CN106778652A - Physical activity recognition methods and device - Google Patents

Physical activity recognition methods and device Download PDF

Info

Publication number
CN106778652A
CN106778652A CN201611221454.7A CN201611221454A CN106778652A CN 106778652 A CN106778652 A CN 106778652A CN 201611221454 A CN201611221454 A CN 201611221454A CN 106778652 A CN106778652 A CN 106778652A
Authority
CN
China
Prior art keywords
data
identified
user
identification
terminal
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
CN201611221454.7A
Other languages
Chinese (zh)
Inventor
徐丽丽
王宇飞
董俊龙
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.)
Neusoft Corp
Original Assignee
Neusoft Corp
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 Neusoft Corp filed Critical Neusoft Corp
Priority to CN201611221454.7A priority Critical patent/CN106778652A/en
Publication of CN106778652A publication Critical patent/CN106778652A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/20—Movements or behaviour, e.g. gesture recognition
    • G06V40/23—Recognition of whole body movements, e.g. for sport training
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Human Computer Interaction (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a kind of physical activity recognition methods and device, it is related to network technique field, activity recognition target can be divided level, the complexity that data characteristics is extracted when can reduce multi-activity target identification.Methods described includes：Treating identifying user when the identification for receiving User Activity state is instructed, the need for being carried in the acquisition identification instruction carries out the identification target of activity recognition, and the terminal data for obtaining the customer mobile terminal to be identified；It is determined that disaggregated model combination process corresponding with the identification target；According to the disaggregated model combination process and the data characteristics of the terminal data that determine, active state to the user to be identified carries out successively multiclass classification, obtain the active state of the user to be identified, wherein, there is hyponymy between the last layer level disaggregated model active state that obtains of classification and next optional active state of hierarchical classification category of model.The present invention is suitable for physical activity identification.

