[go: up one dir, main page]

CN109086667A - Similar activity recognition method based on intelligent terminal - Google Patents

Similar activity recognition method based on intelligent terminal Download PDF

Info

Publication number
CN109086667A
CN109086667A CN201810706613.5A CN201810706613A CN109086667A CN 109086667 A CN109086667 A CN 109086667A CN 201810706613 A CN201810706613 A CN 201810706613A CN 109086667 A CN109086667 A CN 109086667A
Authority
CN
China
Prior art keywords
intelligent terminal
recognition method
acceleration
classifier
activity recognition
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
Application number
CN201810706613.5A
Other languages
Chinese (zh)
Inventor
陈建新
闫娜
周亮
于涛
刘希鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Post and Telecommunication University
Original Assignee
Nanjing Post and Telecommunication University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing Post and Telecommunication University filed Critical Nanjing Post and Telecommunication University
Priority to CN201810706613.5A priority Critical patent/CN109086667A/en
Publication of CN109086667A publication Critical patent/CN109086667A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G06F2218/16—Classification; Matching by matching signal segments
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/03—Arrangements for converting the position or the displacement of a member into a coded form
    • G06F3/033—Pointing devices displaced or positioned by the user, e.g. mice, trackballs, pens or joysticks; Accessories therefor
    • G06F3/0346—Pointing devices displaced or positioned by the user, e.g. mice, trackballs, pens or joysticks; Accessories therefor with detection of the device orientation or free movement in a 3D space, e.g. 3D mice, 6-DOF [six degrees of freedom] pointers using gyroscopes, accelerometers or tilt-sensors
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Human Computer Interaction (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

A kind of similar active recognition methods based on intelligent terminal, includes the following steps: S1, data collection steps, and physical activity data are obtained from accelerometer sensor；S2, pre-treatment step pre-process acquired physical activity data by the method for coordinate transformation, construct five dimensional vectors；S3, active characteristics step is extracted, extracts temporal signatures, frequency domain character and the time and frequency domain characteristics in pretreated physical activity data, form feature set；S4, optimal feature selection step refine feature set using feature extraction algorithm, are optimized, obtain optimal characteristics collection；S5, classifier selection and activity recognition step, choose classifier, optimal characteristics collection are passed in classifier, classification results are finally obtained.The present invention, which is realized, accurately and efficiently to be detected and identifies to physical activity, has very high use and promotional value.

