WO2017004880A1 - Method, device for behavior recognition and computer storage medium - Google Patents

Method, device for behavior recognition and computer storage medium Download PDF

Info

Publication number
WO2017004880A1
WO2017004880A1 PCT/CN2015/089058 CN2015089058W WO2017004880A1 WO 2017004880 A1 WO2017004880 A1 WO 2017004880A1 CN 2015089058 W CN2015089058 W CN 2015089058W WO 2017004880 A1 WO2017004880 A1 WO 2017004880A1
Authority
WO
WIPO (PCT)
Prior art keywords
layer
wavelet
behavior
approximation coefficient
behavior data
Prior art date
Application number
PCT/CN2015/089058
Other languages
French (fr)
Chinese (zh)
Inventor
贺炎
王斌
王忠民
梁琛
张�荣
宋辉
衡霞
范琳
王文浪
贺菲菲
Original Assignee
中兴通讯股份有限公司
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 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Publication of WO2017004880A1 publication Critical patent/WO2017004880A1/en

Links

Images

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/103Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01PMEASURING LINEAR OR ANGULAR SPEED, ACCELERATION, DECELERATION, OR SHOCK; INDICATING PRESENCE, ABSENCE, OR DIRECTION, OF MOVEMENT
    • G01P13/00Indicating or recording presence, absence, or direction, of movement
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00Input 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
    • G06F3/01Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/03Arrangements for converting the position or the displacement of a member into a coded form
    • G06F3/033Pointing devices displaced or positioned by the user, e.g. mice, trackballs, pens or joysticks; Accessories therefor
    • G06F3/0346Pointing 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