Description

Physical activity recognition methods and device
Technical field
The present invention relates to a kind of network technique field, more particularly to a kind of physical activity recognition methods and device.
Background technology
With the development of network technology, physical activity identification is more and more important.Physical activity identification has widely grinds very much Study carefully meaning and value, the society of the health of the mankind, training state, the prediction mankind can be detected by the identification to physical activity Meeting behavior etc..
At present, can realize that physical activity is recognized by multi classifier, specifically according to the User Activity shape for collecting The data characteristics of state data, different classes of active state is divided into by multi classifier by human body active state.
However, multi classifier is when multi-activity target identification is carried out, data characteristics extraction process complexity is higher, especially Under the complex situations such as many, the data characteristics intersection of data characteristics, data characteristics is extracted can have deviation, and then None- identified is obtained The accurate active state of user, so as to the accuracy that physical activity is recognized can be influenceed.
The content of the invention
In view of this, the invention provides a kind of physical activity recognition methods and device, main purpose is to reduce The complexity of Feature Selection during multi-activity target identification, it is to avoid data characteristics extracts physical activity recognition accuracy caused by deviation It is relatively low.
According to one aspect of the invention, there is provided a kind of physical activity recognition methods, the method includes：
When the identification for receiving User Activity state is instructed, obtain during the identification is instructed the need for carrying to be identified User carries out the identification target of activity recognition, and the terminal data for obtaining the customer mobile terminal to be identified；
It is determined that disaggregated model combination process corresponding with the identification target, different identification targets corresponds to different respectively Disaggregated model combination process, comprising one or more the classification moulds arranged according to predefined procedure in the disaggregated model combination process Type；
According to the disaggregated model combination process and the data characteristics of the terminal data that determine, to the use to be identified The active state at family carries out successively multiclass classification, obtains the active state of the user to be identified, wherein, last layer level classification mould There is hyponymy between the type active state that obtains of classification and next optional active state of hierarchical classification category of model.
Further, methods described also includes：
The terminal data of the different user mobile terminal that acquisition is collected into；
According to different active states, corresponding characteristic is trained in extracting the terminal data, obtains activity The corresponding disaggregated model of state；
Determination disaggregated model combination process corresponding with the identification target, specifically includes：
The identification target is parsed, multiple optional active states corresponding with the identification target are obtained；
According to the hyponymy between the multiple optional active state, corresponding disaggregated model is selected to carry out group Close, obtain disaggregated model combination process corresponding with the identification target.
Specifically, it is described to divide according to determining if the identification target is the basic activity identification of the user The data characteristics of class model combination process and the terminal data, the active state to the user to be identified carries out successively multistage Classification, specifically includes：
The location data of terminals feature of the customer mobile terminal to be identified is extracted from the terminal data；
According to the location data of terminals feature, by the decision tree classifier of basic activity identification to the use to be identified The active state at family is classified；
If determining the basic activity state of the user to be identified for ambulatory activities state according to classification results, from described The accelerometer data feature of customer mobile terminal to be identified is extracted in terminal data；
According to the accelerometer data feature, the support vector machine classifier recognized by walking classification is lived to the walking Dynamic state is classified.
Specifically, if determining that the basic activity state of the user to be identified is the work that rides public transportation means according to classification results Dynamic state and the vehicles identification for recognizing that target is user's seating to be identified, then it is described according to determining The data characteristics of disaggregated model combination process and the terminal data, is carried out successively many to the active state of the user to be identified Level classification, specifically also includes：
The location data of terminals feature and terminal signaling number of customer mobile terminal to be identified are extracted from the terminal data According to feature, the terminal signaling is terminal called signal and/or terminal network signal；
According to the location data of terminals feature and the terminal signaling data characteristics, the decision-making recognized by the vehicles Tree Classifier is classified to the active state that rides public transportation means；
If determining that the vehicles that the user to be identified takes are automobile according to classification results, from the automobile pre- Vehicle operation data feature is extracted in vehicle operation data in the fixed cycle time period；
According to the vehicle operation data feature, by the decision tree classifier of vehicle type recognition classify obtains institute State the type of vehicle of automobile.
Specifically, the vehicle operation data feature is included and enlivens driving range, vehicle line species, each scheduled time The accumulative operation duration of vehicle in interval, the determination mode for enlivening driving range includes：
By default DBSCAN Density Clustering functions, to the vehicle location of the automobile within the predetermined period time period Data carry out Density Clustering, obtain each class cluster；
By the central point of the most class cluster of sample point in described each class cluster, it is defined as the active centre point of the automobile, And by arrived a little in the most class cluster of the sample point active centre point apart from maximum, be defined as the automobile Traveling enlivens radius；
Radius is enlivened according to the active centre point and the traveling, determine the automobile enlivens driving range.
Further, if determining the vehicles that the user to be identified takes for automobile and described according to classification results Identification target be the user to be identified passenger identity identification, then it is described according to determine the disaggregated model combination process and The data characteristics of the terminal data, the active state to the user to be identified carries out successively multiclass classification, specifically also includes：
The accelerometer data feature and gyro data that customer mobile terminal to be identified is extracted from the terminal data are special Levy；
According to the accelerometer data feature and the gyro data feature, the K recognized by passenger identity is closest Grader classify and obtains the corresponding passenger identity of the user to be identified.
According to another aspect of the invention, there is provided a kind of physical activity identifying device, the device includes：
Acquiring unit, for when the identification for receiving User Activity state is instructed, obtaining carrying in the identification instruction The need for treat identifying user and carry out the identification target of activity recognition, and obtain the terminal of the customer mobile terminal to be identified Data；
Determining unit, for determining disaggregated model combination process corresponding with the identification target of acquiring unit acquisition, Different identification targets corresponds to different disaggregated model combination process respectively, is included according to pre- in the disaggregated model combination process Fixed tactic one or more disaggregated models；
Taxon, obtains for the disaggregated model combination process determined according to the determining unit and the acquiring unit Terminal data data characteristics, the active state to the user to be identified carries out successively multiclass classification, obtains described waiting to know The active state of other user, wherein, active state and next hierarchical classification model point that last layer level disaggregated model classification is obtained There is hyponymy between the optional active state of class.
Further, described device also includes：Training unit；
The acquiring unit, the terminal data for obtaining the different user mobile terminal being collected into；
The training unit, for according to different active states, extracting corresponding characteristic in the terminal data It is trained, obtains the corresponding disaggregated model of active state；
The determining unit, specifically for being parsed to the identification target, obtains corresponding with the identification target Multiple optional active states；
According to the hyponymy between the multiple optional active state, corresponding disaggregated model is selected to carry out group Close, obtain disaggregated model combination process corresponding with the identification target.
Specifically, the taxon, if specifically for the basic activity that the identification target is the user to be identified Identification, then extract the location data of terminals feature of the customer mobile terminal to be identified from the terminal data；
According to the location data of terminals feature, by the decision tree classifier of basic activity identification to the use to be identified The active state at family is classified；
If determining the basic activity state of the user to be identified for ambulatory activities state according to classification results, from described The accelerometer data feature of customer mobile terminal to be identified is extracted in terminal data；
According to the accelerometer data feature, the support vector machine classifier recognized by walking classification is lived to the walking Dynamic state is classified.
The taxon, if being specifically additionally operable to determine according to classification results the basic activity state of the user to be identified Be to ride public transportation means active state and the identification target is vehicles identification that the user to be identified takes, then from The location data of terminals feature and terminal signaling data characteristics of customer mobile terminal to be identified are extracted in the terminal data, it is described Terminal signaling is terminal called signal and/or terminal network signal；
According to the location data of terminals feature and the terminal signaling data characteristics, the decision-making recognized by the vehicles Tree Classifier is classified to the active state that rides public transportation means；
If determining that the vehicles that the user to be identified takes are automobile according to classification results, from the automobile pre- Vehicle operation data feature is extracted in vehicle operation data in the fixed cycle time period；
According to the vehicle operation data feature, by the decision tree classifier of vehicle type recognition classify obtains institute State the type of vehicle of automobile.
Specifically, the vehicle operation data feature is included and enlivens driving range, vehicle line species, each scheduled time The accumulative operation duration of vehicle in interval, the determining unit is additionally operable to by default DBSCAN Density Clustering functions, in institute The vehicle position data for stating the automobile in the predetermined period time period carries out Density Clustering, obtains each class cluster；
By the central point of the most class cluster of sample point in described each class cluster, it is defined as the active centre point of the automobile, And by arrived a little in the most class cluster of the sample point active centre point apart from maximum, be defined as the automobile Traveling enlivens radius；
Radius is enlivened according to the active centre point and the traveling, determine the automobile enlivens driving range.
Further, the taxon, if being specifically additionally operable to determine that the user to be identified takes according to classification results The vehicles for automobile and it is described identification target be the user to be identified passenger identity identification, then from the number of terminals According to the middle accelerometer data feature and gyro data feature for extracting customer mobile terminal to be identified；
According to the accelerometer data feature and the gyro data feature, the K recognized by passenger identity is closest Grader classify and obtains the corresponding passenger identity of the user to be identified.