Description

Similar active recognition methods based on intelligent terminal
Technical field
The present invention relates to a kind of recognition methods more particularly to a kind of similar active recognition methods based on intelligent terminal, belong to In computer and sensor technical field.
Background technique
With being constantly progressive for computer and sensor technology, physical activity identification (HAR) all has in every field Broad application prospect, such as health care, disease control, movement and body-building.Such as in daily life, tumble is the old day for human beings It often would generally one of problems faced in life, it is likely that it will cause serious consequence, and nowadays, which can pass through HAR system solves.In addition, physical activity identification technology can also be used for the energy consumption of estimation daily routines.
It under such technical background, is continued to develop along with subjects such as artificial intelligence, especially wearable sensing After device (such as acceleration transducer) occurs, the physical activity identification based on wearable sensors is possibly realized.Due to wearable biography Sensor is small in size, light weight, can integrate on the intelligent terminals such as mobile phone, smartwatch.Thus gone based on intelligent terminal It is a kind of more natural, also more convenient, protection privacy man-machine interaction mode to identify.
Currently, also having had already appeared all kinds of relevant researchs in industry, and achievement is striking.
Gieber in 2011 et al. proposes the method for the sensor fusion of smart phone to improve the accuracy rate of identification.Make The activity that user is tracked with triaxial accelerometer obtains current environmental information with microphone.Using hierarchical classification algorithm, melt Secondary classifier is closed to realize activity recognition.
Harasimowicz in 2014 et al. identifies eight kinds of daily routines using the triaxial accelerometer in smart phone.Make 98.5% accuracy rate can be obtained with KNN algorithm.But this method identifies similar active, and accuracy rate only has 94.5%.
Niazi in 2016 constructs the three-level classifier system being made of five random tree classification devices.By three times Classification is to identify five activity groups.This method is lower for upper activity recognition accuracy rate downstairs.
He in 2017 proposes a kind of Automatic Feature Extraction based on wavelet transformation and half cosine fuzzy clustering and identification Physical activity recognition methods.It is initialized using half cosine to eliminate the sensibility that fuzzy clustering is distributed initial center.But it is right It is as a result and unsatisfactory in similar movable identification.
In general, the physical activity identification for being currently based on machine learning can be divided into two classes: the method for view-based access control model and Sensor-based method.In view of privacy and environmental constraints, using sensor-based recognition methods more by industry personnel Concern, and the challenge of this method is how accurately and efficiently to detect and identify physical activity, including similar movement.
In conclusion a kind of similar active recognition methods based on intelligent terminal how is proposed, to realize to physical activity It accurately and efficiently detects and identifies, also just become row those skilled in the art and one of it is expected to solve the problems, such as jointly.
Summary of the invention
In view of the prior art, there are drawbacks described above, and the purpose of the present invention is to propose to a kind of similar actives based on intelligent terminal Recognition methods includes the following steps:
S1, data collection steps obtain physical activity data from accelerometer sensor；
S2, pre-treatment step pre-process acquired physical activity data by the method for coordinate transformation, construction Five dimensional vector out；
S3, extract active characteristics step, extract temporal signatures in pretreated physical activity data, frequency domain character with And time and frequency domain characteristics, form feature set；
S4, optimal feature selection step refine feature set using feature extraction algorithm, are optimized, obtain optimal spy Collection；
S5, classifier selection and activity recognition step, choose classifier, optimal characteristics collection are passed in classifier, finally Obtain classification results.
Preferably, the S1 data collection steps include:
User wears intelligent terminal, and acquires the people of behavior to be identified in real time by the accelerometer sensor in intelligent terminal Body activity data.
Preferably, the S2 pre-treatment step includes:
S21, physical activity data collected are segmented, generate block signal, makes every in the block signal One segment signal corresponds to a kind of behavior of user；
S22, acceleration both vertically and horizontally, coordinate are converted by 3-axis acceleration using Formula of Coordinate System Transformation Conversion formula are as follows:
Wherein, A is the vector sum of 3-axis acceleration, and G is acceleration of gravity,For the angle of G and A, Ax, Ay and Az difference For the output of triaxial accelerometer, Gx, Gy and Gz are respectively acceleration of gravity relevant to X, Y and Z, and Av is normal acceleration, Ah For horizontal acceleration.
S23, in such a way that 3-axis acceleration, vertical direction acceleration and horizontal direction acceleration are combined, structure Produce five dimensional vectors, the five dimensional vectors AH={ Av Ah Ax Ay Az }.
Preferably, physical activity data collected are segmented described in S21, include the following steps: to utilize adding window To generate reference window corresponding with physical activity Data Data signal generated and testing window.
Preferably, the S3 extraction active characteristics step includes:
S31, common feature is extracted, common feature includes average value, variance, standard deviation, maximum value, minimum value, phase relation Number crosses mean value number, peak value and power spectrum；
S32, non-common feature is extracted, non-common feature includes that power spectral density, interquartile range and binary tree are multiple small Wave conversion.
Preferably, the non-common feature of extraction described in S32 includes the following steps:
