyro data;

AccXE represents the expectation of the X-axis component of three-dimensional acceleration data;

The expectation of other axle components is consistent with the representation of variance as above-mentioned accXE, does not just describe one by one.

Obtain data float scaling matrices unitDataCorrection:

$\mathrm{unitDataCorrection}=\left(\begin{array}{ccc}\frac{\mathrm{unitData}\left(\mathrm{1,1}\right)-\mathrm{accXE}}{\mathrm{accXV}}& ...& \frac{\mathrm{unitData}(1,L)-\mathrm{accXE}}{\mathrm{accXV}}\\ ...& ...& ...\\ \frac{\mathrm{unitData}\left(\mathrm{6,1}\right)-\mathrm{gyroZE}}{\mathrm{gyroZV}}& ...& \frac{\mathrm{unitData}(6,L)-\mathrm{gyroZE}}{\mathrm{gyroZV}}\end{array}\right);$

First, setting segmentation hop count: sectionNum, and segmentation yardstick:

And then, carry out piecemeal along line direction to unitDataCorrection, for front sectionNum-1 block, their width is all segmentation yardstick, and for that last block, its width not necessarily just reaches segmentation yardstick.

For the first row data of unitDataCorrection, do following calculating:

Work as i=1 ..., during sectionNum-1, have

$\begin{array}{c}\mathrm{accXRoughShape}=\\ \frac{1}{\sec \mathrm{tionMeasure}}\×{\mathrm{\Σ}}_{j=(i-1)\×\sec \mathrm{tionMeasure}+1}^{i\×\sec \mathrm{tionMeasure}}\mathrm{unitDataCorrection}(1,j)\end{array};$

As i=sectionNum, have

$\begin{array}{c}\mathrm{accXRoughShape}=\frac{1}{l-(\sec \mathrm{tionNum}-1)\×\sec \mathrm{tionMeasure}}\\ \×{\mathrm{\Σ}}_{j=(i-1)\×\sec \mathrm{tionMeasure}+1}^l\mathrm{unitDataCorrection}(1,j)\end{array};$

Wherein, accXRoughShape (i) represents the shape description sequence of data float scaling matrices in X-axis that three-dimensional acceleration data sequence fragment is corresponding, other ratios describe sequence as described in the representation shown of accXRoughShape (i) consistent, just do not describe one by one.

Wherein, for the data of other the row of unitDataCorrection, similar process is done.

Finally can obtain the basic acts feature actionFeature moved:

$\mathrm{actionFeature}=\left(\begin{array}{c}\mathrm{accXE}\\ \mathrm{accXV}\\ {\left\{\mathrm{accXRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{accYE}\\ \mathrm{accYV}\\ {\left\{\mathrm{accYRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{accZE}\\ \mathrm{accZV}\\ {\left\{\mathrm{accZRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroXE}\\ \mathrm{gyroXV}\\ {\left\{\mathrm{gyroXRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroYE}\\ \mathrm{gyroYV}\\ {\left\{\mathrm{gyroYRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroZE}\\ \mathrm{gyroZV}\\ {\left\{\mathrm{gyroZRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\end{array}\right);$

Wherein:

AccXRoughShape represents the shape description sequence of the X-axis component of three-dimensional acceleration data shape;

AccYRoughShape represents the shape description sequence of the Y-axis component of three-dimensional acceleration data shape;

AccZRoughShape represents the shape description sequence of the Z axis component of three-dimensional acceleration data shape;

GyroXRoughShape represents the shape description sequence of the X-axis component of three-dimensional gyro data shape;

GyroYRoughShape represents the shape description sequence of the Y-axis component of three-dimensional gyro data shape;

GyroZRoughShape represents the shape description sequence of the Z axis component of three-dimensional gyro data shape;

S205: judge that whether the feature of the basic acts of extracting is similar, similar to the characteristic standard of basic acts, go to step S206, dissimilar words, go to step S209.Being described in detail as follows of the enforcement of S205:

When program is from the vector value data sequence fragment that the basic acts of motion produces, after extracting one group of signal intensity feature, the characteristic standard of the basic acts feature extracted and current basic acts is carried out similarity system design by program, and then judges whether this basic acts representated by variation characteristic organizing signal and former basic acts belong to same class and move.And the basic acts feature that actionFeature is the motion of Program extraction is set.