A kind of physical activity recognition methods provided by above-mentioned technical proposal, the present invention and device, with current by many Class grader realizes that physical activity is known and compares otherwise that the present invention with needing to treat identifying user it is determined that carry out activity recognition The corresponding disaggregated model combination process of identification target after, according to the disaggregated model combination process and customer mobile terminal to be identified Terminal data data characteristics, the active state for treating identifying user carries out successively multiclass classification, can be by classification task point It is multiclass classification task to solve, and the last layer level disaggregated model active state that obtains of classification and next hierarchical classification category of model There is hyponymy between optional active state, and then some similar human body active states can be first categorized as a system Claim active state, then the active state for finding user's reality is finely divided to the general designation active state, whole process is reduced often The data characteristics number and moving target number of one hierarchical classification, it is to avoid data characteristics it is many, the class of activity is more, data characteristics intersect etc. Complex situations carry out Direct Classification, the complexity that data characteristics is chosen when can reduce multi-activity target identification, it is to avoid data are special Physical activity recognition accuracy is low caused by levying extraction deviation, so as to improve the accuracy of physical activity identification.
Described above is only the general introduction of technical solution of the present invention, in order to better understand technological means of the invention, And can be practiced according to the content of specification, and in order to allow the above and other objects of the present invention, feature and advantage can Become apparent, below especially exemplified by specific embodiment of the invention.
Brief description of the drawings
By reading the detailed description of hereafter preferred embodiment, various other advantages and benefit is common for this area Technical staff will be clear understanding.Accompanying drawing is only used for showing the purpose of preferred embodiment, and is not considered as to the present invention Limitation.And in whole accompanying drawing, identical part is denoted by the same reference numerals.In the accompanying drawings：
Fig. 1 shows a kind of physical activity recognition methods schematic flow sheet provided in an embodiment of the present invention；
Fig. 2 shows a kind of activity state classification example schematic provided in an embodiment of the present invention；
Fig. 3 shows another physical activity recognition methods schematic flow sheet provided in an embodiment of the present invention；
Fig. 4 shows a kind of example schematic of all activity recognitions provided in an embodiment of the present invention；
Fig. 5 shows five disaggregated model example schematics provided in an embodiment of the present invention；
Fig. 6 shows a kind of method flow schematic diagram for determining to enliven driving range provided in an embodiment of the present invention；
Fig. 7 shows a kind of method flow schematic diagram for determining type of vehicle provided in an embodiment of the present invention；
Fig. 8 shows a kind of physical activity identifying device structural representation provided in an embodiment of the present invention；
Fig. 9 shows another physical activity identifying device structural representation provided in an embodiment of the present invention.
Specific embodiment
The exemplary embodiment of the disclosure is more fully described below with reference to accompanying drawings.Although showing the disclosure in accompanying drawing Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here Limited.Conversely, there is provided these embodiments are able to be best understood from the disclosure, and can be by the scope of the present disclosure Complete conveys to those skilled in the art.
A kind of physical activity recognition methods is the embodiment of the invention provides, feature choosing when can reduce multi-activity target identification The complexity for taking, can improve the accuracy of physical activity identification, as shown in figure 1, the method includes：
101st, when the identification for receiving User Activity state is instructed, obtain during identification is instructed the need for carrying to be identified User carries out the identification target of activity recognition, and the terminal data for obtaining customer mobile terminal to be identified.
Wherein, terminal data can be comprising global positioning system (Global Positioning in mobile terminal System, GPS) locator collection data and the data etc. of each sensor collection.And it can be basic people to recognize target The identification target of body activity, the target can identify that the active state of user to be identified is specifically on foot, runs, takes traffic Instrument etc. activity in which；Identification target can also be the identification target of the vehicles, and the target can identify user Which in aircraft, high ferro, train, automobile, subway and other modes of transport the vehicles of seating be specifically.Need explanation It is in embodiments of the present invention, target optionally to be recognized for user to be identified, can be carried out according to different actual demands pre- First set, to meet different user's requests.
In embodiments of the present invention, human body active state can be known by gathering the terminal data of mobile terminal Not, wherein, mobile terminal can be the equipment such as smart mobile phone, panel computer, intelligent watch, Intelligent bracelet.Current mobile terminal In software and hardware it is powerful, integrated kind of sensor is various in mobile terminal, including gyro sensor, Distance-sensing Device, Light indicator, magnetometric sensor etc..And current user is higher to the degree of dependence of mobile terminal, and user can often take Tape movement terminal, therefore, it can by gather these sensor collections in mobile terminal to data be analyzed identification, recognize Go out the specific active state of user to be identified.And can be collected for basis for the executive agent of the embodiment of the present invention The terminal data of customer mobile terminal to be identified carries out the device of physical activity identification, and the device is receiving User Activity state Identification when instructing, identification instruction can be parsed, obtaining needing treating identifying user carries out the identification mesh of activity recognition Mark, and the terminal data of customer mobile terminal to be identified is obtained according to identification instruction.
102nd, disaggregated model combination process corresponding with identification target is determined.
Wherein, different identification targets corresponds to different disaggregated model combination process respectively.In disaggregated model combination process Comprising one or more disaggregated models arranged according to predefined procedure.In embodiments of the present invention, can be previously according to different Hyponymy between active state, by multiple similar active states merge with abstract treatment, and train corresponding Disaggregated model.For example, for active state and the running active state of walking, it is ambulatory activities state that can merge abstract, and For ambulatory activities state, being run and walked line state due to different people has differences, it is necessary to substantial amounts of sample data carries out model Training, therefore from support vector machine classifier as the corresponding disaggregated model of ambulatory activities state, and can collect corresponding Sample data the disaggregated model is trained.
After being finished to each disaggregated model training, the disaggregated model of different levels relation can be obtained.For example, such as Fig. 2 It is shown, User Activity state to be classified by disaggregated model 1, optional active state has activity 1, activity 2, abstract work Dynamic, further, when carrying out activity state classification again to this abstract activities, optional active state has activity 3 and activity 4, Here activity 3 and activity 4 may be considered two similar active states, and abstract activities can be merged by activity 3 and activity 4 The abstract general designation activity for obtaining, thus has upper and lower hierarchical relationship between disaggregated model 1 and disaggregated model 2, disaggregated model 1 is point The disaggregated model of the last layer level of class model 2, so that in each hierarchical classification, reduce the data characteristics for needing to extract, Data characteristics is avoided to extract deviation.
For the embodiment of the present invention, can be according to the hierarchical relationship between specific identification target and disaggregated model, selection Corresponding disaggregated model is combined sequence, obtains disaggregated model combination process corresponding with the identification target, so that basis should Disaggregated model combination process carries out physical activity identification.
103rd, according to the disaggregated model combination process and the data characteristics of terminal data for determining, the activity of identifying user is treated State carries out successively multiclass classification, obtains the active state of user to be identified.
Wherein, the active state that the classification of last layer level disaggregated model is obtained optionally is lived with next hierarchical classification category of model There is hyponymy between dynamic state.For example, classifying the active state for obtaining for walking is lived by last layer level disaggregated model Dynamic state, and next optional active state of hierarchical classification category of model is active state, running active state on foot, is walked here There is hyponymy between row active state and on foot active state, running active state.
For example, identification target is the identification target of basic physical activity, according to corresponding disaggregated model combination process, The active state for treating identifying user carries out successively multiclass classification：Gps data is extracted first from terminal data Data characteristics, the data characteristics includes：Translational speed maximum, according to this feature, is classified by disaggregated model 1, and then Determine that user to be identified is currently at ambulatory activities state, the data for then extracting accelerometer data from terminal data again are special Levy, the data characteristics includes：Acceleration standard deviation, the quantile sum of acceleration 25, the quantile sum of acceleration 75, according to the spy Levy, classified by disaggregated model 2, finally determine user to be identified currently specifically in running active state, it is logical with current Cross multi classifier directly to be compared according to the mode classified of data characteristics extracted, it may be determined that user to be identified is really Which basic human body active state.
A kind of physical activity recognition methods provided in an embodiment of the present invention, is carried out by the active state for treating identifying user Successively multiclass classification, can be decomposed into multiclass classification task by classification task, and last layer level disaggregated model classifies what is obtained There is hyponymy between active state and next optional active state of hierarchical classification category of model, and then can be by some Similar human body active state be first categorized as one general designation active state, then the general designation active state is finely divided find it is to be identified The actual active state of user, whole process reduces the data characteristics number and moving target number of each hierarchical classification, can drop The complexity that data characteristics is chosen during low multi-activity target identification, it is to avoid data characteristics extracts physical activity identification caused by deviation Accuracy rate is low, so as to improve the accuracy of physical activity identification.
In order to preferably understand the method shown in above-mentioned Fig. 1, as refinement and extension to above-mentioned implementation method, Another physical activity recognition methods is the embodiment of the invention provides, as shown in figure 3, the method includes：
201st, when the identification for receiving User Activity state is instructed, obtain during identification is instructed the need for carrying to be identified User carries out the identification target of activity recognition, and the terminal data for obtaining customer mobile terminal to be identified.
For the embodiment of the present invention, the identification target of User Activity state can be according to actual needs selected, that is, need to know Which degree be clipped to.For example, the active state that can be recognized and its partition of the level are as shown in figure 4, to cover people daily universal Active state (static, walking, multiply the vehicles), role (passenger or driver) and the conventional vehicles (aircraft, height Iron, subway, train, bus, motor bus, private car, taxi).Level Four classification can be specifically divided into, 5 disaggregated models are (1. 2. 3. 4. 5.), the data characteristics of each hierarchical classification reference has differences.As shown in figure 4, phase can be selected according to actual needs The identification target of the User Activity state answered, such as identification of basic physical activity, the identification of the vehicles, angle when riding in an automobile The identification target such as vehicle type recognition when color is recognized, ridden in an automobile.
In order to be identified finishing corresponding disaggregated model, it is necessary to train in advance in time to User Activity state, with Just accurate classification results are obtained, therefore, before step 201, also include：The different user mobile terminal that acquisition is collected into Terminal data；According to different active states, corresponding characteristic is trained in extracting terminal data, obtains active state Corresponding disaggregated model.