Definitions

  • the invention relates to a behavior recognition technology, in particular to a behavior recognition method, a device and a computer storage medium.
  • the conventional behavior recognition method is as follows: the behavior data of the human body, that is, the sample data, is collected by the sensor configured in the mobile device, the feature data of the collected sample data is optimized, and the behavioral model is used to identify the human behavior.
  • the human body behavior may be daily movements of the human body such as stationary, walking, running, jumping, sitting, standing, and the like. These daily movements of the human body are relatively easy to recognize and have high recognition accuracy due to the large difference between the movements.
  • the human body will also have such subdivided actions such as going upstairs and going downstairs, walking slowly, brisk walking and standing still.
  • the current behavior recognition method is used to identify the accuracy of this subdivision action. Not high, the probability of misidentification is large.
  • the embodiments of the present invention provide a behavior recognition method, a device, and a computer storage medium, which can improve the accuracy of the subdivision motion recognition and reduce the false recognition probability.
  • the embodiment of the invention provides a behavior recognition method, and the method includes:
  • the obtaining behavior data corresponding to the behavior to be identified includes:
  • the obtaining the wavelet feature value of the behavior data includes:
  • N-layer wavelet decomposition on the behavior data to obtain an approximation coefficient of each layer, wherein the approximation coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition layer number, and is a positive integer greater than or equal to 1;
  • a wavelet feature value of each of the at least one predetermined layer is determined according to the low frequency approximation coefficient of the at least one predetermined layer retained.
  • the method includes:
  • At least one of the following characteristic parameters is calculated: the wavelet energy value of each layer in all the layers retained, the average size of the wavelet peak, and the number of wavelet peaks ;
  • the calculated characteristic parameter of each layer is the wavelet feature value of the corresponding layer.
  • determining, according to the wavelet feature value, the behavior to be identified including:
  • Performing an operation of a preset algorithm on the input signal by the classifier model identifies the behavior to be recognized.
  • the embodiment of the invention further provides a behavior recognition device, the device comprising:
  • the first obtaining unit is configured to obtain behavior data corresponding to the behavior to be identified
  • a second acquiring unit configured to acquire a wavelet feature value of the behavior data
  • the first determining unit is configured to determine the to-be-identified behavior according to the wavelet feature value.
  • the first acquiring unit is further configured to:
  • the second acquiring unit is further configured to:
  • N-layer wavelet decomposition on the behavior data to obtain an approximation coefficient of each layer, wherein the approximation coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition layer number, and is a positive integer greater than or equal to 1;
  • a wavelet feature value of each of the at least one predetermined layer is determined according to the low frequency approximation coefficient of the at least one predetermined layer retained.
  • the second acquiring unit is further configured to:
  • At least one of the following characteristic parameters is calculated: the wavelet energy value of each layer in all the layers retained, the average size of the wavelet peak, and the number of wavelet peaks ;
  • the calculated characteristic parameter of each layer is the wavelet feature value of the corresponding layer.
  • the first determining unit is further configured to:
  • Performing an operation of a preset algorithm on the input signal by the classifier model identifies the behavior to be recognized.
  • the embodiment of the invention further provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are used to execute the foregoing behavior recognition method.
  • the behavior recognition method, the device and the computer storage medium provided by the embodiment of the invention acquire the behavior data corresponding to the behavior to be identified; acquire the wavelet feature value of the behavior data; and determine the to-be-identified behavior according to the wavelet feature value.
  • the wavelet feature value is the characteristic parameter in the frequency domain, and the recognition of the human body segmentation action behavior from the frequency domain angle can improve the recognition accuracy and reduce the false recognition probability.
  • FIG. 1 is a schematic flowchart of implementing a behavior recognition method according to an embodiment of the present invention
  • FIG. 2 is a schematic flowchart of an implementation process of acquiring wavelet feature values of behavior data according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of coordinates of a sample serial number and a sample amplitude according to an embodiment of the present invention
  • FIG. 4 is a schematic structural diagram of a composition of a behavior recognition apparatus according to an embodiment of the present invention.
  • the behavior recognition method of the embodiment of the present invention is applied to a mobile device, which may be a mobile phone, a smart watch, a smart bracelet, a smart glasses, or a tablet PC, a personal Digital assistant PDA, e-reader, etc., are not specifically limited herein.
  • FIG. 1 is a schematic flowchart of an implementation of a behavior recognition method according to an embodiment of the present invention; as shown in FIG. 1, the method includes:
  • Step 11 Obtain behavior data corresponding to the behavior to be identified
  • the behavior to be identified is at least two subdivision action actions, such as two subdivision actions, upstairs and downstairs, and also three subdivision actions, such as walking, brisk walking, and in situ, for example, going upstairs and down.
  • the technical solution of the embodiment of the invention lies in how to accurately identify each subdivision action behavior with high accuracy.
  • the behavior to be identified includes an upstairs and a downstairs
  • the behavior data is an acceleration signal generated by the human body under certain behaviors, such as an acceleration generated when going upstairs and an acceleration generated when going downstairs.
  • the acceleration signal is preferably a synthetic acceleration signal.
  • the acceleration signal can be sensed by an acceleration sensor built into the mobile device.
  • the acceleration sensor is a three-axis acceleration sensor
  • the acceleration in the X direction, the Y direction, and the Z direction in the XYZ three-dimensional coordinate system is sensed by the three-axis acceleration sensor, and the accelerations in the three directions are combined.
  • the resulting synthetic acceleration signal is obtained.
  • the acceleration sensor adopts an acceleration signal under a certain behavior at a predetermined sampling frequency.
  • the acquisition frequency is 50 times/s
  • the behavior data of the human body in the current behavior is collected for 1 s, and the acceleration data is collected within 1 s.
  • the acquiring behavior data corresponding to the behavior to be identified may be acquiring an acceleration signal generated by the human body when the to-be-identified behavior occurs in a predetermined collection period at a predetermined sampling frequency; determining any collected
  • the acceleration signal is the behavior data.
  • the sampling frequency and the sampling period can be flexibly set according to actual conditions.
  • Step 12 Acquire a wavelet feature value of the behavior data.
  • the wavelet feature value includes at least one of the following characteristic parameters: wavelet energy after wavelet decomposition of the behavior data, average size of the wavelet peak, and number of wavelet peaks.
  • the step 12 further includes:
  • Step 121 performing N-layer wavelet decomposition on the behavior data to obtain an approximate coefficient of each layer, where the approximate coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient;
  • N is a positive integer greater than or equal to 1, which can be flexibly set according to experience.
  • the maximum value of N needs to satisfy the signal in the largest layer when the behavior data is decomposed into the largest layer, which can be expressed as a complete behavior. As indicated as a complete upstairs action.
  • the wavelet transform function is called, and the current behavior data collected by the acceleration sensor, the pre-selected wavelet basis function, and the wavelet decomposition layer number N are input as three input signals of the wavelet transform function to The wavelet transform function, the output of the wavelet transform function, that is, the behavior data is decomposed by wavelet into the approximate coefficients of each layer of the N layer.
  • the approximation coefficient of each layer includes at least the low frequency approximation coefficient and the high frequency approximation coefficient; wherein, the low frequency approximation coefficient is that the behavior data of the signal that is decomposed into the layer changes slowly can be embodied as the contour part of the behavior signal, and the high frequency approximation coefficient is decomposed
  • the behavioral data that changes rapidly to the layer's signal can be embodied as a detail part of the behavioral signal.
  • the behavior data is subjected to N-layer wavelet decomposition, and the behavior data is wavelet-transformed into the frequency domain.
  • the wavelet transform based on multi-resolution analysis is preferably used to perform wavelet decomposition on the behavior data, and the selected wavelet basis function is Hal wavelet.
  • the selected wavelet basis function is Hal wavelet.
  • Step 122 retain a low frequency approximation coefficient of at least one predetermined layer
  • the high frequency approximation coefficient is a detail portion of the signal, the detail portion is generally regarded as noise, and in the embodiment of the present invention, the high frequency approximation coefficient of all layers is discarded; the low frequency approximation coefficient of at least one predetermined layer is retained.
  • the low frequency approximation coefficients of which layer or layers are retained are usually obtained empirically.
  • Step 123 Determine a wavelet feature value of each layer in the at least one predetermined layer according to the retained low frequency approximation coefficient of the at least one predetermined layer;
  • the wavelet feature value includes at least one of the following: wavelet energy of each predetermined layer after the wavelet decomposition of the behavior data, an average size of the wavelet peaks, and a number of wavelet peaks.
  • the wavelet feature value includes the foregoing three parameters as an example.
  • the acceleration sensor adopts an acceleration signal under a certain behavior at a predetermined sampling frequency.
  • the acquisition frequency is 50 times/s
  • the behavior data of the human body in the current behavior is collected for 1 s
  • the acceleration data is collected within 1 s.
  • the 50 acceleration signals are correspondingly identified as the first sample signal, the second sample signal, and the 50th sample signal. As shown in Fig.
  • the XOY axis coordinate system the X axis represents the sample signal sequence number from 1 to 50, and the ordinate represents the amplitude of one of the predetermined layers after the wavelet decomposition of the 50 sample signals, that is, the low frequency approximation coefficient of the layer.
  • the magnitude which is a discrete value, is represented by a solid dot.
  • FIG. 3 it is first found that there are a plurality of amplitude peaks in the amplitude corresponding to so many sample signals, and the first amplitude threshold is preset, as shown by a straight line parallel to the X-axis in FIG.
  • the amplitude corresponding to the sample point is the wavelet peak of the layer, and according to this method, the 50 layers in the layer can be found.
  • the average size of the wavelet peaks for this layer is equal to the sum of all wavelet peaks divided by the total number of wavelet peaks for that layer.
  • the wavelet energy of this layer is equal to the sum of the squares of the wavelet coefficients obtained by the wavelet transform of the behavior data.
  • the average size of wavelet peaks and the number of wavelet peaks please refer to the existing related description for the concept of wavelet coefficients, which will not be repeated here.
  • the average size of the peak of the wavelet and the number of peaks of the wavelet are characterized by small peaks.
  • the wavelet energy distribution feature can reflect the energy distribution of the two behaviors of the upper and lower buildings on different wavelet decomposition layers.
  • the wavelet peak characteristics can reflect the amplitude of the acceleration signal generated by the human body under these two behaviors.
  • the combination of these two features can significantly improve the recognition accuracy of behaviors compared with the related techniques in which human behavior is recognized only by time domain characteristics such as mean square error, variance, and the like.
  • the wavelet energy, the average size of the wavelet peaks, and the number of wavelet peaks are characteristic parameters in the frequency domain.
  • the recognition of the human behavior is realized by a characteristic parameter in the frequency domain or a combination of at least two characteristic parameters. .
  • Step 13 Determine the behavior to be identified according to the wavelet feature value.
  • a classifier model is pre-trained, and wavelets of each layer in the at least one predetermined layer are used.
  • the energy, the average size of the wavelet peaks, and the number of wavelet peaks are input to the classifier model as the input signal of the classifier model.
  • the wavelet energy of each layer calculated in the second, third, and fourth layers, and the average size of the wavelet peaks.
  • the number of wavelet peaks is input to the classifier model as an input signal, and the operation of the preset signal is performed by the classifier model to identify the to-be-identified behavior;
  • the preset algorithm may be a decision tree, SVM or speed learning machine algorithm. That is, through the classifier model, it is possible to identify which behavior of the human body to be identified.
  • the input of the classifier model includes behavior data with behavior tags, and the behavior data passes through the N layer wavelet.
  • the wavelet energy of each layer of the second to fourth layers, the average size of the wavelet peaks, and the number of wavelet peaks, using these input signals the parameters of the classifier model are adjusted accordingly, and the parameters of the classifier model are adjusted.
  • the output of the classifier model is consistent with the behavior indicated by the behavior tag, the classifier model is trained at this time.
  • the process of training a classifier model includes: collecting behavior data in an upstairs behavior by a sensor at a predetermined acquisition frequency, and collecting behavior data in a downstairs behavior multiple times, and characterizing the same behavior.
  • the behavior data is recorded in the same file.
  • the acquisition frequency is For example, 50 times/s
  • the behavior data of the human body in the behavior of going upstairs is continuously collected for 10s
  • these 500 behaviors are collected.
  • the data is stored in a first file that records the behavior data of the upstairs collected by the human body when it is in the upstairs behavior. Continuously collect the behavior data of the human body in the behavior of going downstairs for 5s.
  • the second document records the behavior data of the downstairs collected when the human body is in the downstairs behavior. It should be noted that those skilled in the art should know that if the upstairs and the downstairs are regarded as different types of behaviors, the number of files storing different behavior data should be the same as the type of behavior to be identified. It can be seen from the above that the behavior data stored in the first file and the second file is marked with a behavior label (the behavior label can distinguish the behavior corresponding to the behavior data), that is, the behavior stored in the first file. The data corresponds to the behavior of going upstairs, and the behavior data stored in the second file corresponds to the behavior of going downstairs.
  • the classifier model trains the parameters of the model. When the parameters of the classifier model are adjusted such that the output of the classifier model is consistent with the behavior indicated by the behavior tag, the classifier model is well trained.
  • the classifier model After the classifier model is trained, when a certain behavior data is collected by the acceleration sensor, it will be performed according to the aforementioned steps 11 to 13 to identify the behavior corresponding to the behavior data through the trained classifier model.
  • a certain behavior data For the specific training process of the parameters of the classifier model and the concept of the training data set, please refer to the existing related description, which is not described here.
  • the description is made by taking the upper floor and the lower building as an example, and in addition, considering that there is a level between the stairs and the stairs in practical applications, the embodiment can also recognize the upstairs, the downstairs, and the walk.
  • Three subdivided action behaviors can also identify three types: slow walking, fast walking, and in situ. Subdivide action behavior.
  • the collected behavior data is transformed into the frequency domain by wavelet decomposition, and the human behavior recognition is performed from the frequency domain angle, which can improve the recognition accuracy and reduce the false recognition probability compared with the simple human body behavior recognition from the time domain angle. ;
  • the classifier model is The training data set used in the pre-training process is at least one characteristic parameter in the frequency domain, so that the accuracy of the trained classifier model is improved, and the highly accurate classifier model is used to identify the subdivided human behavior, which can be improved. Identify accuracy and reduce the probability of misrecognition.
  • the embodiment of the present invention further provides a behavior recognition device.
  • the device includes: a first acquisition unit 401, a second acquisition unit 402, and a first determination unit 403; ,
  • the first obtaining unit 401 is configured to acquire behavior data corresponding to the behavior to be identified;
  • the second obtaining unit 402 is configured to acquire a wavelet feature value of the behavior data.
  • the first determining unit 403 is configured to determine the to-be-identified behavior according to the wavelet feature value.
  • the first obtaining unit 401 is further configured to: collect, in a predetermined sampling period, an acceleration signal generated by the human body when the to-be-identified behavior occurs in a predetermined sampling period; and determine that any acquired acceleration signal is Behavioral data.
  • the second obtaining unit 402 is further configured to perform N-layer wavelet decomposition on the behavior data to obtain an approximate coefficient of each layer, where the approximate coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition.
  • the number of layers is a positive integer greater than or equal to 1; a low frequency approximation coefficient of at least one predetermined layer is retained; and wavelet feature values of each of the at least one predetermined layer are determined according to the retained low frequency approximation coefficients of the at least one predetermined layer.
  • the 6-layer wavelet decomposition is performed in the wavelet transform function to obtain the high-frequency approximation coefficient and the low-frequency approximation coefficient of each layer; the low-frequency approximation coefficients of the 2nd, 3rd, and 4th layers are retained; the second layer, the third layer, and the The low-frequency approximation coefficient of the 4 layers is calculated by calculating at least one of the following characteristic parameters: the wavelet energy value of each layer in all the layers retained, the average size of the wavelet peaks, and the number of wavelet peaks; determining the calculated characteristics of each layer
  • the parameters are the wavelet eigenvalues of the respective layers.
  • the first determining unit 403 is further configured to: determine that the feature parameters of all layers are input signals of the pre-trained classifier model; input the input signal to the classifier model; and pass the classifier model Performing an operation of a preset algorithm on the input signal identifies the behavior to be recognized.
  • the embodiment of the present invention further provides a behavior recognition device. Since the principle and method for solving the problem are similar, the implementation process and implementation principle of the behavior recognition device can refer to the implementation process and implementation of the foregoing method. The principle description, the repetition will not be repeated.
  • the first obtaining unit 401, the second obtaining unit 402, and the first determining unit 403 may each be a central processing unit (CPU), or a digital signal processing (DSP), or It is implemented by a microprocessor (MPU, Micro Processor Unit) or a Field Programmable Gate Array (FPGA).
  • CPU central processing unit
  • DSP digital signal processing
  • MPU Microprocessor
  • FPGA Field Programmable Gate Array
  • the embodiment of the invention further provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are used to execute the foregoing behavior recognition method.
  • embodiments of the present invention can be provided as a method, system, Or a computer program product. Accordingly, the present invention can take the form of a hardware embodiment, a software embodiment, or a combination of software and hardware. Moreover, the invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) including computer usable program code.
  • the computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device.
  • the apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.
  • These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device.
  • the instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.
  • the behavior data corresponding to the behavior to be identified is acquired first, and then the behavior data is acquired.
  • a wavelet feature value and determining the to-be-identified behavior according to the wavelet feature value.
  • the wavelet feature value is a characteristic parameter in the frequency domain, and the recognition of the human body segmentation action behavior from the frequency domain angle can improve the recognition accuracy and reduce the false recognition probability.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Pathology (AREA)
  • Molecular Biology (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Physiology (AREA)
  • Biophysics (AREA)
  • Human Computer Interaction (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Dentistry (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Character Discrimination (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

A method, a device for behavior recognition and a computer storage medium, the method includes: behavior data corresponding to the behavior to be recognized is obtained (11); a value for a set of wavelets of the behavior data is obtained (12); according to the value for a set of wavelets, the behavior to be recognized is determined (13).