$\mathrm{actionFeature}=\left(\begin{array}{c}\mathrm{accXE}\\ \mathrm{accXV}\\ {\left\{\mathrm{accXRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{accYE}\\ \mathrm{accYV}\\ {\left\{\mathrm{accYRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{accZE}\\ \mathrm{accZV}\\ {\left\{\mathrm{accZRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroXE}\\ \mathrm{gyroXV}\\ {\left\{\mathrm{gyroXRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroYE}\\ \mathrm{gyroYV}\\ {\left\{\mathrm{gyroYRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroZE}\\ \mathrm{gyroZV}\\ {\left\{\mathrm{gyroZRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\end{array}\right)$

And the basic acts characteristic standard that actionFeatureStd is motion is set;

$\mathrm{actionFeatureStd}=\left(\begin{array}{c}\mathrm{accXE}\\ \mathrm{accXV}\\ {\left\{\mathrm{accXRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{accYE}\\ \mathrm{accYV}\\ {\left\{\mathrm{accYRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{accZE}\\ \mathrm{accZV}\\ {\left\{\mathrm{accZRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroXE}\\ \mathrm{gyroXV}\\ {\left\{\mathrm{gyroXRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroYE}\\ \mathrm{gyroYV}\\ {\left\{\mathrm{gyroYRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ \mathrm{gyroZE}\\ \mathrm{gyroZV}\\ {\left\{\mathrm{gyroZRoughShape}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\end{array}\right)$

AccXRoughShape represents the shape description sequence of the X-axis component of three-dimensional acceleration data shape;

AccYRoughShape represents the shape description sequence of the Y-axis component of three-dimensional acceleration data shape;

AccZRoughShape represents the shape description sequence of the Z axis component of three-dimensional acceleration data shape;

GyroXRoughShape represents the shape description sequence of the X-axis component of three-dimensional gyro data shape;

GyroYRoughShape represents the shape description sequence of the Y-axis component of three-dimensional gyro data shape;

GyroZRoughShape represents the shape description sequence of the Z axis component of three-dimensional gyro data shape;

Further, setting acceleration information shape similarity threshold values accRoughShapeDifferThreshold, 0.1≤accRoughShapeDifferThreshold≤0.3,

Setting gyro data shape similarity threshold values gyroRoughShapeDifferThreshold, 0.2≤gyroRoughShapeDifferThreshold≤0.4,

Setting characteristic similarity threshold values featureSimilarityEvaluateThreshold? { 3,4}

Setting initialization feature similarity enumerator featureSimilarityEvaluate=0;

Specific algorithm comprises:

A, basic acts feature actionFeature and the shape of basic acts characteristic standard actionFeatureStd on each component is utilized to describe to produce shape difference degree,

$\mathrm{actionFeatureStd}=\left(\begin{array}{c}{\left\{\mathrm{accXRoughShapeDiffer}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ {\left\{\mathrm{accYRoughShapeDiffer}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ {\left\{\mathrm{accZRoughShapeDiffer}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ {\left\{\mathrm{gyroXRoughShapeDiffer}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ {\left\{\mathrm{gyroYRoughShapeDiffer}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\\ {\left\{\mathrm{gyroZRoughShapeDiffer}\left(i\right)\right\}}_{1\≤i\≤\sec \mathrm{tionNum}}\end{array}\right).$

Wherein:

AccXRoughShapeDiffer represents the shape difference of the shape description sequence of the X-axis component of two three-dimensional acceleration data;

AccYRoughShapeDiffer represents the shape difference of the shape description sequence of the Y-axis component of two three-dimensional acceleration data;

AccZRoughShapeDiffer represents the shape difference of the shape description sequence of the Z axis component of two three-dimensional acceleration data;

GyroXRoughShapeDiffer represents the shape difference of the shape description sequence of the X-axis component of two three-dimensional gyro data;

GyroYRoughShapeDiffer represents the shape difference of the shape description sequence of the Y-axis component of two three-dimensional gyro data;

GyroZRoughShapeDiffer represents the shape difference of the shape description sequence of the Z axis component of two three-dimensional gyro data;

Do following calculating:

As 1≤i≤sectionNum,

accXRoughShapeDiffer(i)＝actionFeature.accXRoughShape(i)-actionFeatureStd.accXRoughShape(i)；

The algorithm of other dimensions is described above, just repeats no more at this.