For example, the terminal data of the mobile terminal of different sample of users can be collected in advance, corresponding characteristic is extracted Train classification models, as shown in figure 4, optimum, degree of accuracy highest disaggregated model can be chosen per first-level class, to carry The degree of accuracy of mass activity identification high, disaggregated model data sources at different levels and data characteristics are as shown in figure 5, each disaggregated model is specific It is as follows：
For disaggregated model 1., the model data source is gps data, and the data characteristics of reference includes speed Maximum, it is possible to achieve inactive state, ambulatory status, the classification of the state that rides public transportation means.Because the aspect of model data are few, Feature has obvious boundary, can select decision tree classifier so that disaggregated model execution efficiency is higher, and extracts not same The characteristic of the terminal data of this user trains the disaggregated model, for example, extracting different sample of users is respectively at static shape The characteristic of speed of mobile terminal maximum when state, ambulatory status, the state that rides public transportation means, instructs according to this feature data Practice decision tree classifier, obtain that inactive state, ambulatory status, the disaggregated model of the state classification that rides public transportation means can be realized；
For disaggregated model 2., the model data source is accelerometer data, and the data characteristics of reference includes acceleration standard Difference, the quantile sum of acceleration 25, the quantile sum of acceleration 75, it is possible to achieve walk line state, the classification of running state.Due to The model belongs to two disaggregated models, and different people is run and walks line state and has differences, therefore the model needs substantial amounts of sample Notebook data carries out model training, so from support vector machine classifier, for example, extract different sample of users being respectively on foot Mobile terminal acceleration standard deviation, the quantile sum of acceleration 25, the quantile sum of acceleration 75 when state, running state Characteristic, according to this feature data Training Support Vector Machines grader, obtains that line state, running state classification can be realized away Disaggregated model；
For disaggregated model 3., the model data source is gps data, and the data characteristics of reference includes speed Average value, GPS satellite number, terminal signaling average strength, it is possible to achieve aircraft, high ferro, automobile, subway, fire The identification of the vehicles such as car, wherein, the terminal signaling can be comprising terminal called signal and network signal etc..Due to the model The training data of needs is few, from decision tree classifier so that model training and execution efficiency are higher, for example, can extract not It is speed of mobile terminal average value during with sample of users difference airplane, high ferro, automobile, subway, the vehicles such as train, complete Ball positioning system satellite number, the characteristic of terminal signaling average strength, decision tree classification is trained according to this feature data Device, obtains realizing the disaggregated model of the vehicles such as aircraft, high ferro, automobile, subway, train identification；
For disaggregated model 4., the model data source is accelerometer data, gyro data, and the data characteristics of reference includes Acceleration standard deviation, the quantile sum of acceleration 25, the quantile sum of acceleration 75, gyro data standard deviation, in driver In the driving procedure mobile terminal place in the car, do not carry with the premise of, 4. can realize driving by disaggregated model The person of sailing, the identification of passenger.Intersect because driver and passenger's feature are present, from K nearest neighbour classification devices, can not only realize shape State is classified, and can also provide the probable value for belonging to certain classification, and model caller can combine classification and probable value and carry out characteristic Design, for example, mobile terminal acceleration standard deviation when can extract different sample of users respectively as driver, passenger, plus The quantile sum of speed 25, the quantile sum of acceleration 75, the characteristic of gyro data standard deviation, according to this feature data Training K nearest neighbour classification devices, obtain that driver, the disaggregated model of passenger's identification can be realized；
For disaggregated model 5., the aspect of model is chosen and assorting process is different from other four disaggregated models, because multiplying The data almost indifference of visitor's terminal collection when private car, bus, motor bus, taxi is taken, therefore the model is based on Service data in various vehicle some cycles, the type to vehicle is classified, and vehicle operation data can be by being placed on Particular terminal is collected in vehicle, it is also possible to which the mobile terminal being placed on by user in vehicle is collected.According to the mobile end of user End data, if it is identical with the course of a certain vehicle to detect the course of user in the same time period, and gait of march Also it is consistent, i.e., passenger keeps geo-stationary with a certain vehicle, then illustrates that type of vehicle as passenger of the vehicle takes The type of vehicle of vehicle.The data source of the model is service data in vehicle some cycles, and the data characteristics of reference is included actively The accumulative operation duration of vehicle in driving range, vehicle line species, each predetermined time interval, realize private car, bus, The identification of the type of vehicle such as motor bus, taxi, because the training characteristics data that the model needs are few, from decision tree classification Device so that model training and execution efficiency are higher, for example, can extract different sample of users take respectively private car, bus, Service data feature when motor bus, taxi in the vehicle some cycles of mobile terminal collection, decision-making is trained according to this feature Tree Classifier, obtains that private car, bus, motor bus, the disaggregated model of taxi identification can be realized.Wherein, the scheduled time Interval can be set according to the actual requirements, such as each predetermined time interval can for each season, monthly, per two weeks Can add up driving time for vehicle Deng the accumulative operation duration of, vehicle；Vehicle line species includes drawing driving data by stroke Point, stroke is the driving task for once having beginning and end, and the circuit with identical beginning and end may be considered a kind of line Road；
And it can be the conventional operation area of vehicle to enliven driving range, it is necessary first to exclude the non-recurrent Operational Zone of vehicle Domain to enliven driving range calculating influence, specific calculation process as shown in fig. 6, including：By default DBSCAN Density Clusterings Function, the vehicle position data to the automobile within the predetermined period time period carries out Density Clustering, obtains each class cluster；By each class The central point of the most class cluster of sample point in cluster, is defined as the active centre point of automobile, and by institute in the most class cluster of sample point A little arrive active centre point apart from maximum, the traveling for being defined as automobile enlivens radius；Lived according to active centre point and traveling Jump radius, determine automobile enlivens driving range.Wherein, presetting DBSCAN Density Clusterings function can be poly- according to DBSCAN density Class algorithm carries out writing setting in advance.Specifically, the calculating of driving range is enlivened based on the position data in vehicle some cycles, Position data is clustered with DBSCAN density clustering algorithms first, exclude occasionally working line position to result of calculation Influence.Compared with K-means methods, DBSCAN density clustering algorithms need not in advance know the class number of clusters amount for needing to be formed, It can be found that the cluster class of arbitrary shape, can recognize that noise spot.The most class cluster of the sample point that is obtained after cluster is vehicle warp Normal operation area.The central point of class cluster is the active centre of vehicle, and the computational methods of active centre are that have a longitude and latitude The position of average, and radius of action be arrive a little in class cluster central point apart from maximum.
If it should be noted that obtain disaggregated model classification accuracy it is undesirable, can by increase data source, Characteristic type, the mode of Optimum Classification model lift the classification accuracy of disaggregated model.
202nd, disaggregated model combination process corresponding with identification target is determined.
Step 202 is specifically included：Identification target is parsed, multiple optional activities corresponding with identification target are obtained State；According to the hyponymy between multiple optional active states, select corresponding disaggregated model to be combined, obtain with The corresponding disaggregated model combination process of identification target.
For example, as shown in figure 4, when identification 1 (static, walking, the seating traffic work that identification target is basic physical activity Tool) when, disaggregated model combination process corresponding with the target be disaggregated model 1.；When the knowledge that identification target is basic physical activity When other 2 (static, walk, run, riding public transportation means), disaggregated model combination process corresponding with the target be disaggregated model 1. →②；When identification 1 (aircraft, high ferro, train, automobile, the subway) that target is the vehicles is recognized, corresponding with the target point Class model combination process be disaggregated model 1. → 3.；When identification target is identification 2 (aircraft, high ferro, train, the vapour of the vehicles Car, subway, taxi, private car, motor bus, bus) when, disaggregated model combination process corresponding with the target is classification mould Type 1. → 3. → 4. → 5.；When recognizing that target is recognized for passenger, disaggregated model combination process corresponding with the target is classification Model 1. → 3. → 4.；When it is all activity recognitions to recognize target, 1. → 2. → 3. → 4. → 5..It is thereby achieved that Identification target and disaggregated model combination process are as shown in table 1：
Table 1
For the embodiment of the present invention, it is possible to achieve the identification of multiple target physical activity, different user's requests can be met, and In order to illustrate the physical activity recognition methods of the embodiment of the present invention, as a example by recognizing the identification that target is all activities, hold successively Row following steps.
203rd, the location data of terminals feature of customer mobile terminal to be identified is extracted from terminal data.
Wherein, location data of terminals feature includes translational speed maximum.
In embodiments of the present invention, mobile terminal is carried with user, by mobile terminal terminal location not in the same time Data can determine the translational speed maximum of mobile terminal, and then determine the translational speed maximum of user.
204th, according to location data of terminals feature, the decision tree classifier recognized by basic activity treats identifying user Active state is classified.
For example, the gps data according to mobile terminal in certain period of time, calculates the translational speed maximum of user to be identified Value, then the translational speed maximum of user to be identified is brought into the decision tree classifier of basic activity identification, i.e. step 201 In disaggregated model 1. in, in determining that the translational speed maximum belongs to inactive state, ambulatory status, the state that rides public transportation means Which speed maximum range in, and then can realize inactive state, ambulatory status, the state that rides public transportation means point Class, the decision tree classifier recognized by basic activity is carried out classification and can improve execution efficiency, when detecting user to be identified Active state when being inactive state, other specific data features, use can be extracted from the mobile terminal data of the user In further detect the specific active state of the user be have a meal, sleep, which active state in the activity of taking exercises etc., It should be noted that now mobile terminal is the terminal device that user to be identified wears with oneself, intelligent watch, intelligence are specifically as follows Energy bracelet etc..
If 205a, the basic activity state of user to be identified is determined according to classification results for ambulatory activities state, from end The accelerometer data feature of customer mobile terminal to be identified is extracted in end data.
Wherein, accelerometer data feature includes acceleration standard deviation, the quantile sum of acceleration 25, the quantile of acceleration 75 Sum.
206a, according to accelerometer data feature, the support vector machine classifier recognized by walking classification is to ambulatory activities State is classified.
For example, by the acceleration standard deviation of customer mobile terminal to be identified, the quantile sum of acceleration 25,75 points of acceleration Digit sum these data characteristicses are brought into the support vector machine classifier of walking classification identification, i.e., the classification mould in step 201 In type 2., it is calculated in the range of which data characteristics during these data characteristicses belong to walk line state, running state, and then The classification of line state and running state can be realized away.If determining the active state of user to be identified to walk according to classification results State, and user to be identified rolling average speed close to running average speed, then may further determine that the user In footrace state.