Description

行为识别方法、设备及计算机存储介质Behavior recognition method, device and computer storage medium 技术领域Technical field
本发明涉及到行为识别技术,具体涉及一种行为识别方法、设备及计算机存储介质。The invention relates to a behavior recognition technology, in particular to a behavior recognition method, a device and a computer storage medium.
背景技术Background technique
近年来,随着手机、智能手表、智能手环等移动设备的普及,基于移动设备的人体行为识别技术成为了研究热点。其中,惯用的行为识别方法如下所述:通过移动设备中配置的传感器来采集人体的行为数据即样本数据,对所采集的样本数据进行特征优选,并通过建立的行为模型来识别人体行为。所述人体行为可以为静止、走路、跑步、跳、坐、站立等人体日常动作。这些人体日常动作由于动作间的差异较大,所以比较容易识别且识别准确度较高。但是考虑到在实际应用中,人体也会出现诸如上楼和下楼,慢走、快走和原地踏步等这样细分的动作,目前的行为识别方法对于识别这种细分动作的准确度并不高、误识别概率较大。In recent years, with the popularization of mobile devices such as mobile phones, smart watches and smart bracelets, human behavior recognition technology based on mobile devices has become a research hotspot. Among them, the conventional behavior recognition method is as follows: the behavior data of the human body, that is, the sample data, is collected by the sensor configured in the mobile device, the feature data of the collected sample data is optimized, and the behavioral model is used to identify the human behavior. The human body behavior may be daily movements of the human body such as stationary, walking, running, jumping, sitting, standing, and the like. These daily movements of the human body are relatively easy to recognize and have high recognition accuracy due to the large difference between the movements. However, considering the actual application, the human body will also have such subdivided actions such as going upstairs and going downstairs, walking slowly, brisk walking and standing still. The current behavior recognition method is used to identify the accuracy of this subdivision action. Not high, the probability of misidentification is large.
发明内容Summary of the invention
为解决现有存在的技术问题,本发明实施例提供一种行为识别方法、设备及计算机存储介质,能够提高对细分动作识别的准确度,减少误识别概率。In order to solve the existing technical problems, the embodiments of the present invention provide a behavior recognition method, a device, and a computer storage medium, which can improve the accuracy of the subdivision motion recognition and reduce the false recognition probability.
本发明实施例的技术方案是这样实现的:The technical solution of the embodiment of the present invention is implemented as follows:
本发明实施例提供了一种行为识别方法,所述方法包括:The embodiment of the invention provides a behavior recognition method, and the method includes:
获取待识别行为对应的行为数据; Obtaining behavior data corresponding to the behavior to be identified;
获取所述行为数据的小波特征值;Obtaining a wavelet feature value of the behavior data;
依据所述小波特征值,确定所述待识别行为。Determining the behavior to be identified according to the wavelet feature value.
前述方案中,所述获取待识别行为对应的行为数据,包括:In the foregoing solution, the obtaining behavior data corresponding to the behavior to be identified includes:
以预定的采样频率在预定的采集周期内采集人体在发生所述待识别行为时产生的加速度信号;Acquiring an acceleration signal generated by the human body when the behavior to be recognized occurs in a predetermined sampling period at a predetermined sampling frequency;
确定所采集到的任意加速度信号为所述行为数据。It is determined that any acquired acceleration signal is the behavior data.
前述方案中,所述获取所述行为数据的小波特征值,包括:In the foregoing solution, the obtaining the wavelet feature value of the behavior data includes:
将所述行为数据进行N层小波分解,得到每层的近似系数,所述近似系数包括高频近似系数和低频近似系数,N为小波分解层数、为大于等于1的正整数;Performing N-layer wavelet decomposition on the behavior data to obtain an approximation coefficient of each layer, wherein the approximation coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition layer number, and is a positive integer greater than or equal to 1;
保留至少一个预定层的低频近似系数;Retaining a low frequency approximation coefficient of at least one predetermined layer;
依据所保留的至少一个预定层的低频近似系数,确定所述至少一个预定层中每层的小波特征值。A wavelet feature value of each of the at least one predetermined layer is determined according to the low frequency approximation coefficient of the at least one predetermined layer retained.
前述方案中,所述方法包括:In the foregoing solution, the method includes:
至少将所述行为数据、预定的小波分解层数N=6输入至预先设定的小波变换函数中进行6层小波分解,得到每层的高频近似系数和低频近似系数;At least the behavior data, the predetermined wavelet decomposition layer number N=6 is input into a preset wavelet transform function to perform 6-layer wavelet decomposition, to obtain a high-frequency approximation coefficient and a low-frequency approximation coefficient of each layer;
保留第2层、第3层及第4层的低频近似系数;Retain the low frequency approximation coefficients of Layer 2, Layer 3 and Layer 4;
利用第2层、第3层及第4层的低频近似系数,计算以下其中至少一种特征参数:所保留下的所有层中每层的小波能量值、小波峰值的平均大小、小波峰值个数;Using the low frequency approximation coefficients of the second layer, the third layer, and the fourth layer, at least one of the following characteristic parameters is calculated: the wavelet energy value of each layer in all the layers retained, the average size of the wavelet peak, and the number of wavelet peaks ;
确定所计算出的每层的特征参数为相应层的所述小波特征值。It is determined that the calculated characteristic parameter of each layer is the wavelet feature value of the corresponding layer.
前述方案中,所述依据所述小波特征值,确定所述待识别行为,包括:In the foregoing solution, determining, according to the wavelet feature value, the behavior to be identified, including:
确定所有层的特征参数为预先训练的分类器模型的输入信号;Determining the characteristic parameters of all layers as input signals of the pre-trained classifier model;
将所述输入信号输入至所述分类器模型; Inputting the input signal to the classifier model;
通过所述分类器模型对所述输入信号进行预设算法的运算识别出所述待识别行为。Performing an operation of a preset algorithm on the input signal by the classifier model identifies the behavior to be recognized.
本发明实施例还提供了一种行为识别设备,所述设备包括:The embodiment of the invention further provides a behavior recognition device, the device comprising:
第一获取单元,配置为获取待识别行为对应的行为数据;The first obtaining unit is configured to obtain behavior data corresponding to the behavior to be identified;
第二获取单元,配置为获取所述行为数据的小波特征值;a second acquiring unit, configured to acquire a wavelet feature value of the behavior data;
第一确定单元,配置为依据所述小波特征值,确定所述待识别行为。The first determining unit is configured to determine the to-be-identified behavior according to the wavelet feature value.
前述方案中,所述第一获取单元,还配置为:In the foregoing solution, the first acquiring unit is further configured to:
以预定的采样频率在预定的采集周期内采集人体在发生所述待识别行为时产生的加速度信号;Acquiring an acceleration signal generated by the human body when the behavior to be recognized occurs in a predetermined sampling period at a predetermined sampling frequency;
确定所采集到的任意加速度信号为所述行为数据。It is determined that any acquired acceleration signal is the behavior data.
前述方案中,所述第二获取单元,还配置为:In the foregoing solution, the second acquiring unit is further configured to:
将所述行为数据进行N层小波分解,得到每层的近似系数,所述近似系数包括高频近似系数和低频近似系数,N为小波分解层数、为大于等于1的正整数;Performing N-layer wavelet decomposition on the behavior data to obtain an approximation coefficient of each layer, wherein the approximation coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition layer number, and is a positive integer greater than or equal to 1;
保留至少一个预定层的低频近似系数;Retaining a low frequency approximation coefficient of at least one predetermined layer;
依据所保留的至少一个预定层的低频近似系数,确定所述至少一个预定层中每层的小波特征值。A wavelet feature value of each of the at least one predetermined layer is determined according to the low frequency approximation coefficient of the at least one predetermined layer retained.
前述方案中,所述第二获取单元,还配置为:In the foregoing solution, the second acquiring unit is further configured to:
至少将所述行为数据、预定的小波分解层数N=6输入至预先设定的小波变换函数中进行6层小波分解,得到每层的高频近似系数和低频近似系数;At least the behavior data, the predetermined wavelet decomposition layer number N=6 is input into a preset wavelet transform function to perform 6-layer wavelet decomposition, to obtain a high-frequency approximation coefficient and a low-frequency approximation coefficient of each layer;
保留第2层、第3层及第4层的低频近似系数;Retain the low frequency approximation coefficients of Layer 2, Layer 3 and Layer 4;
利用第2层、第3层及第4层的低频近似系数,计算以下其中至少一种特征参数:所保留下的所有层中每层的小波能量值、小波峰值的平均大小、小波峰值个数; Using the low frequency approximation coefficients of the second layer, the third layer, and the fourth layer, at least one of the following characteristic parameters is calculated: the wavelet energy value of each layer in all the layers retained, the average size of the wavelet peak, and the number of wavelet peaks ;
确定所计算出的每层的特征参数为相应层的所述小波特征值。It is determined that the calculated characteristic parameter of each layer is the wavelet feature value of the corresponding layer.
前述方案中,所述第一确定单元,还配置为:In the foregoing solution, the first determining unit is further configured to:
确定所有层的特征参数为预先训练的分类器模型的输入信号;Determining the characteristic parameters of all layers as input signals of the pre-trained classifier model;
将所述输入信号输入至所述分类器模型;Inputting the input signal to the classifier model;
通过所述分类器模型对所述输入信号进行预设算法的运算识别出所述待识别行为。Performing an operation of a preset algorithm on the input signal by the classifier model identifies the behavior to be recognized.
本发明实施例还提供了一种计算机存储介质,所述计算机存储介质中存储有计算机可执行指令,所述计算机可执行指令用于执行前述的行为识别方法。The embodiment of the invention further provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are used to execute the foregoing behavior recognition method.
本发明实施例提供的行为识别方法、设备及计算机存储介质,获取待识别行为对应的行为数据;获取所述行为数据的小波特征值;依据所述小波特征值,确定所述待识别行为。小波特征值为频域上的特征参数,从频域角度进行人体细分动作行为的识别,可提高识别准确性,减少误识别概率。The behavior recognition method, the device and the computer storage medium provided by the embodiment of the invention acquire the behavior data corresponding to the behavior to be identified; acquire the wavelet feature value of the behavior data; and determine the to-be-identified behavior according to the wavelet feature value. The wavelet feature value is the characteristic parameter in the frequency domain, and the recognition of the human body segmentation action behavior from the frequency domain angle can improve the recognition accuracy and reduce the false recognition probability.