B, calculating shape difference degree are described in the expectation and variance of shape difference degree on each component:

$\mathrm{accXRoughShapeDifferE}=\frac{1}{\sec \mathrm{tionNum}}{\mathrm{\Σ}}_{i=1}^{\sec \mathrm{tionNum}}\mathrm{accXRoughShapeDiffer}\left(i\right);$

$\begin{array}{c}\mathrm{accXRoughShapeDifferV}=\\ \sqrt[2]{\frac{1}{\sec \mathrm{tionNum}}{\mathrm{\Σ}}_{i=1}^{\sec \mathrm{tionNum}}{(\mathrm{accXRoughShapeDiffer}\left(i\right)-\mathrm{accXRoughShapeDifferE})}^2}\end{array};$

The algorithm of the expectation and variance of the upper shape difference degree of other dimensions is described above, just repeats no more at this.

Similarity degree between the feature of the basic acts that C, judgement are extracted and the characteristic standard of motion basic acts:

If accXRoughShapeDifferV≤accRoughShapeDifferThreshold, so featureSimilarityEvaluate is from adding 1;

If accYRoughShapeDifferV≤accRoughShapeDifferThreshold, so featureSimilarityEvaluate is from adding 1;

If accZRoughShapeDifferV≤accRoughShapeDifferThreshold, so featureSimilarityEvaluate is from adding 1;

If gyroXRoughShapeDifferV≤gyroRoughShapeDifferThreshold, so featureSimilarityEvaluate is from adding 1;

If gyroYRoughShapeDifferV≤gyroRoughShapeDifferThreshold, so featureSimilarityEvaluate is from adding 1;

If gyroZRoughShapeDifferV≤gyroRoughShapeDifferThreshold, so featureSimilarityEvaluate is from adding 1;

If featureSimilarityEvaluate >=featureSimilarityEvaluateThreshold, so program thinks that the feature of the basic acts of extracting is similar to the characteristic standard of motion basic acts;

If featureSimilarityEvaluate < is featureSimilarityEvaluateThreshold, so program thinks that the characteristic standard of the feature of the basic acts of extracting and motion basic acts is dissimilar;

S206: by the basic acts characteristic storage extracted in the queue of basic acts feature; The new characteristic standard of basic acts is generated by existing basic acts feature and the feature extracted.

Being described in detail as follows of the enforcement of S206:

If program finds that the basic acts feature extracted is similar to the characteristic standard of current basic acts, so program can utilize the characteristic standard of basic acts feature and the current basic acts of extracting to regenerate new basic acts characteristic standard.

Arranging actionFeature is the basic acts characteristic standard that the basic acts characteristic sum of the motion of Program extraction arranges that actionFeatureStd is motion;

Described algorithm steps specifically comprises:

$\mathrm{actionFeatureStd}.\mathrm{accXE}=\frac{1}{2}(\mathrm{actionFeature}.\mathrm{accXE}+\mathrm{actionFeatureStd}.\mathrm{accXE});$

$\mathrm{actionFeatureStd}.\mathrm{accXV}=\frac{1}{2}(\mathrm{actionFeature}.\mathrm{accXV}+\mathrm{actionFeatureStd}.\mathrm{accXV});$

As 1≤i≤sectionNum,

$\begin{array}{c}\mathrm{actionFeatureStd}.\mathrm{accXRoughShape}\left(i\right)=\\ \frac{1}{2}(\mathrm{actionFeature}.\mathrm{accXRoughShape}\left(i\right)+\mathrm{actionFeatureStd}.\mathrm{accXRoughShape}\left(i\right))\end{array};$

The algorithm of other dimensions is described above, just repeats no more at this.

Wherein, actionFeatureStd.accXRoughShape is new basic acts characteristic standard.

S207: judge whether the quantity of basic acts feature in the queue of basic acts feature reaches setting requirement, is go to step S208, words that no, go to step S212.

S208: utilize the basic acts feature queue of this motion to calculate the statistical nature of the feature of the basic acts of this motion, and be stored in flash memorizer, as the description of this motion basic acts.