If the step 205b arranged side by side with step 205a, the basic activity state for determining user to be identified according to classification results are Ride public transportation means active state, then extracted from terminal data customer mobile terminal to be identified location data of terminals feature and Terminal signaling data characteristics.
Wherein, location data of terminals feature includes translational speed average value, GPS satellite number, terminal signaling Data characteristics includes terminal signaling average strength, and the terminal signaling can be terminal called signal and/or network signal.
206b, according to location data of terminals feature and terminal signaling data characteristics, the decision tree recognized by the vehicles Grader is classified to the active state that rides public transportation means.
For example, the translational speed average value of customer mobile terminal to be identified, GPS satellite number, terminal are believed Number average strength these data characteristicses are brought into the decision tree classifier of vehicles identification, i.e., the classification mould in step 201 Type 3. in, which the data characteristics model during these data characteristicses belong to aircraft, high ferro, automobile, subway, train etc. be calculated Enclose interior, and then the identification of the vehicles such as aircraft, high ferro, automobile, subway, train can be realized, recognized by the vehicles Decision tree classifier carries out classification and can improve execution efficiency.If determining the traffic work that user to be identified takes according to classification results It is aircraft to have, and further, can be combined with the change of the user terminal location, determines the line of flight, is searched now corresponding Flight frequency, and then determine that the user is now taking the aircraft of which frame flight；Similarly, however, it is determined that user to be identified takes The vehicles be the one kind in high ferro, train, the change of the user terminal location is further can be combined with, it is determined that traveling road Line, and finally determine which time train the user is now taking；If determining what user to be identified took according to classification results The vehicles are subway, further can be combined with the user terminal location, and terminal location change, determine that the user exists By the subway, and which circuit sit is in which city, and subway travel direction etc..
If 207b, determining that the vehicles that user to be identified takes are automobile according to classification results, from terminal data Extract the accelerometer data feature and gyro data feature of customer mobile terminal.
Wherein, accelerometer data feature includes acceleration standard deviation, the quantile sum of acceleration 25, the quantile of acceleration 75 Sum, and gyro data feature includes gyro data standard deviation.
208b, according to accelerometer data feature and gyro data feature, the K nearest neighbour classifications recognized by passenger identity Device classify and obtains the corresponding passenger identity of user.
For example, driver in driving procedure mobile terminal place in the car, do not carry with the premise of, will treat The acceleration standard deviation of identifying user mobile terminal, the quantile sum of acceleration 25, the quantile sum of acceleration 75, gyroscope number During the K nearest neighbour classification devices of passenger identity identification, i.e. disaggregated model in step 201 are brought into according to standard deviation 4., it is calculated These data characteristicses belong in the range of which data characteristics in driver, passenger, and then can realize driver, passenger Identification.And intersect because driver and passenger's feature are present, from the K nearest neighbour classification devices of passenger identity identification, not only may be used To realize state classification, the probable value for belonging to certain classification can also be provided.
After user to be identified determines passenger identity, the vehicle that the user now rides in an automobile can also be further determined that Type, that is, perform step 209b.
Vehicle operation data feature is extracted in 209b, the vehicle operation data from automobile within the predetermined period time period.
Wherein, vehicle operation data feature is included and enlivened in driving range, vehicle line species, each predetermined time interval The accumulative operation duration of vehicle.The predetermined period time period can according to the actual requirements carry out option and installment.For example, during predetermined period Between section can be vehicle type recognition decision tree classifier the training data corresponding time period.
210b, according to vehicle operation data feature, classified by the decision tree classifier of vehicle type recognition The type of vehicle of automobile.
For the embodiment of the present invention, the identification to type of vehicle needs to refer to the data run in the vehicle some cycles, Specific identification process may be referred to Fig. 7, and the mobile terminal data according to user to be identified first is detected, if detecting same The course of user to be identified is identical with the course of a certain vehicle in one time period, and gait of march is also consistent, I.e. passenger keeps geo-stationary with a certain vehicle, then illustrate the vehicle class of type of vehicle as passenger's ride-on vehicles of the vehicle Type, then according to the data run in some cycles of the vehicle, is carried out by the decision tree classifier of vehicle type recognition Classification, wherein, the data that the vehicle is run in some cycles can by being placed on vehicle in particular terminal collect, it is also possible to The mobile terminal being placed on by user in vehicle is collected.
For example, by the vehicle enlivened in driving range, vehicle line species, each predetermined time interval of vehicle to be identified Accumulative these data characteristicses of operation duration are brought into the decision tree classifier of vehicle type recognition, i.e., the classification mould in step 201 Type 5. in, be calculated which data that these data characteristicses belong in private car, bus, motor bus, taxi etc. special In the range of levying, and then the identification of the type of vehicle such as private car, bus, motor bus, taxi can be realized.
Concrete application scene for the embodiment of the present invention can be with as follows, but not limited to this, including：
The physical activity recognition methods for providing according to embodiments of the present invention, configures corresponding client in customer mobile terminal End application, each active state experienced in user every day can be monitored by the client application, help the user can be with It is better understood by the health status of itself.Specifically, active state knowledge is carried out when the client application needs to treat identifying user When other, can accordingly recognize that target recognizes target for basic activity by default setting first, according to customer mobile terminal to be identified Gps data in the current certain period of time for collecting, calculates the translational speed maximum of user to be identified, then will be to be identified The translational speed maximum of user carries out calculating classification in being brought into the decision tree classifier of basic activity identification, is tied according to classification Fruit and current affiliated time period, record the active state of the user to be identified.If the active state of user to be identified is walking State or the state that rides public transportation means, can also further set new identification target and to extract customer mobile terminal to be identified new Characteristic, then carry out the active state identification of more fine-grained user to be identified, and then user to be identified can be identified Active state whether be to walk line state or running state or in taking ' aircraft, high ferro, automobile, subway, train ' etc. Which vehicles, finally according to recognition result and it is current belonging to time period, record the moving type of the user to be identified State, so as to user understand one day in which active state itself experienced, each active state all experienced how long.
Another physical activity recognition methods provided in an embodiment of the present invention, it is possible to achieve multiple target physical activity identification, Successively multiclass classification is carried out by the active state for treating identifying user, classification task multiclass classification task can be decomposed into, Whole process reduces the data characteristics number and moving target number of each hierarchical classification, while can choose most suitable per first-level class Preferably, degree of accuracy highest sorting technique, the complexity that data characteristics is chosen when can reduce multi-activity target identification, it is to avoid data Physical activity recognition accuracy is low caused by feature extraction deviation, so as to improve the accuracy of physical activity identification.
Further, implementing as Fig. 1 methods describeds, the embodiment of the invention provides a kind of physical activity identification Device, as shown in figure 8, described device includes：Acquiring unit 31, determining unit 32, taxon 33.
Acquiring unit 31, can be used for, when the identification for receiving User Activity state is instructed, obtaining the identification instruction Identifying user is treated the need for middle carrying carries out the identification target of activity recognition, and obtains the customer mobile terminal to be identified Terminal data.Acquiring unit 31 can be the main functional modules of acquisition terminal data and identification target in the present apparatus, and Triggering determining unit 32 is operated, wherein, identification target can be identification target, the knowledge of the vehicles of basic physical activity Other target etc..
Determining unit 32, is determined for disaggregated model group corresponding with the identification target that the acquiring unit 31 is obtained Interflow journey, different identification target corresponds to different disaggregated model combination process respectively, is wrapped in the disaggregated model combination process Containing one or more disaggregated models arranged according to predefined procedure.Determining unit 32 is determination disaggregated model combination stream in the present apparatus The main functional modules of journey, for the embodiment of the present invention, can be according to the level between specific identification target and disaggregated model Relation, selects corresponding disaggregated model to be combined sequence, obtains disaggregated model combination process corresponding with the identification target, with Just physical activity identification is carried out according to the disaggregated model combination process.
Taxon 33, can be used for the disaggregated model combination process determined according to the determining unit 32 and the acquisition The data characteristics of the terminal data that unit is obtained, the active state to the user to be identified carries out successively multiclass classification, obtains The active state of the user to be identified, wherein, active state and next level point that last layer level disaggregated model classification is obtained There is hyponymy between the optional active state of class model classification.For example, classified by last layer level disaggregated model obtaining Active state to multiply vehicles active state, and next optional active state of hierarchical classification category of model to live by air, by plane Dynamic state, multiply high ferro active state, by bus active state, by train active state of taking the subway, active state, traffic is multiplied here Tool activity state and active state by air, by plane, multiply high ferro active state, by bus active state, active state of taking the subway, multiply There is hyponymy between train active state.
It should be noted that each functional unit involved by a kind of physical activity identifying device provided in an embodiment of the present invention Other are accordingly described, and may be referred to the correspondence description in Fig. 1, be will not be repeated here.
A kind of physical activity identifying device provided in an embodiment of the present invention, including：Acquiring unit, determining unit, grouping sheet Unit, compared with realizing that physical activity is known otherwise by multi classifier at present, determines to treat knowledge with needs in determining unit After other user carries out the corresponding disaggregated model combination process of identification target of activity recognition, taxon is according to the disaggregated model group The data characteristics of the terminal data of the customer mobile terminal to be identified that interflow journey and acquiring unit are obtained, by treating identifying user Active state carry out successively multiclass classification, classification task can be decomposed into multiclass classification task, and last layer level classification There is hyponymy between active state that category of model is obtained and next optional active state of hierarchical classification category of model, And then some similar human body active states can be first categorized as a general designation active state, then the general designation active state is carried out The actual active state of user is found in subdivision, and whole process reduces the data characteristics number and moving target of each hierarchical classification Number, the complexity that data characteristics is chosen when can reduce multi-activity target identification, it is to avoid data characteristics extracts people caused by deviation Body activity recognition accuracy rate is low, so as to improve the accuracy of physical activity identification.