附图说明DRAWINGS
图1为本发明实施例的行为识别方法的实现流程示意图;1 is a schematic flowchart of implementing a behavior recognition method according to an embodiment of the present invention;
图2为本发明实施例的获取行为数据的小波特征值的实现流程示意图;2 is a schematic flowchart of an implementation process of acquiring wavelet feature values of behavior data according to an embodiment of the present invention;
图3为本发明实施例的样本序号与样本幅值的坐标示意图;3 is a schematic diagram of coordinates of a sample serial number and a sample amplitude according to an embodiment of the present invention;
图4为本发明实施例的行为识别设备的组成结构示意图。FIG. 4 is a schematic structural diagram of a composition of a behavior recognition apparatus according to an embodiment of the present invention.
具体实施方式detailed description
本发明实施例的行为识别方法,应用于一移动设备中,该移动设备可以手机、智能手表、智能手环、智能眼镜,也可以为平板电脑PAD、个人 数字助理PDA、电子阅读器等,此处不做具体限定。The behavior recognition method of the embodiment of the present invention is applied to a mobile device, which may be a mobile phone, a smart watch, a smart bracelet, a smart glasses, or a tablet PC, a personal Digital assistant PDA, e-reader, etc., are not specifically limited herein.
图1为本发明实施例的行为识别方法的实现流程示意图;如图1所示,所述方法包括:1 is a schematic flowchart of an implementation of a behavior recognition method according to an embodiment of the present invention; as shown in FIG. 1, the method includes:
步骤11:获取待识别行为对应的行为数据;Step 11: Obtain behavior data corresponding to the behavior to be identified;
所述待识别行为为至少两种细分动作行为,例如上楼和下楼这两种细分动作,还例如慢走、快走和原地踏步这三种细分动作,还例如上楼、下楼和步行这三种细分动作。本发明实施例的技术方案就在于如何高准确度的识别每种细分动作行为。The behavior to be identified is at least two subdivision action actions, such as two subdivision actions, upstairs and downstairs, and also three subdivision actions, such as walking, brisk walking, and in situ, for example, going upstairs and down. The three subdivisions of the building and the walk. The technical solution of the embodiment of the invention lies in how to accurately identify each subdivision action behavior with high accuracy.
本实施例中,以待识别行为包括上楼和下楼为例,所述行为数据为人体在某种行为下产生的加速度信号,例如在上楼时产生的加速度、下楼时产生的加速度,该加速度信号优选为合成加速度信号。该加速度信号可通过内置在移动设备中的加速度传感器而感应到。当该加速度传感器为三轴加速度传感器时,通过该三轴加速度传感器感应到的是XYZ三维坐标系下在X方向、Y方向和Z方向上的加速度,并将这三个方向上的加速度进行合成得到的所述合成加速度信号。In this embodiment, the behavior to be identified includes an upstairs and a downstairs, and the behavior data is an acceleration signal generated by the human body under certain behaviors, such as an acceleration generated when going upstairs and an acceleration generated when going downstairs. The acceleration signal is preferably a synthetic acceleration signal. The acceleration signal can be sensed by an acceleration sensor built into the mobile device. When the acceleration sensor is a three-axis acceleration sensor, the acceleration in the X direction, the Y direction, and the Z direction in the XYZ three-dimensional coordinate system is sensed by the three-axis acceleration sensor, and the accelerations in the three directions are combined. The resulting synthetic acceleration signal is obtained.
通常,加速度传感器以预定的采样频率采用处于某种行为下的加速度信号,如以采集频率为50次/s为例,采集人体在发生当前行为下的行为数据1s,在这1s内共采集到50次/s*1s=50个的行为数据即加速度信号,视这1s为预定的一个采样周期。本实施例中,所述获取待识别行为对应的行为数据可以为以预定的采样频率在预定的采集周期内采集人体在发生所述待识别行为时产生的加速度信号;确定所采集到的任意个加速度信号为所述行为数据。并通过后续的技术方案识别该行为具体为哪种行为如为上楼行为还是下楼行为。其中,所述采样频率与采样周期可根据实际情况而灵活设定。Generally, the acceleration sensor adopts an acceleration signal under a certain behavior at a predetermined sampling frequency. For example, the acquisition frequency is 50 times/s, and the behavior data of the human body in the current behavior is collected for 1 s, and the acceleration data is collected within 1 s. The behavior data of 50 times/s*1s=50 is the acceleration signal, and this 1s is regarded as a predetermined sampling period. In this embodiment, the acquiring behavior data corresponding to the behavior to be identified may be acquiring an acceleration signal generated by the human body when the to-be-identified behavior occurs in a predetermined collection period at a predetermined sampling frequency; determining any collected The acceleration signal is the behavior data. And through the subsequent technical solutions to identify which behavior is specifically for the behavior of going upstairs or going downstairs. The sampling frequency and the sampling period can be flexibly set according to actual conditions.
步骤12:获取所述行为数据的小波特征值; Step 12: Acquire a wavelet feature value of the behavior data.
这里,所述小波特征值包括以下至少其中一种特征参数:行为数据经小波分解后的小波能量、小波峰值的平均大小、小波峰值个数。Here, the wavelet feature value includes at least one of the following characteristic parameters: wavelet energy after wavelet decomposition of the behavior data, average size of the wavelet peak, and number of wavelet peaks.
如图2所示,所述步骤12进一步包括:As shown in FIG. 2, the step 12 further includes:
步骤121:将所述行为数据进行N层小波分解,得到每层的近似系数,所述近似系数包括高频近似系数和低频近似系数;Step 121: performing N-layer wavelet decomposition on the behavior data to obtain an approximate coefficient of each layer, where the approximate coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient;
这里,N为大于等于1的正整数,可根据经验而灵活设定,通常N的最大取值需要满足在将行为数据分解为最大层时该最大层中的信号能够表示为一个完整的行为动作如表示为一个完整的上楼动作。本实施例中,优选小波分解层数N=6。在预先配置好的库函数中,调用小波变换函数,并将通过加速度传感器采集到的当前行为数据、预先选取好的小波基函数以及小波分解层数N作为小波变换函数的三个输入信号输入至小波变换函数,小波变换函数的输出即行为数据在经小波分解为N层每层的近似系数。每层的近似系数至少包括低频近似系数和高频近似系数;其中,低频近似系数为被分解到该层的信号变化缓慢的行为数据可体现为行为信号的轮廓部分,高频近似系数为被分解到该层的信号变化迅速的行为数据可体现为行为信号的细节部分。本步骤中将行为数据进行N层小波分解即将行为数据经小波变换至频域上。Here, N is a positive integer greater than or equal to 1, which can be flexibly set according to experience. Usually, the maximum value of N needs to satisfy the signal in the largest layer when the behavior data is decomposed into the largest layer, which can be expressed as a complete behavior. As indicated as a complete upstairs action. In the present embodiment, the number of wavelet decomposition layers is preferably N=6. In the pre-configured library function, the wavelet transform function is called, and the current behavior data collected by the acceleration sensor, the pre-selected wavelet basis function, and the wavelet decomposition layer number N are input as three input signals of the wavelet transform function to The wavelet transform function, the output of the wavelet transform function, that is, the behavior data is decomposed by wavelet into the approximate coefficients of each layer of the N layer. The approximation coefficient of each layer includes at least the low frequency approximation coefficient and the high frequency approximation coefficient; wherein, the low frequency approximation coefficient is that the behavior data of the signal that is decomposed into the layer changes slowly can be embodied as the contour part of the behavior signal, and the high frequency approximation coefficient is decomposed The behavioral data that changes rapidly to the layer's signal can be embodied as a detail part of the behavioral signal. In this step, the behavior data is subjected to N-layer wavelet decomposition, and the behavior data is wavelet-transformed into the frequency domain.
本实施例中优选采用基于多分辨率分析的小波变换对行为数据进行小波分解,所选用的小波基函数为哈尔小波。关于前述的高频、低频近似系数以及小波基函数的概念及用途请参见现有相关说明,此次不再赘述。In this embodiment, the wavelet transform based on multi-resolution analysis is preferably used to perform wavelet decomposition on the behavior data, and the selected wavelet basis function is Hal wavelet. For the above concepts and uses of the high frequency, low frequency approximation coefficients and wavelet basis functions, please refer to the existing related descriptions, and will not be repeated here.
步骤122:保留至少一个预定层的低频近似系数;Step 122: retain a low frequency approximation coefficient of at least one predetermined layer;
这里,因为高频近似系数为信号的细节部分,细节部分通常被视为噪声,本发明实施例中舍弃所有层的高频近似系数;保留至少一个预定层的低频近似系数。以N=6为例,当行为数据进行N=6层的小波分解后,在所得到第1~6层的低频近似系数中,选取第2、3、4层的低频近似系数作为 所保留下的至少一个预定层的低频近似系数。保留下哪个或哪几个层的低频近似系数通常根据经验而获得。Here, since the high frequency approximation coefficient is a detail portion of the signal, the detail portion is generally regarded as noise, and in the embodiment of the present invention, the high frequency approximation coefficient of all layers is discarded; the low frequency approximation coefficient of at least one predetermined layer is retained. Taking N=6 as an example, when the behavior data is subjected to wavelet decomposition of N=6 layers, the low-frequency approximation coefficients of the second, third and fourth layers are selected as the low-frequency approximation coefficients of the first to sixth layers. A low frequency approximation coefficient of at least one predetermined layer retained. The low frequency approximation coefficients of which layer or layers are retained are usually obtained empirically.
步骤123:依据所保留的至少一个预定层的低频近似系数,确定所述至少一个预定层中每层的小波特征值;Step 123: Determine a wavelet feature value of each layer in the at least one predetermined layer according to the retained low frequency approximation coefficient of the at least one predetermined layer;