The statistical analysis of based drive basic acts feature queue, specifically comprises:

Based on the feature extraction of sextuple data sequence fragment, obtain the feature queue of the basic acts of several basic acts structural feature motion of a certain motion; Calculate expectation and the variance of the ordered series of numbers that the data on each basic acts feature correspondence position are formed; By expectation and the variance of above-mentioned generation, form the knowledge point of the basic acts of a certain motion; Knowledge base is formed by several knowledge points.

Wherein, the specific implementation extracting the knowledge point of basic acts from basic acts characteristic sequence comprises:

Basic acts number of features actionFeatureNum is set, obtains several basic acts features: { actionFeature (k) } _{1≤k≤actionFeatureNum},

Wherein:

$\mathrm{actionFeature}\left(k\right)=\left(\begin{array}{c}\mathrm{accXE}\\ \mathrm{accXV}\\ {\left\{\mathrm{accXRoughShape}\left(i\right)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ \mathrm{accYE}\\ \mathrm{accYV}\\ {\left\{\mathrm{accYRoughShape}\left(i\right)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ \mathrm{accZE}\\ \mathrm{accZV}\\ {\left\{\mathrm{accZRoughShape}\left(i\right)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ \mathrm{gyroXE}\\ \mathrm{gyroXV}\\ {\left\{\mathrm{gyroXRoughShape}\left(i\right)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ \mathrm{gyroYE}\\ \mathrm{gyroYV}\\ {\left\{\mathrm{gyroYRoughShape}\left(i\right)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ \mathrm{gyroZE}\\ \mathrm{gyroZV}\\ {\left\{\mathrm{gyroZRoughShape}\left(i\right)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\end{array}\right);$

The knowledge point of this action can be produced by statistical computation:

$\mathrm{actionFeature}\left(k\right)=\left(\begin{array}{c}(\mathrm{accXEE},\mathrm{accXEV})\\ (\mathrm{accXVE},\mathrm{accXVV})\\ {\left\{(\mathrm{accXRoughShape}\left(i\right).E,\mathrm{accXRoughShape}\left(i\right).V)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ (\mathrm{accYEE},\mathrm{accYEV})\\ (\mathrm{accYVE},\mathrm{accYVV})\\ {\left\{(\mathrm{accYRoughShape}\left(i\right).E,\mathrm{accYRoughShape}\left(i\right).V)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ (\mathrm{accZEE},\mathrm{accZEV})\\ (\mathrm{accZVE},\mathrm{accZVV})\\ {\left\{(\mathrm{accZRoughShape}\left(i\right).E,\mathrm{accZRoughShape}\left(i\right).V)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ (\mathrm{gyroXEE},\mathrm{gyroXEV})\\ (\mathrm{gyroXVE},\mathrm{gyroXVV})\\ {\left\{(\mathrm{gyroXRoughShape}\left(i\right).E,\mathrm{gyroXRoughShape}\left(i\right).V)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ (\mathrm{gyroYEE},\mathrm{gyroYEV})\\ (\mathrm{gyroYVE},\mathrm{gyroYVV})\\ {\left\{(\mathrm{gyroYRoughShape}\left(i\right).E,\mathrm{gyroYRoughShape}\left(i\right).V)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\\ (\mathrm{gyroZEE},\mathrm{gyroZEV})\\ (\mathrm{gyroZVE},\mathrm{gyroZVV})\\ {\left\{(\mathrm{gyroZRoughShape}\left(i\right).E,\mathrm{gyroZRoughShape}\left(i\right).V)\right\}}_{1\&le;i\&le;\sec \mathrm{tionNum}}\end{array}\right);$

Wherein,

$\mathrm{accXEE}=\frac{1}{\mathrm{actionFeatureNum}}{\mathrm{\&Sigma;}}_{i=1}^{\mathrm{actionFeatureNum}}\mathrm{actionFeature}\left(i\right).\mathrm{accXE};$

$\mathrm{accXEV}=$

$\sqrt[2]{\frac{1}{\mathrm{actionFeatureNum}}{\mathrm{\&Sigma;}}_{i=1}^{\mathrm{actionFeatureNum}}{(\mathrm{actionFeature}\left(i\right).\mathrm{accXE}-\mathrm{accXEE})}^2};$