Further, implementing as Fig. 3 methods describeds, the embodiment of the invention provides another physical activity and knows Other device, as shown in figure 9, described device includes：Acquiring unit 41, determining unit 42, taxon 43.
Acquiring unit 41, can be used for, when the identification for receiving User Activity state is instructed, obtaining the identification instruction Identifying user is treated the need for middle carrying carries out the identification target of activity recognition, and obtains the customer mobile terminal to be identified Terminal data.
Determining unit 42, is determined for disaggregated model group corresponding with the identification target that the acquiring unit 41 is obtained Interflow journey, different identification target corresponds to different disaggregated model combination process respectively, is wrapped in the disaggregated model combination process Containing one or more disaggregated models arranged according to predefined procedure.
Taxon 43, can be used for the disaggregated model combination process determined according to the determining unit 42 and the acquisition The data characteristics of the terminal data that unit is obtained, the active state to the user to be identified carries out successively multiclass classification, obtains The active state of the user to be identified, wherein, active state and next level point that last layer level disaggregated model classification is obtained There is hyponymy between the optional active state of class model classification.
It is identified in time, it is necessary to training in advance finishes corresponding classification mould in order to treat identifying user active state Type, to obtain accurate classification results, further, described device also includes：Training unit 44.
The acquiring unit 41, can be used for obtaining the terminal data of the different user mobile terminal being collected into.
The training unit 44, can be used for according to different active states, extract corresponding spy in the terminal data Levy data to be trained, obtain the corresponding disaggregated model of active state.
The determining unit 42, specifically can be used for parsing the identification target, obtain and the identification target Corresponding multiple optional active states；According to the hyponymy between the multiple optional active state, selection is corresponding Disaggregated model be combined, obtain and the identification corresponding disaggregated model combination process of target.For the embodiment of the present invention, Can realize that multiple target physical activity is recognized, different user's requests can be met.
The taxon 43, if specifically can be used for the basic activity that the identification target is the user to be identified knowing Not, then the location data of terminals feature of the customer mobile terminal to be identified is extracted from the terminal data；According to the end End position data characteristics, the decision tree classifier recognized by basic activity is divided the active state of the user to be identified Class, can cause that execution efficiency is higher；If determining that the basic activity state of the user to be identified is walking according to classification results Active state, then extract the accelerometer data feature of customer mobile terminal to be identified from the terminal data；Added according to described Speed counts feature, and the support vector machine classifier recognized by walking classification is classified to the ambulatory activities state.
The taxon 43, if specifically can be also used for determining living substantially for the user to be identified according to classification results Dynamic state is ride public transportation means active state and the vehicles knowledge for recognizing that target is user's seating to be identified Not, then the location data of terminals feature and terminal signaling data that customer mobile terminal to be identified is extracted from the terminal data are special Levy；According to the location data of terminals feature and the terminal signaling data characteristics, the decision tree point recognized by the vehicles Class device is classified to the active state that rides public transportation means；If determining what the user to be identified took according to classification results The vehicles are automobile, then the automobile is extracted from the history terminal data of the customer mobile terminal in the predetermined period time Vehicle operation data feature in section；According to the vehicle operation data feature, by the decision tree classification of vehicle type recognition Device classify and obtains the type of vehicle of the automobile.
Alternatively, vehicle operation data feature is included and enlivens driving range, vehicle line species, each predetermined time interval The accumulative operation duration of interior vehicle.Wherein, predetermined time interval can be set according to the actual requirements, such as each scheduled time Interval can for each season, monthly, per two weeks etc., the accumulative operation duration of vehicle can add up driving time for vehicle；OK Sailing circuit grade includes dividing driving data by stroke, and stroke is the driving task for once having beginning and end, with identical The circuit of beginning and end may be considered a kind of circuit.
The determining unit 42, can be also used for by default DBSCAN Density Clustering functions, in the predetermined period The vehicle position data of the automobile carries out Density Clustering in time period, obtains each class cluster；By sample in described each class cluster The central point of the most class clusters of point, is defined as the active centre point of the automobile, and by institute in the most class cluster of the sample point A little arrive the active centre point apart from maximum, the traveling for being defined as the automobile enlivens radius；According in described enlivening Heart point and the traveling enliven radius, and determine the automobile enlivens driving range.Wherein, DBSCAN Density Clustering functions are preset Can in advance be carried out writing setting according to DBSCAN density clustering algorithms, compared with K-means methods, DBSCAN Density Clusterings are calculated Method need not in advance know the class number of clusters amount for needing to be formed, it can be found that the cluster class of arbitrary shape, can recognize that noise spot.
The taxon 43, if specifically can be also used for determining the friendship that the user to be identified takes according to classification results Logical instrument is automobile and the passenger identity identification for recognizing that target is the user to be identified, then from the terminal data Extract the accelerometer data feature and gyro data feature of customer mobile terminal to be identified；According to the accelerometer data feature With the gyro data feature, by the K nearest neighbour classifications device of passenger identity identification classify obtains the use to be identified The corresponding passenger identity in family.Classified by K nearest neighbour classification devices, can not only be realized state classification, category can also be given In the probable value of certain classification.
It should be noted that each functional unit involved by another physical activity identifying device provided in an embodiment of the present invention Other corresponding descriptions, may be referred to the correspondence description in Fig. 3, will not be repeated here.
Another physical activity identifying device provided in an embodiment of the present invention, including：Acquiring unit, determining unit, classification Unit, training unit, it is possible to achieve multiple target physical activity is recognized, carried out successively to the active state of user by taxon Multiclass classification, can be decomposed into multiclass classification task by classification task, and the data that whole process reduces each hierarchical classification are special Number and moving target number are levied, while optimum, degree of accuracy highest sorting technique can be chosen per first-level class, can be reduced many The complexity that data characteristics is chosen when moving target is recognized, it is to avoid physical activity identification is accurate caused by data characteristics extracts deviation Rate is low, so as to improve the accuracy of physical activity identification.
In the above-described embodiments, the description to each embodiment all emphasizes particularly on different fields, and does not have the portion described in detail in certain embodiment Point, may refer to the associated description of other embodiment.
It is understood that the correlated characteristic in the above method and device can be referred to mutually.In addition, in above-described embodiment " first ", " second " etc. be, for distinguishing each embodiment, and not represent the quality of each embodiment.
It is apparent to those skilled in the art that, for convenience and simplicity of description, the system of foregoing description, The specific work process of device and unit, may be referred to the corresponding process in preceding method embodiment, will not be repeated here.
Algorithm and display be not inherently related to any certain computer, virtual system or miscellaneous equipment provided herein. Various general-purpose systems can also be used together with based on teaching in this.As described above, construct required by this kind of system Structure be obvious.Additionally, the present invention is not also directed to any certain programmed language.It is understood that, it is possible to use it is various Programming language realizes the content of invention described herein, and the description done to language-specific above is to disclose this hair Bright preferred forms.
In specification mentioned herein, numerous specific details are set forth.It is to be appreciated, however, that implementation of the invention Example can be put into practice in the case of without these details.In some instances, known method, structure is not been shown in detail And technology, so as not to obscure the understanding of this description.
Similarly, it will be appreciated that in order to simplify one or more that the disclosure and helping understands in each inventive aspect, exist Above to the description of exemplary embodiment of the invention in, each feature of the invention is grouped together into single implementation sometimes In example, figure or descriptions thereof.However, the method for the disclosure should be construed to reflect following intention：I.e. required guarantor The application claims of shield features more more than the feature being expressly recited in each claim.More precisely, such as following Claims reflect as, inventive aspect is all features less than single embodiment disclosed above.Therefore, Thus the claims for following specific embodiment are expressly incorporated in the specific embodiment, and wherein each claim is in itself All as separate embodiments of the invention.
Those skilled in the art are appreciated that can be carried out adaptively to the module in the equipment in embodiment Change and they are arranged in one or more equipment different from the embodiment.Can be the module or list in embodiment Unit or component be combined into a module or unit or component, and can be divided into addition multiple submodule or subelement or Sub-component.In addition at least some in such feature and/or process or unit exclude each other, can use any Combine to all features disclosed in this specification (including adjoint claim, summary and accompanying drawing) and so disclosed appoint Where all processes or unit of method or equipment are combined.Unless expressly stated otherwise, this specification (including adjoint power Profit is required, summary and accompanying drawing) disclosed in each feature can the alternative features of or similar purpose identical, equivalent by offer carry out generation Replace.
Although additionally, it will be appreciated by those of skill in the art that some embodiments described herein include other embodiments In included some features rather than further feature, but the combination of the feature of different embodiments means in of the invention Within the scope of and form different embodiments.For example, in the following claims, embodiment required for protection is appointed One of meaning mode can be used in any combination.
All parts embodiment of the invention can be realized with hardware, or be run with one or more processor Software module realize, or with combinations thereof realize.It will be understood by those of skill in the art that can use in practice Microprocessor or digital signal processor (DSP) come realize a kind of physical activity recognition methods according to embodiments of the present invention and The some or all functions of some or all parts in device.The present invention is also implemented as being retouched here for execution Some or all equipment or program of device (for example, computer program and computer program product) of the method stated. It is such to realize that program of the invention be stored on a computer-readable medium, or can have one or more signal Form.Such signal can be downloaded from internet website and obtained, or on carrier signal provide, or with it is any its He provides form.
It should be noted that above-described embodiment the present invention will be described rather than limiting the invention, and ability Field technique personnel can design alternative embodiment without departing from the scope of the appended claims.In the claims, Any reference symbol being located between bracket should not be configured to limitations on claims.Word "comprising" is not excluded the presence of not Element listed in the claims or step.Word "a" or "an" before element is not excluded the presence of as multiple Element.The present invention can come real by means of the hardware for including some different elements and by means of properly programmed computer It is existing.If in the unit claim for listing equipment for drying, several in these devices can be by same hardware branch To embody.The use of word first, second, and third does not indicate that any order.These words can be explained and run after fame Claim.