所述小波特征值包括以下至少其中一种:行为数据经小波分解后的每个预定层的小波能量、小波峰值的平均大小、小波峰值个数。本实施例中,以小波特征值包括前述三个参数为例。通常,加速度传感器以预定的采样频率采用处于某种行为下的加速度信号,如以采集频率为50次/s为例,采集人体在发生当前行为下的行为数据1s,在这1s内共采集到50次/s*1s=50个的行为数据即加速度信号,将这50个加速度信号对应标识为第1个样本信号、第2个样本信号…第50个样本信号。如图3所示的XOY轴坐标系,X轴代表样本信号的序号为从1到50,纵坐标代表这50个样本信号经小波分解后其中一个预定层的幅值即该层的低频近似系数的幅值、为离散值,用实心圆点来表示。在图3中,先找出在这么多的样本信号对应的幅值中共有多个幅度峰值,预先设定第一幅度阈值,如图3中平行于X轴的直线所示,某个样本点的前一个样本点与后一个样本点各自所对应的幅度均超过第一幅度阈值时,视该样本点对应的幅度为该层的小波峰值,依据此方法可查找出在该层的这50个样本信号中的所有小波峰值的个数,以及所有小波峰值的大小。该层的小波峰值的平均大小等于所有小波峰值之和除以该层的小波峰值的总个数。该层的小波能量等于行为数据经小波变换得到的小波系数的平方和。对于小波能量、小波峰值的平均大小及小波峰值个数的计算方法,小波系数的概念具体请参见现有相关说明,此次不再赘述。The wavelet feature value includes at least one of the following: wavelet energy of each predetermined layer after the wavelet decomposition of the behavior data, an average size of the wavelet peaks, and a number of wavelet peaks. In this embodiment, the wavelet feature value includes the foregoing three parameters as an example. Generally, the acceleration sensor adopts an acceleration signal under a certain behavior at a predetermined sampling frequency. For example, the acquisition frequency is 50 times/s, and the behavior data of the human body in the current behavior is collected for 1 s, and the acceleration data is collected within 1 s. The behavior data of 50 times/s*1s=50 is the acceleration signal, and the 50 acceleration signals are correspondingly identified as the first sample signal, the second sample signal, and the 50th sample signal. As shown in Fig. 3, the XOY axis coordinate system, the X axis represents the sample signal sequence number from 1 to 50, and the ordinate represents the amplitude of one of the predetermined layers after the wavelet decomposition of the 50 sample signals, that is, the low frequency approximation coefficient of the layer. The magnitude, which is a discrete value, is represented by a solid dot. In FIG. 3, it is first found that there are a plurality of amplitude peaks in the amplitude corresponding to so many sample signals, and the first amplitude threshold is preset, as shown by a straight line parallel to the X-axis in FIG. When the amplitude corresponding to each of the previous sample point and the latter sample point exceeds the first amplitude threshold, the amplitude corresponding to the sample point is the wavelet peak of the layer, and according to this method, the 50 layers in the layer can be found. The number of all wavelet peaks in the sample signal, as well as the size of all wavelet peaks. The average size of the wavelet peaks for this layer is equal to the sum of all wavelet peaks divided by the total number of wavelet peaks for that layer. The wavelet energy of this layer is equal to the sum of the squares of the wavelet coefficients obtained by the wavelet transform of the behavior data. For the calculation method of wavelet energy, the average size of wavelet peaks and the number of wavelet peaks, please refer to the existing related description for the concept of wavelet coefficients, which will not be repeated here.
其中,称小波峰值的平均大小和小波峰值个数为小波峰特征。小波能量分布特征可反映上楼、下楼这两种行为在不同小波分解层上的能量分布情况,小波峰特征可反映人体在这两种行为下产生的加速度信号的幅度, 这两种特征的结合比相关技术中仅通过时域特性如均方差、方差等来识别人体行为相比较,可以显著提高行为的识别准确率。此外,小波能量、小波峰值的平均大小和小波峰值个数均为频域上的特征参数,本发明实施例通过频域上的一个特征参数或至少两个特征参数的组合实现对人体行为的识别。Among them, the average size of the peak of the wavelet and the number of peaks of the wavelet are characterized by small peaks. The wavelet energy distribution feature can reflect the energy distribution of the two behaviors of the upper and lower buildings on different wavelet decomposition layers. The wavelet peak characteristics can reflect the amplitude of the acceleration signal generated by the human body under these two behaviors. The combination of these two features can significantly improve the recognition accuracy of behaviors compared with the related techniques in which human behavior is recognized only by time domain characteristics such as mean square error, variance, and the like. In addition, the wavelet energy, the average size of the wavelet peaks, and the number of wavelet peaks are characteristic parameters in the frequency domain. In the embodiment of the present invention, the recognition of the human behavior is realized by a characteristic parameter in the frequency domain or a combination of at least two characteristic parameters. .
步骤13:依据所述小波特征值,确定所述待识别行为。Step 13: Determine the behavior to be identified according to the wavelet feature value.
本实施例中,利用数据挖掘中的决策树、支持向量机(SVM,Support Vector Machine)或极速学习机等算法,预先训练好一个分类器模型,将所述至少一个预定层中每层的小波能量、小波峰值的平均大小、小波峰值个数作为该分类器模型的输入信号输入至分类器模型中如将第2、3、4层中计算出的每层的小波能量、小波峰值的平均大小、小波峰值个数作为输入信号输入至分类器模型中,通过所述分类器模型对所述输入信号进行预设算法的运算识别出所述待识别行为;所述预设算法可以为决策树、SVM或极速学习机算法。即通过该分类器模型即可识别出待识别行为人体的哪种行为。In this embodiment, using a decision tree in a data mining, a support vector machine (SVM) or a speed learning machine, a classifier model is pre-trained, and wavelets of each layer in the at least one predetermined layer are used. The energy, the average size of the wavelet peaks, and the number of wavelet peaks are input to the classifier model as the input signal of the classifier model. The wavelet energy of each layer calculated in the second, third, and fourth layers, and the average size of the wavelet peaks. And the number of wavelet peaks is input to the classifier model as an input signal, and the operation of the preset signal is performed by the classifier model to identify the to-be-identified behavior; the preset algorithm may be a decision tree, SVM or speed learning machine algorithm. That is, through the classifier model, it is possible to identify which behavior of the human body to be identified.
其中,所述分类器模型的具体公式及其各项参数请参见现有相关说明,此处不赘述。以小波分解层数N=6、预定层为第2~4层为例,在分类器模型的训练过程中,分类器模型的输入包括带有行为标签的行为数据、该行为数据经N层小波分解后第2~4层每层的小波能量、小波峰值的平均大小以及小波峰值个数,利用这些输入信号,分类器模型的各项参数进行相应的调整,当分类器模型的各项参数调整至使分类器模型的输出结果与行为标签指示的行为相一致时,说明此时分类器模型训练好。具体的,训练一个分类器模型的过程包括:以预定的采集频率,通过传感器多次采集处于上楼行为中的行为数据以及多次采集处于下楼行为的行为数据,并将表征为同一个行为的行为数据记录在同一个文件中。举个例子,以采集频率为 50次/s为例,持续采集人体在上楼行为中的行为数据10s,在这10s内共采集到可表征为上楼行为的50*10=500个的行为数据,并将这500个行为数据保存在第一文件中,该第一文件记录的是人体在处于上楼行为时所采集到的上楼行为数据。持续采集人体在下楼行为中的行为数据5s,在这5s内共采集到可表征为下楼行为的50*5=250个的行为数据,并将这250个行为数据保存在第二文件中,该第二文件记录的是在人体在处于下楼行为时所采集到的下楼行为数据。需要说明的是,本领域技术人员应该而知,如果将上楼和下楼视为行为的不同种类,那么存储有不同行为数据的文件的数量应该与待识别行为的种类相同。由上可知,第一文件和第二文件中存储的行为数据是带有行为标签的(通过行为标签能够区分出某个行为数据对应的行为为哪种行为),即第一文件中存储的行为数据对应于上楼行为,第二文件中存储的行为数据对应于下楼行为。将这两个文件中存储的所有行为数据进行N=6层的小波分解,得到每个行为数据在第2~4层中每层的小波能量、小波峰值的平均大小以及小波峰值个数,将带有行为标签的部分或所有行为数据、该部分或所有行为数据经N=6层小波分解后第2~4层每层的小波特征值作为分类模型的训练数据集,并训练数据集代入至分类器模型以训练该模型的各项参数,当分类器模型的各项参数调整至使分类器模型的输出结果与行为标签指示的行为相一致时,说明此时分类器模型训练好。在分类器模型训练好之后,当通过加速度传感器采集到某个行为数据时,将按照前述步骤11~13进行,以通过训练好的分类器模型来识别该行为数据对应的行为。对分类器模型的各项参数的具体训练过程及训练数据集的概念请参见现有相关说明,此处不赘述。For the specific formula of the classifier model and its various parameters, please refer to the existing related description, which is not described here. Taking the wavelet decomposition layer number N=6 and the predetermined layer as the second to fourth layers as an example, in the training process of the classifier model, the input of the classifier model includes behavior data with behavior tags, and the behavior data passes through the N layer wavelet. After the decomposition, the wavelet energy of each layer of the second to fourth layers, the average size of the wavelet peaks, and the number of wavelet peaks, using these input signals, the parameters of the classifier model are adjusted accordingly, and the parameters of the classifier model are adjusted. When the output of the classifier model is consistent with the behavior indicated by the behavior tag, the classifier model is trained at this time. Specifically, the process of training a classifier model includes: collecting behavior data in an upstairs behavior by a sensor at a predetermined acquisition frequency, and collecting behavior data in a downstairs behavior multiple times, and characterizing the same behavior. The behavior data is recorded in the same file. For example, the acquisition frequency is For example, 50 times/s, the behavior data of the human body in the behavior of going upstairs is continuously collected for 10s, and 50*10=500 behavioral data that can be characterized as upstairs behavior are collected in these 10s, and these 500 behaviors are collected. The data is stored in a first file that records the behavior data of the upstairs collected by the human body when it is in the upstairs behavior. Continuously collect the behavior data of the human body in the behavior of going downstairs for 5s. In this 5s, collect 50*5=250 behavioral data that can be characterized as downstairs behavior, and save the 250 behavior data in the second file. The second document records the behavior data of the downstairs collected when the human body is in the downstairs behavior. It should be noted that those skilled in the art should know that if the upstairs and the downstairs are regarded as different types of behaviors, the number of files storing different behavior data should be the same as the type of behavior to be identified. It can be seen from the above that the behavior data stored in the first file and the second file is marked with a behavior label (the behavior label can distinguish the behavior corresponding to the behavior data), that is, the behavior stored in the first file. The data corresponds to the behavior of going upstairs, and the behavior data stored in the second file corresponds to the behavior of going downstairs. The N=6 layer wavelet decomposition is performed on all the behavior data stored in the two files, and the wavelet energy, the average size of the wavelet peaks, and the wavelet peak number of each layer of the behavior data in the second to fourth layers are obtained. Part or all of the behavior data with the behavior label, the part or all of the behavior data is decomposed by the N=6 layer wavelet, and the wavelet eigenvalues of each layer of the 2nd to 4th layers are used as the training data set of the classification model, and the training data set is substituted into The classifier model trains the parameters of the model. When the parameters of the classifier model are adjusted such that the output of the classifier model is consistent with the behavior indicated by the behavior tag, the classifier model is well trained. After the classifier model is trained, when a certain behavior data is collected by the acceleration sensor, it will be performed according to the aforementioned steps 11 to 13 to identify the behavior corresponding to the behavior data through the trained classifier model. For the specific training process of the parameters of the classifier model and the concept of the training data set, please refer to the existing related description, which is not described here.