$\mathrm{accXVE}=\frac{1}{\mathrm{actionFeatureNum}}{\mathrm{\&Sigma;}}_{i=1}^{\mathrm{actionFeatureNum}}\mathrm{actionFeature}\left(i\right).\mathrm{accXV};$

$\begin{array}{c}\mathrm{accXVV}=\\ \sqrt[2]{\frac{1}{\mathrm{actionFeatureNum}}{\mathrm{\&Sigma;}}_{i=1}^{\mathrm{actionFeatureNum}{(\mathrm{actionFeature}\left(i\right).\mathrm{accXV}-\mathrm{accXVE})}^2}}\end{array};$

As 1≤k≤sectionNum, have

$\begin{array}{c}\mathrm{accXRoughShape}\left(k\right).E=\\ \frac{1}{\mathrm{actionFeatureNum}}{\mathrm{\&Sigma;}}_{i=1}^{\mathrm{actionFeatureNum}}\mathrm{actionFeature}\left(i\right).\mathrm{accXRoughShape}\left(k\right)\end{array},$

$\begin{array}{c}\mathrm{accXRoughShape}\left(k\right).V=\\ \sqrt[2]{\frac{1}{\mathrm{actionFeatureNum}}\underset{i=1}{\overset{\mathrm{actionFeatureNum}}{\mathrm{\&Sigma;}}}{(\mathrm{actionFeature}\left(i\right).\mathrm{accXRoughShape}\left(k\right)-\mathrm{accXRoughShape}\left(k\right).E)}^2}\end{array};$

The algorithm of other dimensions is described above, just repeats no more at this.

S209: basic acts suspects that enumerator is from adding 1.

S210: judge whether the suspection enumerator of basic acts feature reaches suspection threshold values, is go to step S211, words that no, go to step S212.

S211: slip pre-treatment, empties the queue of basic acts feature; Empty basic acts characteristic standard; Counter O reset is suspected in basic acts.

S212: process of sliding, overrides processed data with the sensing data of not processed mistake, and buffer queue is filled full with the sensing data newly collected; Regenerate the first data queue and the second data queue with new data cached queue, go to step S202.

Fig. 3 is the structured flowchart of the present invention's intelligence object wearing device.The invention provides a kind of Intelligent worn device, such as: movement spire lamella, it comprises characteristic extracting module 301, sports immunology storehouse 303 and sensing module 304.Wherein, this characteristic extracting module 301 is by sensing module 304, and by adopting aforesaid feature extracting method can set up sports immunology storehouse 303, this sports immunology storehouse 303 comprises the description of the basic acts of at least one motion.Those of ordinary skill in the art can be expressly understood, module mentioned here can pass through hardware implementing, and the mode that also can add necessary general hardware platform by software realizes.

See Fig. 4, provide a kind of characteristic extracting module embodiment of Intelligent worn device, it roughly comprises: first module 401, realizes the function of step S101 in Fig. 1 in order to correspondence; Second unit 402, realizes the function of step S102 in Fig. 1 in order to correspondence; 3rd unit 403, realizes the function of step S103 in Fig. 1 in order to correspondence; 4th unit 404, realizes the function of step S104 in Fig. 1 in order to correspondence; And the 5th unit 405, the function of step S105 in Fig. 1 is realized in order to correspondence.Those of ordinary skill in the art can be expressly understood, module mentioned here and/or unit can pass through hardware implementing, and the mode that also can add necessary general hardware platform by software realizes.

The invention provides a kind of general discovery cyclical signal and the algorithm of information extraction: from the six-vector value sequence that synchronous 3-axis acceleration signal and three-axis sensor signal are formed, search similarity vector value signal section, and the feature of extracted vector value signal section, finally from multiple vector value signal characteristic, extract corresponding statistical property calculation process.Algorithm of the present invention is not limited to the application on bracelet, can be embedded in a lot of system and apply.

Should be understood that, above embodiment only in order to technical scheme of the present invention to be described, is not intended to limit, for a person skilled in the art, technical scheme described in above-described embodiment can be modified, or equivalent replacement is carried out to wherein portion of techniques feature; And these amendments and replacement, all should belong to the protection domain of claims of the present invention.