Claims (10)

1. a kind of physical activity recognition methods, it is characterised in that including：
When the identification for receiving User Activity state is instructed, to obtain during the identification is instructed and treat identifying user the need for carrying Carry out the identification target of activity recognition, and the terminal data for obtaining the customer mobile terminal to be identified；
It is determined that with the identification corresponding disaggregated model combination process of target, different identification targets correspond to different classification respectively Model combination process, comprising one or more disaggregated models arranged according to predefined procedure in the disaggregated model combination process；
According to the disaggregated model combination process and the data characteristics of the terminal data that determine, to the user's to be identified Active state carries out successively multiclass classification, obtains the active state of the user to be identified, wherein, last layer level disaggregated model point There is hyponymy between active state that class is obtained and next optional active state of hierarchical classification category of model.
2. physical activity recognition methods according to claim 1, it is characterised in that methods described also includes：
The terminal data of the different user mobile terminal that acquisition is collected into；
According to different active states, corresponding characteristic is trained in extracting the terminal data, obtains active state Corresponding disaggregated model；
Determination disaggregated model combination process corresponding with the identification target, specifically includes：
The identification target is parsed, multiple optional active states corresponding with the identification target are obtained；
According to the hyponymy between the multiple optional active state, select corresponding disaggregated model to be combined, obtain To disaggregated model combination process corresponding with the identification target.
3. physical activity recognition methods according to claim 1, it is characterised in that if the identification target is the user Basic activity identification, then it is described according to determine the disaggregated model combination process and the terminal data data characteristics, Active state to the user to be identified carries out successively multiclass classification, specifically includes：
The location data of terminals feature of the customer mobile terminal to be identified is extracted from the terminal data；
According to the location data of terminals feature, by the decision tree classifier of basic activity identification to the user's to be identified Active state is classified；
If the basic activity state that the user to be identified is determined according to classification results is ambulatory activities state, from the terminal The accelerometer data feature of extracting data customer mobile terminal to be identified；
According to the accelerometer data feature, the support vector machine classifier recognized by walking classification is to the ambulatory activities shape State is classified.
4. physical activity recognition methods according to claim 3, it is characterised in that if being treated according to classification results determine The basic activity state of identifying user for ride public transportation means active state and it is described identification target for the user to be identified multiplies The vehicles identification of seat, then it is described special according to the disaggregated model combination process and the data of the terminal data for determining Levy, the active state to the user to be identified carries out successively multiclass classification, specifically also includes：
The location data of terminals feature and terminal signaling data that customer mobile terminal to be identified is extracted from the terminal data are special Levy, the terminal signaling is terminal called signal and/or terminal network signal；
According to the location data of terminals feature and the terminal signaling data characteristics, the decision tree point recognized by the vehicles Class device is classified to the active state that rides public transportation means；
If determining that the vehicles that the user to be identified takes are automobile according to classification results, from the automobile in predetermined week Vehicle operation data feature is extracted in vehicle operation data in time period phase；
According to the vehicle operation data feature, by the decision tree classifier of vehicle type recognition classify obtains the vapour The type of vehicle of car.
5. physical activity recognition methods according to claim 4, it is characterised in that the vehicle operation data feature is included The accumulative operation duration of vehicle in driving range, vehicle line species, each predetermined time interval is enlivened, it is described actively to travel model The determination mode enclosed includes：
By default DBSCAN Density Clustering functions, to the vehicle position data of the automobile within the predetermined period time period Density Clustering is carried out, each class cluster is obtained；
By the central point of the most class cluster of sample point in described each class cluster, it is defined as the active centre point of the automobile, and will Arrived a little in the most class cluster of the sample point active centre point apart from maximum, be defined as the traveling of the automobile Enliven radius；
Radius is enlivened according to the active centre point and the traveling, determine the automobile enlivens driving range.
6. physical activity recognition methods according to claim 4, it is characterised in that if being treated according to classification results determine The vehicles that identifying user is taken are the passenger identity identification that automobile and the identification target are the user to be identified, then The disaggregated model combination process and the data characteristics of the terminal data according to determination, to the user's to be identified Active state carries out successively multiclass classification, specifically also includes：
The accelerometer data feature and gyro data feature of customer mobile terminal to be identified are extracted from the terminal data；
According to the accelerometer data feature and the gyro data feature, the K nearest neighbour classifications recognized by passenger identity Device classify and obtains the corresponding passenger identity of the user to be identified.
7. a kind of physical activity identifying device, it is characterised in that including：
Acquiring unit, for when the identification for receiving User Activity state is instructed, obtaining the need carried in the identification instruction Treating identifying user carries out the identification target of activity recognition, and the number of terminals for obtaining the customer mobile terminal to be identified According to；
Determining unit is different for determining disaggregated model combination process corresponding with the identification target of acquiring unit acquisition Identification target correspond to different disaggregated model combination process respectively, included according to predetermined suitable in the disaggregated model combination process One or more disaggregated models of sequence arrangement；
Taxon, for the end that the disaggregated model combination process determined according to the determining unit and the acquiring unit are obtained The data characteristics of end data, the active state to the user to be identified carries out successively multiclass classification, obtains the use to be identified The active state at family, wherein, the active state that last layer level disaggregated model classification is obtained can with next hierarchical classification category of model There is hyponymy between the active state of choosing.
8. physical activity identifying device according to claim 7, it is characterised in that described device also includes：Training unit；
The acquiring unit, the terminal data for obtaining the different user mobile terminal being collected into；
The training unit, for according to different active states, corresponding characteristic to be carried out in the extraction terminal data Training, obtains the corresponding disaggregated model of active state；
The determining unit, specifically for being parsed to the identification target, obtains multiple corresponding with the identification target Optional active state；
According to the hyponymy between the multiple optional active state, select corresponding disaggregated model to be combined, obtain To disaggregated model combination process corresponding with the identification target.
9. physical activity identifying device according to claim 7, it is characterised in that
The taxon, if specifically for the basic activity identification that the identification target is the user to be identified, from institute State the location data of terminals feature that the customer mobile terminal to be identified is extracted in terminal data；
According to the location data of terminals feature, by the decision tree classifier of basic activity identification to the user's to be identified Active state is classified；
If the basic activity state that the user to be identified is determined according to classification results is ambulatory activities state, from the terminal The accelerometer data feature of extracting data customer mobile terminal to be identified；
According to the accelerometer data feature, the support vector machine classifier recognized by walking classification is to the ambulatory activities shape State is classified.
10. physical activity identifying device according to claim 9, it is characterised in that
The taxon, if being specifically additionally operable to determine the basic activity state of the user to be identified to multiply according to classification results Sit vehicles active state and the identification target is vehicles identification that the user to be identified takes, then from described The location data of terminals feature and terminal signaling data characteristics of customer mobile terminal to be identified, the terminal are extracted in terminal data Signal is terminal called signal and/or terminal network signal；
According to the location data of terminals feature and the terminal signaling data characteristics, the decision tree point recognized by the vehicles Class device is classified to the active state that rides public transportation means；
If determining that the vehicles that the user to be identified takes are automobile according to classification results, from the automobile in predetermined week Vehicle operation data feature is extracted in vehicle operation data in time period phase；
According to the vehicle operation data feature, by the decision tree classifier of vehicle type recognition classify obtains the vapour The type of vehicle of car.
CN201611221454.7A 2016-12-26 2016-12-26 Physical activity recognition methods and device Pending CN106778652A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611221454.7A CN106778652A (en) 2016-12-26 2016-12-26 Physical activity recognition methods and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611221454.7A CN106778652A (en) 2016-12-26 2016-12-26 Physical activity recognition methods and device