前述方案中,是以识别上楼和下楼为例进行的说明,此外考虑到在实际应用中在楼梯与楼梯之间还存在有平地,本实施例还可以识别上楼、下楼和步行这三种细分动作行为,也可以识别慢走、快走和原地踏步这三种 细分动作行为。In the foregoing scheme, the description is made by taking the upper floor and the lower building as an example, and in addition, considering that there is a level between the stairs and the stairs in practical applications, the embodiment can also recognize the upstairs, the downstairs, and the walk. Three subdivided action behaviors can also identify three types: slow walking, fast walking, and in situ. Subdivide action behavior.
本发明实施例的技术方案的优势在于:The technical solutions of the embodiments of the present invention have the following advantages:
1)将采集到的行为数据经小波分解变换至频域上,从频域角度进行人体行为识别,与单纯的从时域角度进行人体行为识别相比,可提高识别准确性,减少误识别概率;1) The collected behavior data is transformed into the frequency domain by wavelet decomposition, and the human behavior recognition is performed from the frequency domain angle, which can improve the recognition accuracy and reduce the false recognition probability compared with the simple human body behavior recognition from the time domain angle. ;
2)通过将频域上的一个特征参数或至少两个特征参数的组合输入至预先训练好的分类器模型中,以通过预先训练好的分类器模型识别出人体行为;其中该分类器模型在预先的训练过程中采用的训练数据集为频域上至少一个特征参数,使训练好的分类器模型的准确度提高,由准确度较高的分类器模型来识别细分的人体行为,可提高识别准确度,减少误识别概率。2) by inputting a characteristic parameter in the frequency domain or a combination of at least two characteristic parameters into the pre-trained classifier model to identify the human behavior through the pre-trained classifier model; wherein the classifier model is The training data set used in the pre-training process is at least one characteristic parameter in the frequency domain, so that the accuracy of the trained classifier model is improved, and the highly accurate classifier model is used to identify the subdivided human behavior, which can be improved. Identify accuracy and reduce the probability of misrecognition.
基于前述的行为识别方法,本发明实施例还提供了一种行为识别设备,如图4所示,所述设备包括:第一获取单元401、第二获取单元402、第一确定单元403;其中,Based on the foregoing behavior recognition method, the embodiment of the present invention further provides a behavior recognition device. As shown in FIG. 4, the device includes: a first acquisition unit 401, a second acquisition unit 402, and a first determination unit 403; ,
第一获取单元401,配置为获取待识别行为对应的行为数据;The first obtaining unit 401 is configured to acquire behavior data corresponding to the behavior to be identified;
第二获取单元402,配置为获取所述行为数据的小波特征值;The second obtaining unit 402 is configured to acquire a wavelet feature value of the behavior data.
第一确定单元403,配置为依据所述小波特征值,确定所述待识别行为。The first determining unit 403 is configured to determine the to-be-identified behavior according to the wavelet feature value.
其中,所述第一获取单元401,还配置为:以预定的采样频率在预定的采集周期内采集人体在发生所述待识别行为时产生的加速度信号;确定所采集到的任意加速度信号为所述行为数据。The first obtaining unit 401 is further configured to: collect, in a predetermined sampling period, an acceleration signal generated by the human body when the to-be-identified behavior occurs in a predetermined sampling period; and determine that any acquired acceleration signal is Behavioral data.
其中,所述第二获取单元402,还配置为:将所述行为数据进行N层小波分解,得到每层的近似系数,所述近似系数包括高频近似系数和低频近似系数,N为小波分解层数、为大于等于1的正整数;保留至少一个预定层的低频近似系数;依据所保留的至少一个预定层的低频近似系数,确定所述至少一个预定层中每层的小波特征值。进一步的,所述第二获取单元402至少将所述行为数据、预定的小波分解层数N=6输入至预先设定的 小波变换函数中进行6层小波分解,得到每层的高频近似系数和低频近似系数;保留第2层、第3层及第4层的低频近似系数;利用第2层、第3层及第4层的低频近似系数,计算以下其中至少一种特征参数:所保留下的所有层中每层的小波能量值、小波峰值的平均大小、小波峰值个数;确定所计算出的每层的特征参数为相应层的所述小波特征值。The second obtaining unit 402 is further configured to perform N-layer wavelet decomposition on the behavior data to obtain an approximate coefficient of each layer, where the approximate coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition. The number of layers is a positive integer greater than or equal to 1; a low frequency approximation coefficient of at least one predetermined layer is retained; and wavelet feature values of each of the at least one predetermined layer are determined according to the retained low frequency approximation coefficients of the at least one predetermined layer. Further, the second obtaining unit 402 inputs at least the behavior data and the predetermined wavelet decomposition layer number N=6 to a preset one. The 6-layer wavelet decomposition is performed in the wavelet transform function to obtain the high-frequency approximation coefficient and the low-frequency approximation coefficient of each layer; the low-frequency approximation coefficients of the 2nd, 3rd, and 4th layers are retained; the second layer, the third layer, and the The low-frequency approximation coefficient of the 4 layers is calculated by calculating at least one of the following characteristic parameters: the wavelet energy value of each layer in all the layers retained, the average size of the wavelet peaks, and the number of wavelet peaks; determining the calculated characteristics of each layer The parameters are the wavelet eigenvalues of the respective layers.
其中,所述第一确定单元403,还配置为:确定所有层的特征参数为预先训练的分类器模型的输入信号;将所述输入信号输入至所述分类器模型;通过所述分类器模型对所述输入信号进行预设算法的运算识别出所述待识别行为。The first determining unit 403 is further configured to: determine that the feature parameters of all layers are input signals of the pre-trained classifier model; input the input signal to the classifier model; and pass the classifier model Performing an operation of a preset algorithm on the input signal identifies the behavior to be recognized.
为实现上述方法,本发明实施例还提供了一种行为识别设备,由于该设备解决问题的原理与方法相似,因此,行为识别设备的实施过程及实施原理均可以参见前述方法的实施过程及实施原理描述,重复之处不再赘述。In order to implement the above method, the embodiment of the present invention further provides a behavior recognition device. Since the principle and method for solving the problem are similar, the implementation process and implementation principle of the behavior recognition device can refer to the implementation process and implementation of the foregoing method. The principle description, the repetition will not be repeated.
在实际应用中,所述第一获取单元401、第二获取单元402、第一确定单元403均可由中央处理单元(CPU,Central Processing Unit)、或数字信号处理(DSP,Digital Signal Processor)、或微处理器(MPU,Micro Processor Unit)、或现场可编程门阵列(FPGA,Field Programmable Gate Array)等来实现。In a practical application, the first obtaining unit 401, the second obtaining unit 402, and the first determining unit 403 may each be a central processing unit (CPU), or a digital signal processing (DSP), or It is implemented by a microprocessor (MPU, Micro Processor Unit) or a Field Programmable Gate Array (FPGA).
本发明实施例还提供了一种计算机存储介质,所述计算机存储介质中存储有计算机可执行指令,所述计算机可执行指令用于执行前述的行为识别方法。The embodiment of the invention further provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are used to execute the foregoing behavior recognition method.
本领域技术人员应当理解,图4中所示的行为识别设备中的各处理模块的实现功能可参照前述行为识别方法的相关描述而理解。本领域技术人员应当理解,图4所示的行为识别设备中各处理单元的功能可通过运行于处理器上的程序而实现,也可通过具体的逻辑电路而实现。Those skilled in the art should understand that the implementation functions of the processing modules in the behavior recognition device shown in FIG. 4 can be understood by referring to the related description of the foregoing behavior recognition method. It should be understood by those skilled in the art that the functions of the processing units in the behavior recognition device shown in FIG. 4 can be implemented by a program running on a processor, or can be implemented by a specific logic circuit.
本领域内的技术人员应明白,本发明的实施例可提供为方法、系统、 或计算机程序产品。因此,本发明可采用硬件实施例、软件实施例、或结合软件和硬件方面的实施例的形式。而且,本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器和光学存储器等)上实施的计算机程序产品的形式。Those skilled in the art will appreciate that embodiments of the present invention can be provided as a method, system, Or a computer program product. Accordingly, the present invention can take the form of a hardware embodiment, a software embodiment, or a combination of software and hardware. Moreover, the invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) including computer usable program code.
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器,使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。The present invention has been described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (system), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or FIG. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine for the execution of instructions for execution by a processor of a computer or other programmable data processing device. Means for implementing the functions specified in one or more of the flow or in a block or blocks of the flow chart.
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中,使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品,该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。The computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device. The apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上,使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理,从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device. The instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.
以上所述,仅为本发明的较佳实施例而已,并非用于限定本发明的保护范围。The above is only the preferred embodiment of the present invention and is not intended to limit the scope of the present invention.
工业实用性Industrial applicability
本实施例中,先获取待识别行为对应的行为数据,再获取该行为数据 的小波特征值,并依据所述小波特征值确定所述待识别行为。其中,小波特征值为频域上的特征参数,从频域角度进行人体细分动作行为的识别,可提高识别准确性,减少误识别概率。 In this embodiment, the behavior data corresponding to the behavior to be identified is acquired first, and then the behavior data is acquired. a wavelet feature value, and determining the to-be-identified behavior according to the wavelet feature value. Among them, the wavelet feature value is a characteristic parameter in the frequency domain, and the recognition of the human body segmentation action behavior from the frequency domain angle can improve the recognition accuracy and reduce the false recognition probability.