S321, power spectral density is extracted, power spectral density includes amplitude statistics and the big feature of shape Statistics two, power spectrum The amplitude statistics formula of degree is,
Wherein, ciIt is the frequency width of i-th of window, μaRepresent mean value, σaRepresent standard deviation, γaThe degree of bias is represented,Represent peak Degree,
The shape Statistics formula of power spectral density is,
Wherein, μsRepresent mean value, σsFor standard deviation, γsFor the degree of bias,Kurtosis is represented,
S322, interquartile range is extracted；
S323, Dual-tree Complex Wavelet, wavelet energy E are carried outwCalculation formula be,
Wherein, N is Decomposition order, and k is time variable, DjkIt is high frequency coefficient.
Preferably, the S4 optimal feature selection step includes: by Sequential Floating Selection sequence It is combined before floating to selection feature selecting algorithm and reliefF algorithm, dimension-reduction treatment is carried out to feature set, obtains optimal characteristics Collection.
Preferably, classifier described in S5 is SVM, KNN, randomForest or MLP.
Preferably, classifier described in S5 is MLP.
Preferably, classified using the method for 10 times of cross validations.
Compared with prior art, advantages of the present invention is mainly reflected in the following aspects:
The present invention by 3-axis acceleration and normal acceleration and horizontal acceleration are combined, constructed one five tie up to The mode of amount eliminates the influence of directional effect, ensure that the accuracy rate of activity recognition procedure.Meanwhile the present invention is from time domain, frequency Three aspects in domain and time-frequency domain have studied the feature of activity data, and optimal characteristics collection is selected to sentence to carry out the identification of similar active It is disconnected, improve the reliability of identification process.The present invention carries out activity recognition as classifier using neural network (MLP), effectively Ground ensure that result precision of the invention.It, can be in addition, the present invention also provides reference for other relevant issues in same domain Expansion extension is carried out on this basis, is applied in field in the constructing plan of other similar recognition methods, is had very wide Application prospect.
In conclusion the invention proposes a kind of similar active recognition methods based on intelligent terminal, has very high make With and promotional value.
Just attached drawing in conjunction with the embodiments below, the embodiment of the present invention is described in further detail, so that of the invention Technical solution is more readily understood, grasps.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is mobile phone coordinate schematic diagram；
Fig. 3 is world coordinates schematic diagram；
Fig. 4 is one of the acceleration comparison diagram vertically and horizontally of four kinds of similar actives；
Fig. 5 is the two of the acceleration comparison diagram vertically and horizontally of four kinds of similar actives；
Fig. 6 is the three of the acceleration comparison diagram vertically and horizontally of four kinds of similar actives；
Fig. 7 is the four of the acceleration comparison diagram vertically and horizontally of four kinds of similar actives；
Fig. 8 is characterized the flow chart schematic diagram of selection algorithm.
Specific embodiment
As shown in FIG. 1 to FIG. 8, the similar active recognition methods based on intelligent terminal that present invention discloses a kind of, including such as Lower step:
S1, data collection steps obtain physical activity data from accelerometer sensor.
S2, pre-treatment step pre-process acquired physical activity data by the method for coordinate transformation, construction Five dimensional vector out.
S3, extract active characteristics step, extract temporal signatures in pretreated physical activity data, frequency domain character with And time and frequency domain characteristics, form feature set.
S4, optimal feature selection step refine feature set using feature extraction algorithm, are optimized, obtain optimal spy Collection.
S5, classifier selection and activity recognition step, choose classifier, optimal characteristics collection are passed in classifier, finally Obtain classification results.
The S1 data collection steps include:
The executing subject of the embodiment of the present invention is intelligent terminal with accelerometer sensor, such as smart phone etc..To Identification behavior is the active state of user, for example walks, runs, is seated or stands.User wears intelligent terminal, and passes through Accelerometer sensor acquires the physical activity data of behavior to be identified in real time.
The S2 pre-treatment step includes:
S21, due to the physical activity data of acquisition be one section of continuous data flow, in order to facilitate back feature extraction and The training of recognition mode, the signal that the present invention generates physical activity data carry out adding window segmentation, generate block signal, wherein Each segment signal in block signal corresponds to a kind of behavior of user.Adding window also can be reduced data waste simultaneously, reduce complexity.
After S22, data segmentation, the influence in order to avoid mobile phone center to activity recognition utilizes Formula of Coordinate System Transformation Acceleration both vertically and horizontally is converted by 3-axis acceleration, as shown in Fig. 2, Formula of Coordinate System Transformation are as follows:
Wherein, A is the vector sum of 3-axis acceleration, and G is acceleration of gravity,For the angle of G and A, Ax, Ay and Az difference For the output of triaxial accelerometer, Gx, Gy and Gz are respectively acceleration of gravity relevant to X, Y and Z, and Av is normal acceleration, Ah For horizontal acceleration.
S23, it is as shown in Figure 4 to 7 be four kinds of similar actives vertical direction acceleration and horizontal direction acceleration.Though So, the tendency of four kinds of movable acceleration vertically and horizontally substantially is different.But it for the same activity, surveys It tries different people and is likely to be obtained different acceleration.Vertically and horizontally the acceleration of method can not completely distinguish similar work It is dynamic.Therefore, it by way of combining 3-axis acceleration, vertical direction acceleration and horizontal direction acceleration, constructs One five dimensional vector, the five dimensional vectors AH={ Av Ah Ax Ay Az }, to improve the accuracy rate of activity recognition.