Claims (25)

1. a feature extracting method for human motion, is characterized in that, comprises successively:

Carry out pretreatment, it comprises the process carrying out one first data queue in order to the to weigh physical activity amplitude three-dimensional gyroscope component in the data queue collected being converted to an one dimension, and carries out the process of one second data queue in order to weigh the change beat that human cyclin the moves three-dimensional acceleration component in the data queue collected being converted to an one dimension;

Carry out finding out the beginning of special exercise section and the process of end in the first described data queue;

Carry out the process finding out the beat chain in described special exercise section in the second described data queue, provide beat chain on average to describe;

Carry out the extraction process of basic acts feature;

To extract and whether the quantity of the basic acts feature of preserving reaches the judgement process of setting requested number, if so, generated the description of basic acts of motion; Otherwise, carry out slip process, and return and above-mentioned in the first data queue, find out the beginning of special exercise section and the process of end, carry out circular treatment;

Described slip process comprises: override processed data with the sensing data of not processed mistake, and is filled by buffer queue full with the sensing data newly collected; The first data queue and the second data queue is regenerated with new data cached queue.

2. the feature extracting method of human motion according to claim 1, is characterized in that: the process of the beat chain that described carrying out finds out in described special exercise section in the second described data queue also provides beginning and the end of the number of beat and multiple beat.

3. the feature extracting method of human motion according to claim 2, it is characterized in that: described beat chain on average describes the average description of the multiple beats referred in described beat chain, the acquisition of the described average description to multiple beat have employed the feature extraction based on wavelet analysis.

4. the feature extracting method of human motion according to claim 3, is characterized in that: the described feature extraction based on wavelet analysis comprises: for each sequence of points in each beat, carry out the normalized of deviation relative variance.

5. the feature extracting method of human motion according to claim 1, is characterized in that: the process of the beat chain that described carrying out finds out in described special exercise section in the second described data queue have employed secondary clustering method.

6. the feature extracting method of human motion according to claim 5, is characterized in that: the first order cluster of described secondary clustering method have employed C-means algorithm idea, main employing difference comparsion technology in categorizing process.

7. the feature extracting method of human motion according to claim 6, is characterized in that: the second level cluster of described secondary clustering method have employed C-means algorithm idea, main employing similarity system design technology in categorizing process.

8. the feature extracting method of human motion according to claim 1, is characterized in that: every one dimension component that the described extraction process of carrying out basic acts feature have employed subtend numerical quantity group carries out wavelet analysis.

9. the feature extracting method of human motion according to claim 1, it is characterized in that: the described extraction process of carrying out basic acts feature have employed the similarity system design based on basic acts feature and basic acts characteristic standard, and basic acts characteristic standard is filled along with the queue of basic acts feature and dynamically upgrades.

10. the feature extracting method of human motion according to claim 9, it is characterized in that: described basic acts feature and the similarity system design of basic acts characteristic standard perform under a suspection mechanism, when suspection reaches the degree of agreement, the queue of basic acts feature and basic acts characteristic standard will be emptied.

The feature extracting method of 11. human motions according to claim 9, is characterized in that: described basic acts feature have employed multilamellar with the similarity system design of basic acts characteristic standard and compares.

The feature extracting method of 12. human motions according to claim 11, is characterized in that: described multilamellar compares the similarity system design between the average description comprising beat chain.

The feature extracting method of 13. human motions according to claim 1, is characterized in that: described special exercise section refers to the longest one of span in the multiple motor segments found in the first described data queue.

The feature extracting method of 14. human motions according to claim 1, is characterized in that: the description of the basic acts of described motion comprises average and the mean-square value of the every one dimension component to numerical quantity group of each basic acts feature.

The feature extracting method of 15. human motions according to claim 1, it is characterized in that: the process in order to one first data queue weighing physical activity amplitude that the three-dimensional gyroscope component in the data queue collected is converted to an one dimension by described carrying out refers to: for each sequence of points, after averaging respectively to the three-dimensional gyroscope component of each history point, the value of square root is first sued for peace average process again.