Claims (11)

  1. 一种行为识别方法,所述方法包括:A behavior recognition method, the method comprising:
    获取待识别行为对应的行为数据;Obtaining behavior data corresponding to the behavior to be identified;
    获取所述行为数据的小波特征值;Obtaining a wavelet feature value of the behavior data;
    依据所述小波特征值,确定所述待识别行为。Determining the behavior to be identified according to the wavelet feature value.
  2. 根据权利要求1所述的方法,其中,所述获取待识别行为对应的行为数据,包括:The method according to claim 1, wherein the obtaining the behavior data corresponding to the behavior to be identified comprises:
    以预定的采样频率在预定的采集周期内采集人体在发生所述待识别行为时产生的加速度信号;Acquiring an acceleration signal generated by the human body when the behavior to be recognized occurs in a predetermined sampling period at a predetermined sampling frequency;
    确定所采集到的加速度信号为所述行为数据。The acquired acceleration signal is determined to be the behavior data.
  3. 根据权利要求1或2所述的方法,其中,所述获取所述行为数据的小波特征值,包括:The method according to claim 1 or 2, wherein the acquiring the wavelet feature value of the behavior data comprises:
    将所述行为数据进行N层小波分解,得到每层的近似系数,所述近似系数包括高频近似系数和低频近似系数,N为小波分解层数、为大于等于1的正整数;Performing N-layer wavelet decomposition on the behavior data to obtain an approximation coefficient of each layer, wherein the approximation coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition layer number, and is a positive integer greater than or equal to 1;
    保留至少一个预定层的低频近似系数;Retaining a low frequency approximation coefficient of at least one predetermined layer;
    依据所保留的至少一个预定层的低频近似系数,确定所述至少一个预定层中每层的小波特征值。A wavelet feature value of each of the at least one predetermined layer is determined according to the low frequency approximation coefficient of the at least one predetermined layer retained.
  4. 根据权利要求3所述的方法,其中,所述方法包括:The method of claim 3 wherein said method comprises:
    至少将所述行为数据、预定的小波分解层数N=6输入至预先设定的小波变换函数中进行6层小波分解,得到每层的高频近似系数和低频近似系数;At least the behavior data, the predetermined wavelet decomposition layer number N=6 is input into a preset wavelet transform function to perform 6-layer wavelet decomposition, to obtain a high-frequency approximation coefficient and a low-frequency approximation coefficient of each layer;
    保留第2层、第3层及第4层的低频近似系数;Retain the low frequency approximation coefficients of Layer 2, Layer 3 and Layer 4;
    利用第2层、第3层及第4层的低频近似系数,计算以下其中至少一 种特征参数:所保留下的所有层中每层的小波能量值、小波峰值的平均大小、小波峰值个数;Calculate at least one of the following using the low frequency approximation coefficients of the second, third, and fourth layers Characteristic parameters: the wavelet energy value of each layer in all layers retained, the average size of wavelet peaks, and the number of wavelet peaks;
    确定所计算出的每层的特征参数为相应层的所述小波特征值。It is determined that the calculated characteristic parameter of each layer is the wavelet feature value of the corresponding layer.
  5. 根据权利要求4所述的方法,其中,所述依据所述小波特征值,确定所述待识别行为,包括:The method according to claim 4, wherein the determining the behavior to be identified according to the wavelet feature value comprises:
    确定所有层的特征参数为预先训练的分类器模型的输入信号;Determining the characteristic parameters of all layers as input signals of the pre-trained classifier model;
    将所述输入信号输入至所述分类器模型;Inputting the input signal to the classifier model;
    通过所述分类器模型对所述输入信号进行预设算法的运算识别出所述待识别行为。Performing an operation of a preset algorithm on the input signal by the classifier model identifies the behavior to be recognized.
  6. 一种行为识别设备,所述设备包括:A behavior recognition device, the device comprising:
    第一获取单元,用于获取待识别行为对应的行为数据;a first acquiring unit, configured to acquire behavior data corresponding to the behavior to be identified;
    第二获取单元,用于获取所述行为数据的小波特征值;a second acquiring unit, configured to acquire a wavelet feature value of the behavior data;
    第一确定单元,用于依据所述小波特征值,确定所述待识别行为。The first determining unit is configured to determine the to-be-identified behavior according to the wavelet feature value.
  7. 根据权利要求6所述的设备,其中,所述第一获取单元,还配置为:The device according to claim 6, wherein the first obtaining unit is further configured to:
    以预定的采样频率在预定的采集周期内采集人体在发生所述待识别行为时产生的加速度信号;Acquiring an acceleration signal generated by the human body when the behavior to be recognized occurs in a predetermined sampling period at a predetermined sampling frequency;
    确定所采集到的加速度信号为所述行为数据。The acquired acceleration signal is determined to be the behavior data.
  8. 根据权利要求6或7所述的设备,其中,所述第二获取单元,还配置为:The device according to claim 6 or 7, wherein the second obtaining unit is further configured to:
    将所述行为数据进行N层小波分解,得到每层的近似系数,所述近似系数包括高频近似系数和低频近似系数,N为小波分解层数、为大于等于1的正整数;Performing N-layer wavelet decomposition on the behavior data to obtain an approximation coefficient of each layer, wherein the approximation coefficient includes a high frequency approximation coefficient and a low frequency approximation coefficient, and N is a wavelet decomposition layer number, and is a positive integer greater than or equal to 1;
    保留至少一个预定层的低频近似系数;Retaining a low frequency approximation coefficient of at least one predetermined layer;
    依据所保留的至少一个预定层的低频近似系数,确定所述至少一个预定层中每层的小波特征值。 A wavelet feature value of each of the at least one predetermined layer is determined according to the low frequency approximation coefficient of the at least one predetermined layer retained.
  9. 根据权利要求8所述的设备,其中,所述第二获取单元,还配置为:The device according to claim 8, wherein the second obtaining unit is further configured to:
    至少将所述行为数据、预定的小波分解层数N=6输入至预先设定的小波变换函数中进行6层小波分解,得到每层的高频近似系数和低频近似系数;At least the behavior data, the predetermined wavelet decomposition layer number N=6 is input into a preset wavelet transform function to perform 6-layer wavelet decomposition, to obtain a high-frequency approximation coefficient and a low-frequency approximation coefficient of each layer;
    保留第2层、第3层及第4层的低频近似系数;Retain the low frequency approximation coefficients of Layer 2, Layer 3 and Layer 4;
    利用第2层、第3层及第4层的低频近似系数,计算以下其中至少一种特征参数:所保留下的所有层中每层的小波能量值、小波峰值的平均大小、小波峰值个数;Using the low frequency approximation coefficients of the second layer, the third layer, and the fourth layer, at least one of the following characteristic parameters is calculated: the wavelet energy value of each layer in all the layers retained, the average size of the wavelet peak, and the number of wavelet peaks ;
    确定所计算出的每层的特征参数为相应层的所述小波特征值。It is determined that the calculated characteristic parameter of each layer is the wavelet feature value of the corresponding layer.
  10. 根据权利要求9所述的设备,其中,所述第一确定单元,还配置为:The device according to claim 9, wherein the first determining unit is further configured to:
    确定所有层的特征参数为预先训练的分类器模型的输入信号;Determining the characteristic parameters of all layers as input signals of the pre-trained classifier model;
    将所述输入信号输入至所述分类器模型;Inputting the input signal to the classifier model;
    通过所述分类器模型对所述输入信号进行预设算法的运算识别出所述待识别行为。Performing an operation of a preset algorithm on the input signal by the classifier model identifies the behavior to be recognized.
  11. 一种计算机存储介质,所述计算机存储介质中存储有计算机可执行指令,所述计算机可执行指令用于执行权利要求1至5任一项所述的方法。 A computer storage medium having stored therein computer executable instructions for performing the method of any one of claims 1 to 5.
PCT/CN2015/089058 2015-07-08 2015-09-07 Method, device for behavior recognition and computer storage medium WO2017004880A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510400436.4 2015-07-08
CN201510400436.4A CN106339071A (en) 2015-07-08 2015-07-08 Method and device for identifying behaviors