Physical activity data collected are segmented described in S21, include the following steps: to generate using adding window with The corresponding reference window of physical activity Data Data signal generated and testing window.
The S3 extracts active characteristics step
S31, common feature is extracted, common feature includes average value, variance, standard deviation, maximum value, minimum value, phase relation Number crosses mean value number, peak value and power spectrum.
S32, non-common feature is extracted, for general activity, common feature facilitates movable identification, but for phase As activity, the effect of common feature is with regard to some discounts.Non- common feature includes power spectral density (PSD), quartile Spacing (IQR) and Dual-tree Complex Wavelet (DT-CWT).
The non-common feature of extraction described in S32 includes the following steps:
S321, power spectral density is extracted, power spectral density includes amplitude statistics and the big feature of shape Statistics two, power spectrum The amplitude statistics formula of degree is,
Wherein, ciIt is the frequency width of i-th of window, μaRepresent mean value, σaRepresent standard deviation, γaThe degree of bias is represented,Represent peak Degree,
The shape Statistics formula of power spectral density is,
Wherein, μsRepresent mean value, σsFor standard deviation, γsFor the degree of bias,Kurtosis is represented,
S322, interquartile range is extracted.IQR be also referred to as interquartile range or it is interior away from, in descriptive statistics be equal to third A and first quartile difference.Image space difference is as standard deviation, and IQR is also the statistical distribution characteristic of gauge signal, but it is not It is influenced by exceptional value and extreme value, the similar movement of mean value can be distinguished.
S323, Dual-tree Complex Wavelet is carried out, wavelet analysis can show the spectral information of non-stationary signal.Then, Frequency spectrum analysis method can only provide the frequency content of a fixed signal such as Fourier transformation (FT).In dynamic moving signal, Over time, frequency changes rapidly, therefore wavelet transformation can describe the different bases under axis at any given time The intensity of band.So we are converted into wavelet coefficient using discrete wavelet transformer original signal of changing commanders.But wavelet transform has Have the shortcomings that obvious: lacking translation invariance, it means that the minor change of input signal will lead to wavelet coefficient in different rulers Spend the variation of lower Energy distribution.In research of the invention, for example, it is upper downstairs, walking and hurry up etc. to move and be for timeinvariance Sensitive, so we need to find a new small echo, i.e. Dual-tree Complex Wavelet Transformation (DT-CWT).Dual-tree Complex Wavelet is not The analysis ability of traditional real valued wavelet transform multi-resolution characteristics and Time-Frequency Localization is only maintained, and there is the choosing of better direction The features such as selecting property, translation invariance and limited data redundancy.
Wavelet energy EwCalculation formula be,
Wherein, N is Decomposition order, and k is time variable, DjkIt is high frequency coefficient.
Since there are many feature that the present invention extracts, some of them is characterized in redundancy.In addition, the feature of selection is more, Complexity is higher.Accordingly, it is considered to the tradeoff between computation complexity and accuracy of identification, the S4 optimal feature selection step packet It includes: to selection feature selecting algorithm and reliefF algorithm phase before Sequential Floating Selection sequence is floated In conjunction with to feature set progress dimension-reduction treatment, acquisition optimal characteristics collection.
The basic thought operated herein is as follows: the feature set descending sort that will initially be extracted using reliefF algorithm is reached Feature set R.Since R complete or collected works, a subset X is selected in every wheel feature non-selected in R, reaches accuracy rate after subset is added To highest, subset Z is then selected in selected feature, make to reject accuracy rate after subset Z take to it is optimal.Detailed process is shown in figure 8。
Classifier described in S5 is SVM, KNN, randomForest or MLP.In the present embodiment, the classifier is most It is preferred that being selected as MLP.
In addition, the present invention is classified using the method for 10 times of cross validations.
The present invention by 3-axis acceleration and normal acceleration and horizontal acceleration are combined, constructed one five tie up to The mode of amount eliminates the influence of directional effect, ensure that the accuracy rate of activity recognition procedure.Meanwhile the present invention is from time domain, frequency Three aspects in domain and time-frequency domain have studied the feature of activity data, and optimal characteristics collection is selected to sentence to carry out the identification of similar active It is disconnected, improve the reliability of identification process.The present invention carries out activity recognition as classifier using neural network (MLP), effectively Ground ensure that result precision of the invention.It, can be in addition, the present invention also provides reference for other relevant issues in same domain Expansion extension is carried out on this basis, is applied in field in the constructing plan of other similar recognition methods, is had very wide Application prospect.
In conclusion the invention proposes a kind of similar active recognition methods based on intelligent terminal, has very high make With and promotional value.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie In the case where without departing substantially from spirit and essential characteristics of the invention, the present invention can be realized in other specific forms.Therefore, no matter From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims Variation is included within the present invention, and any reference signs in the claims should not be construed as limiting the involved claims.
In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art The other embodiments being understood that.