The feature extracting method of 16. human motions according to claim 1, it is characterized in that: the process that the three-dimensional acceleration component in the data queue collected is converted to one second data queue of the change beat in order to weigh human cyclin motion of an one dimension by described carrying out refers to: for each sequence of points, to the process that three-dimensional acceleration component is sued for peace.

The feature extracting method of 17. human motions according to claim 1, it is characterized in that: the process of the description of the basic acts of described generation campaign comprises: utilize the basic acts feature queue of motion to calculate the statistical nature of the feature of the basic acts of this motion, and be stored in flash memorizer, as the description of the basic acts of this motion.

The feature extracting method of 18. human motions according to claim 1, it is characterized in that: the process of the beat chain that described carrying out finds out in described special exercise section in the second described data queue comprises: according to the starting position found in the first data queue and end position, search the information of the beat chain in the special exercise section in the second data queue, comprise the characteristic information of beat waveform, the number of beat, the starting position of each beat and final position.

The feature extracting method of 19. human motions according to claim 18, it is characterized in that: the described extraction process of carrying out basic acts feature comprises: extract the wave character of the three axis accelerometer signal segment synchronous with beat and three-axis gyroscope signal segment as basic acts feature according to the starting position of each beat of the information of beat chain and final position from sensing data buffer queue, if first time extracts basic acts feature, first of feature queue element is then it can be used as to store, and with its characteristic standard as basic acts.

The feature extracting method of 20. human motions according to claim 1, is characterized in that: the described extraction process of carrying out basic acts feature comprises further: judge that whether the feature of current basic acts of extracting is similar to the characteristic standard of basic acts.

The feature extracting method of 21. human motions according to claim 20, it is characterized in that: described judges that the feature of current basic acts of extracting comprises to whether the characteristic standard of basic acts is similar: the waveform of current basic acts feature is described the waveform deducting basic acts characteristic standard and describes, obtain difference; Calculate the expectation of the waveform difference in each dimension; Calculate the variance of the waveform difference in each dimension; Judge: if the number that the variance of waveform difference is less than definite value exceedes certain setting number, just think that these two wave characters are similar, otherwise dissimilar.

The feature extracting method of 22. human motions according to claim 20, it is characterized in that: judge that if described the feature of current basic acts of the extracting result whether similar to the characteristic standard of basic acts is that similar, the described extraction process of carrying out basic acts feature comprises further: by the basic acts characteristic storage extracted in the queue of basic acts feature; The new characteristic standard of basic acts is generated by existing basic acts feature and the feature extracted.

The feature extracting method of 23. human motions according to claim 20, it is characterized in that: judge that if described the feature of current basic acts of the extracting result whether similar to the characteristic standard of basic acts is for dissimilar, carry out suspection number of times to add up, and then judge whether the suspection enumerator of basic acts feature reaches suspection threshold value, be, advanced line slip pre-treatment, then carry out slip process, no, directly carry out slip process.

The feature extracting method of 24. human motions according to claim 23, is characterized in that: described slip pre-treatment comprises: the queue emptying basic acts feature, empties basic acts characteristic standard, the suspection enumerator of basic acts feature.

25. 1 kinds of Intelligent worn device, is characterized in that, comprise a module, and in order to complete the feature extraction of human motion, described module comprises:

Pretreatment unit, in order to carry out the process of one first data queue in order to the to weigh physical activity amplitude three-dimensional gyroscope component in the data queue collected being converted to an one dimension, and carry out the process of one second data queue in order to weigh the change beat that human cyclin the moves three-dimensional acceleration component in the data queue collected being converted to an one dimension;

First searches unit, in order to carry out finding out the beginning of special exercise section and the process of end in the first described data queue;

Second searches unit, in order to carry out the process finding out the beat chain in described special exercise section in the second described data queue, provides beat chain on average to describe;

Feature extraction unit, in order to carry out the extraction process of basic acts feature; And

Sports immunology generation unit, in order to extract and whether the quantity of the basic acts feature of preserving reaches the judgement process of setting requested number, is generate the description of basic acts of motion; Otherwise, carry out slip process, and turn back to above-mentioned first and search unit and carry out circular treatment, described slip process comprises: override processed data with the sensing data of not processed mistake, and is filled by buffer queue full with the sensing data newly collected; The first data queue and the second data queue is regenerated with new data cached queue.