Publications (1)

Publication Number Publication Date
WO2017004880A1 true WO2017004880A1 (en) 2017-01-12

Family

ID=57684611

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/089058 WO2017004880A1 (en) 2015-07-08 2015-09-07 Method, device for behavior recognition and computer storage medium

Country Status (2)

Country Link
CN (1) CN106339071A (en)
WO (1) WO2017004880A1 (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112383488A (en) * 2020-10-26 2021-02-19 中山大学 Content identification method suitable for encrypted and non-encrypted data streams
CN113887812A (en) * 2021-10-14 2022-01-04 广东电网有限责任公司 Clustering-based small sample load prediction method, device, equipment and storage medium
CN114081504A (en) * 2021-11-23 2022-02-25 青岛理工大学 Driving intention identification method and system based on electroencephalogram signals

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018232557A1 (en) * 2017-06-19 2018-12-27 深圳市酷浪云计算有限公司 Exercise movement recognition method and apparatus, and electronic device
CN107862276B (en) * 2017-11-01 2021-05-18 北京旋极信息技术股份有限公司 Behavior identification method and terminal
CN109241926A (en) * 2018-09-19 2019-01-18 重庆信络威科技有限公司 A kind of driving behavior recognition methods based on wavelet analysis
CN110017834B (en) * 2019-04-15 2021-12-24 歌尔科技有限公司 Usage object determination method, usage object determination apparatus, and storage medium
CN113030890B (en) * 2019-12-24 2023-11-21 大富科技(安徽)股份有限公司 Target identification method and device based on vehicle-mounted radar

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090161915A1 (en) * 2007-12-24 2009-06-25 National Chiao Tung University Of Taiwan Multi-person pose recognition system using a zigbee wireless sensor network
CN103699795A (en) * 2013-12-20 2014-04-02 东软熙康健康科技有限公司 Exercise behavior identification method and device and exercise intensity monitoring system
CN103886323A (en) * 2013-09-24 2014-06-25 清华大学 Behavior identification method based on mobile terminal and mobile terminal
US20140278208A1 (en) * 2013-03-15 2014-09-18 Aliphcom Feature extraction and classification to determine one or more activities from sensed motion signals
CN104598880A (en) * 2015-03-06 2015-05-06 中山大学 Behavior identification method based on fuzzy support vector machine

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090161915A1 (en) * 2007-12-24 2009-06-25 National Chiao Tung University Of Taiwan Multi-person pose recognition system using a zigbee wireless sensor network
US20140278208A1 (en) * 2013-03-15 2014-09-18 Aliphcom Feature extraction and classification to determine one or more activities from sensed motion signals
CN103886323A (en) * 2013-09-24 2014-06-25 清华大学 Behavior identification method based on mobile terminal and mobile terminal
CN103699795A (en) * 2013-12-20 2014-04-02 东软熙康健康科技有限公司 Exercise behavior identification method and device and exercise intensity monitoring system
CN104598880A (en) * 2015-03-06 2015-05-06 中山大学 Behavior identification method based on fuzzy support vector machine

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112383488A (en) * 2020-10-26 2021-02-19 中山大学 Content identification method suitable for encrypted and non-encrypted data streams
CN113887812A (en) * 2021-10-14 2022-01-04 广东电网有限责任公司 Clustering-based small sample load prediction method, device, equipment and storage medium
CN113887812B (en) * 2021-10-14 2023-07-07 广东电网有限责任公司 Clustering-based small sample load prediction method, device, equipment and storage medium
CN114081504A (en) * 2021-11-23 2022-02-25 青岛理工大学 Driving intention identification method and system based on electroencephalogram signals
CN114081504B (en) * 2021-11-23 2024-03-01 青岛理工大学 Driving intention recognition method and system based on electroencephalogram signals

Also Published As

Publication number Publication date
CN106339071A (en) 2017-01-18

Similar Documents

Publication Publication Date Title
WO2017004880A1 (en) Method, device for behavior recognition and computer storage medium
Xu et al. Human activity recognition based on convolutional neural network
Uddin et al. Human activity recognition from wearable sensors using extremely randomized trees
KR102619981B1 (en) Gesture classification apparatus and method using electromyogram signals
Sousa et al. A comparative analysis of the impact of features on human activity recognition with smartphone sensors
Christie et al. Acoustics based terrain classification for legged robots
Uddin et al. A guided random forest based feature selection approach for activity recognition
Rasekh et al. Human activity recognition using smartphone
Zabidi et al. Detection of asphyxia in infants using deep learning convolutional neural network (CNN) trained on Mel frequency cepstrum coefficient (MFCC) features extracted from cry sounds
Zhang et al. Human activity recognition based on time series analysis using U-Net
Janko et al. A new frontier for activity recognition: The Sussex-Huawei locomotion challenge
Giorgi et al. Try walking in my shoes, if you can: Accurate gait recognition through deep learning
CN110647788A (en) Human daily behavior classification method based on micro-Doppler characteristics
KR20220113242A (en) On-device activity recognition
Mace et al. Accelerometer-based hand gesture recognition using feature weighted naïve bayesian classifiers and dynamic time warping
CN112037929A (en) Classification method based on multi-modal machine learning, online new coronary pneumonia early warning model training method and early warning method
Sideridis et al. Gesturekeeper: Gesture recognition for controlling devices in iot environments
CN113642432A (en) Method for identifying human body posture by convolutional neural network based on covariance matrix transformation
US10782791B2 (en) Time domain feature transform for user gestures
Alam et al. Leveraging Sensor Fusion and Sensor-Body Position for Activity Recognition for Wearable Mobile Technologies.
Liu et al. A lightweight double-channel depthwise separable convolutional neural network for multimodal fusion gait recognition
She et al. RadarSpecAugment: A simple data augmentation method for radar-based human activity recognition
Gao et al. MMTSA: Multi-Modal Temporal Segment Attention Network for Efficient Human Activity Recognition
Liang et al. Fall detection using lifting wavelet transform and support vector machine
Ison et al. Beyond user-specificity for emg decoding using multiresolution muscle synergy analysis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 15897536

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 15897536

Country of ref document: EP

Kind code of ref document: A1