Claims (10)

1.一种基于智能终端的相似活动识别方法，其特征在于，包括如下步骤：1. A similar activity recognition method based on an intelligent terminal, characterized in that, comprising the steps: S1、数据采集步骤，从加速计传感器中获取人体活动数据；S1, the data acquisition step, acquiring human body activity data from the accelerometer sensor; S2、预处理步骤，通过坐标转化的方法对所获取的人体活动数据进行预处理，构造出五维向量；S2. The preprocessing step is to preprocess the acquired human activity data by means of coordinate conversion to construct a five-dimensional vector; S3、提取活动特征步骤，提取预处理过的人体活动数据中的时域特征、频域特征以及时频域特征，形成特征集；S3, the step of extracting activity features, extracting time domain features, frequency domain features and time-frequency domain features in the preprocessed human activity data to form a feature set; S4、最优特征选择步骤，利用特征提取算法对特征集进行提炼、优化，得到最优特征集；S4, the optimal feature selection step, using the feature extraction algorithm to refine and optimize the feature set to obtain the optimal feature set; S5、分类器选择和活动识别步骤，选取分类器，将最优特征集传入分类器中，最终得到分类结果。S5, the classifier selection and activity recognition step, select a classifier, transfer the optimal feature set to the classifier, and finally obtain the classification result. 2.根据权利要求1所述的基于智能终端的相似活动识别方法，其特征在于，所述S1数据采集步骤包括：2. the similar activity recognition method based on intelligent terminal according to claim 1, is characterized in that, described S1 data collection step comprises: 用户佩戴智能终端，并通过智能终端内的加速计传感器实时采集待识别行为的人体活动数据。The user wears a smart terminal and collects real-time human activity data of the behavior to be identified through the accelerometer sensor in the smart terminal. 3.根据权利要求1所述的基于智能终端的相似活动识别方法，其特征在于，所述S2预处理步骤包括：3. the similar activity recognition method based on intelligent terminal according to claim 1, is characterized in that, described S2 preprocessing step comprises: S21、对所采集的人体活动数据进行分段，生成分段信号，使所述分段信号中的每一段信号对应用户的一种行为；S21. Segment the collected human body activity data to generate segmented signals, so that each segment of the segmented signal corresponds to a behavior of the user; S22、利用坐标转换公式将三轴加速度转化为垂直方向和水平方向的加速度，坐标转换公式为：S22. Using the coordinate conversion formula to convert the three-axis acceleration into the acceleration in the vertical direction and the horizontal direction, the coordinate conversion formula is: 其中，A为三轴加速度的矢量和，G为重力加速度，为G和A的夹角，Ax、Ay和Az分别为三轴加速计的输出，Gx、Gy和Gz分别为与X、Y和Z相关的重力加速度，Av为垂直加速度，Ah为水平加速度。Among them, A is the vector sum of three-axis acceleration, G is the acceleration of gravity, is the angle between G and A, Ax, Ay, and Az are the outputs of the three-axis accelerometer, Gx, Gy, and Gz are the gravitational accelerations related to X, Y, and Z, respectively, Av is the vertical acceleration, and Ah is the horizontal acceleration. S23、通过将三轴加速度、垂直方向加速度以及水平方向加速度相结合的方式，构造出一个五维向量，所述五维向量AH＝{Av Ah Ax Ay Az}。S23. Construct a five-dimensional vector by combining the three-axis acceleration, the vertical acceleration and the horizontal acceleration, and the five-dimensional vector AH={Av Ah Ax Ay Az}. 4.根据权利要求3所述的基于智能终端的相似活动识别方法，其特征在于，S21中所述对所采集的人体活动数据进行分段，包括如下步骤：利用加窗来生成与人体活动数据数据所生成的信号相对应的参考窗和测试窗。4. The similar activity recognition method based on an intelligent terminal according to claim 3, wherein the segmenting of the collected human activity data in S21 comprises the steps of: utilizing windowing to generate the same activity data as the human activity data The signal generated by the data corresponds to the reference window and the test window. 5.根据权利要求1所述的基于智能终端的相似活动识别方法，其特征在于，所述S3提取活动特征步骤包括：5. the similar activity recognition method based on intelligent terminal according to claim 1, is characterized in that, described S3 extracts activity characteristic step and comprises: S31、提取常用特征，常用特征包括平均值、方差、标准差、最大值、最小值、相关系数、过均值数、峰值以及功率谱；S31. Extract common features, common features include average value, variance, standard deviation, maximum value, minimum value, correlation coefficient, over-average number, peak value and power spectrum; S32、提取非常用特征，非常用特征包括功率谱密度、四分位数间距和二元树复小波变换。S32. Extract unused features, which include power spectral density, interquartile range, and binary tree complex wavelet transform. 6.根据权利要求5所述的基于智能终端的相似活动识别方法，其特征在于，S32所述提取非常用特征包括如下步骤：6. the similar activity recognition method based on intelligent terminal according to claim 5, is characterized in that, S32 described extracting non-common feature comprises the steps: S321、提取功率谱密度，功率谱密度包括幅度统计和形状统计两大特征，功率谱密度的幅度统计公式为，S321. Extract the power spectral density. The power spectral density includes two major features of amplitude statistics and shape statistics. The amplitude statistical formula of the power spectral density is, 其中，ci是第i个窗口的频幅，μa代表均值，σa代表标准差，γa代表偏度，代表峰度，Among them, ci is the frequency amplitude of the i-th window, μ a represents the mean, σ a represents the standard deviation, γ a represents the skewness, stands for kurtosis, 功率谱密度的形状统计公式为，The shape statistics formula of power spectral density is, 其中，μs代表均值，σs为标准差，γs为偏度，代表峰度， Among them, μ s represents the mean, σ s is the standard deviation, γ s is the skewness, stands for kurtosis, S322、提取四分位数间距；S322. Extract interquartile range; S323、进行二元树复小波变换，小波能量Ew的计算公式为，S323, perform binary tree complex wavelet transform, the calculation formula of wavelet energy Ew is, 其中，N是分解层数，k是时间变量，Djk是高频系数。Among them, N is the number of decomposition layers, k is the time variable, and D jk is the high-frequency coefficient. 7.根据权利要求1所述的基于智能终端的相似活动识别方法，其特征在于，所述S4最优特征选择步骤包括：将Sequential Floating Selection序列浮动前向选择特征选择算法和reliefF算法相结合，对特征集进行降维处理，获取最优特征集。7. the similar activity recognition method based on intelligent terminal according to claim 1, is characterized in that, described S4 optimal feature selection step comprises: Sequential Floating Selection sequence floating forward selection feature selection algorithm and reliefF algorithm are combined, Perform dimensionality reduction processing on the feature set to obtain the optimal feature set. 8.根据权利要求1所述的基于智能终端的相似活动识别方法，其特征在于：S5中所述分类器为SVM、KNN、randomForest或MLP。8. The intelligent terminal-based similar activity recognition method according to claim 1, characterized in that: the classifier in S5 is SVM, KNN, randomForest or MLP. 9.根据权利要求1所述的基于智能终端的相似活动识别方法，其特征在于：S5中所述分类器为MLP。9. The intelligent terminal-based similar activity recognition method according to claim 1, characterized in that: the classifier in S5 is MLP. 10.根据权利要求1所述的基于智能终端的相似活动识别方法，其特征在于：采用10倍交叉验证的方法进行分类。10. The intelligent terminal-based similar activity recognition method according to claim 1, characterized in that: a 10-fold cross-validation method is used for classification.
CN201810706613.5A 2018-07-02 2018-07-02 Similar activity recognition method based on intelligent terminal Pending CN109086667A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810706613.5A CN109086667A (en) 2018-07-02 2018-07-02 Similar activity recognition method based on intelligent terminal

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810706613.5A CN109086667A (en) 2018-07-02 2018-07-02 Similar activity recognition method based on intelligent terminal

Publications (1)

Publication Number Publication Date
CN109086667A true CN109086667A (en) 2018-12-25

Family

ID=64836855

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810706613.5A Pending CN109086667A (en) 2018-07-02 2018-07-02 Similar activity recognition method based on intelligent terminal

Country Status (1)

Country Link
CN (1) CN109086667A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109784418A (en) * 2019-01-28 2019-05-21 东莞理工学院 Human behavior recognition method and system based on feature recombination
CN110569898A (en) * 2019-09-02 2019-12-13 河海大学 A method of human behavior recognition
CN111089604A (en) * 2019-12-10 2020-05-01 中国科学院深圳先进技术研究院 Body-building exercise identification method based on wearable sensor
CN111141284A (en) * 2019-12-28 2020-05-12 西安交通大学 Intelligent building personnel thermal comfort and thermal environment management system and method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009090584A2 (en) * 2008-01-18 2009-07-23 Koninklijke Philips Electronics N.V. Method and system for activity recognition and its application in fall detection
CN103886323A (en) * 2013-09-24 2014-06-25 清华大学 Behavior identification method based on mobile terminal and mobile terminal
CN105137498A (en) * 2015-09-17 2015-12-09 鲁东大学 Underground target detection and recognition system and method based on feature fusion
CN108170274A (en) * 2017-12-29 2018-06-15 南京邮电大学 A kind of action identification method based on wearable device

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009090584A2 (en) * 2008-01-18 2009-07-23 Koninklijke Philips Electronics N.V. Method and system for activity recognition and its application in fall detection
CN103886323A (en) * 2013-09-24 2014-06-25 清华大学 Behavior identification method based on mobile terminal and mobile terminal
CN105137498A (en) * 2015-09-17 2015-12-09 鲁东大学 Underground target detection and recognition system and method based on feature fusion
CN108170274A (en) * 2017-12-29 2018-06-15 南京邮电大学 A kind of action identification method based on wearable device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
杨蓉: "设备位置在智能手机行为识别中的研究与应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109784418A (en) * 2019-01-28 2019-05-21 东莞理工学院 Human behavior recognition method and system based on feature recombination
CN110569898A (en) * 2019-09-02 2019-12-13 河海大学 A method of human behavior recognition
CN111089604A (en) * 2019-12-10 2020-05-01 中国科学院深圳先进技术研究院 Body-building exercise identification method based on wearable sensor
CN111089604B (en) * 2019-12-10 2021-09-07 中国科学院深圳先进技术研究院 Fitness motion recognition method based on wearable sensor
CN111141284A (en) * 2019-12-28 2020-05-12 西安交通大学 Intelligent building personnel thermal comfort and thermal environment management system and method

Similar Documents

Publication Publication Date Title
Ha et al. Convolutional neural networks for human activity recognition using multiple accelerometer and gyroscope sensors
Ge et al. Facial expression recognition based on deep learning
Gurbuz et al. Radar-based human-motion recognition with deep learning: Promising applications for indoor monitoring
Batool et al. Sensors technologies for human activity analysis based on SVM optimized by PSO algorithm
Arora et al. AutoFER: PCA and PSO based automatic facial emotion recognition
Yin et al. A systematic review of human activity recognition based on mobile devices: Overview, progress and trends
Alsheikh et al. Deep activity recognition models with triaxial accelerometers.
Zhong et al. Sensor orientation invariant mobile gait biometrics
Chaudhry et al. Bio-inspired dynamic 3d discriminative skeletal features for human action recognition
Minnen et al. Recognizing and discovering human actions from on-body sensor data
Venkataraman et al. Shape distributions of nonlinear dynamical systems for video-based inference
CN101561868A (en) Human motion emotion identification method based on Gauss feature
Mohammadzade et al. Dynamic time warping-based features with class-specific joint importance maps for action recognition using Kinect depth sensor
Tian et al. Single wearable accelerometer-based human activity recognition via kernel discriminant analysis and QPSO-KELM classifier
CN109086667A (en) Similar activity recognition method based on intelligent terminal
Cirujeda et al. 4DCov: A nested covariance descriptor of spatio-temporal features for gesture recognition in depth sequences
Wu et al. Recognizing activities of the elderly using wearable sensors: a comparison of ensemble algorithms based on boosting
CN102567715A (en) Human body action hierarchical identification method based on pyroelectric infrared detection
Ahmad et al. Multidomain multimodal fusion for human action recognition using inertial sensors
Huan et al. Human complex activity recognition with sensor data using multiple features
Vishwakarma et al. Three‐dimensional human activity recognition by forming a movement polygon using posture skeletal data from depth sensor
Luqian et al. Human activity recognition using time series pattern recognition model-based on tsfresh features
Chen et al. A light-weight deep human activity recognition algorithm using multi-knowledge distillation
Zeng et al. Accelerometer-based gait recognition via deterministic learning
Fu et al. Unconstrained workout activity recognition on unmodified commercial off-the-shelf smartphones

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20181225