Publications (1)

Publication Number Publication Date
CN106778652A true CN106778652A (en) 2017-05-31

Family

ID=58925629

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611221454.7A Pending CN106778652A (en) 2016-12-26 2016-12-26 Physical activity recognition methods and device

Country Status (1)

Country Link
CN (1) CN106778652A (en)

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107146445A (en) * 2017-06-30 2017-09-08 北京奇虎科技有限公司 Mobile terminal traffic information reminding method, device and mobile terminal
CN107277154A (en) * 2017-06-30 2017-10-20 北京奇虎科技有限公司 Information-pushing method, device and mobile terminal
CN107341226A (en) * 2017-06-30 2017-11-10 北京奇虎科技有限公司 Information displaying method, device and mobile terminal
CN107368850A (en) * 2017-06-30 2017-11-21 北京奇虎科技有限公司 User Activity state switching method, device and mobile terminal based on mobile terminal
CN107391605A (en) * 2017-06-30 2017-11-24 北京奇虎科技有限公司 Information-pushing method, device and mobile terminal based on geographical position
CN107396306A (en) * 2017-06-30 2017-11-24 北京奇虎科技有限公司 User Activity state identification method, device and mobile terminal based on mobile terminal
CN107391604A (en) * 2017-06-30 2017-11-24 北京奇虎科技有限公司 Map styles display methods, device and mobile terminal based on active state
CN107657277A (en) * 2017-09-22 2018-02-02 上海斐讯数据通信技术有限公司 A kind of human body unusual checking based on big data and decision method and system
CN107662613A (en) * 2017-09-27 2018-02-06 西北工业大学 A kind of extreme driving behavior recognition methods and system based on mobile intelligent perception
CN108319960A (en) * 2018-01-09 2018-07-24 和芯星通（上海）科技有限公司 Activity recognition system and method, equipment and storage medium based on probability graph model
CN108334833A (en) * 2018-01-26 2018-07-27 和芯星通（上海）科技有限公司 Activity recognition method and system, equipment and storage medium based on FFT model
CN109086704A (en) * 2018-07-23 2018-12-25 杭州电子科技大学 A kind of physical activity recognition methods classified based on rarefaction representation and Softmax
WO2019056680A1 (en) * 2017-09-19 2019-03-28 平安科技（深圳）有限公司 Travel mode identification method and apparatus, terminal device, and storage medium
CN109816043A (en) * 2019-02-02 2019-05-28 拉扎斯网络科技（上海）有限公司 Method and device for determining user identification model, electronic equipment and storage medium
CN110321940A (en) * 2019-06-24 2019-10-11 清华大学 The feature extraction of aircraft telemetry and classification method and device
CN110519681A (en) * 2019-07-18 2019-11-29 北京无线体育俱乐部有限公司 Campaign item identifying system, method, equipment and storage medium
WO2020097968A1 (en) * 2018-11-16 2020-05-22 Beijing Didi Infinity Technology And Development Co., Ltd. Artificial intelligent systems and methods for identifying drunk passenger by car hailing order
CN111626222A (en) * 2020-05-28 2020-09-04 深圳市商汤科技有限公司 Pet detection method, device, equipment and storage medium
CN111723835A (en) * 2019-03-21 2020-09-29 北京嘀嘀无限科技发展有限公司 Vehicle movement track distinguishing method and device and electronic equipment
CN111814523A (en) * 2019-04-12 2020-10-23 北京京东尚科信息技术有限公司 Human body activity recognition method and device
CN112026686A (en) * 2019-06-04 2020-12-04 上海汽车集团股份有限公司 Method and device for automatically adjusting position of vehicle seat
CN113361451A (en) * 2021-06-24 2021-09-07 福建万福信息技术有限公司 Ecological environment target identification method based on multi-level model and preset point automatic adjustment
US11403553B2 (en) 2018-11-16 2022-08-02 Beijing Didi Infinity Technology And Development Co., Ltd. Artificial intelligent systems and methods for identifying a drunk passenger by a car hailing order
CN115527373A (en) * 2022-01-05 2022-12-27 荣耀终端有限公司 Riding tool identification method and device
CN116369876A (en) * 2023-04-13 2023-07-04 中国人民解放军空军特色医学中心 Pilot distress call device

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130138436A1 (en) * 2011-11-26 2013-05-30 Microsoft Corporation Discriminative pretraining of deep neural networks
CN105101092A (en) * 2015-09-01 2015-11-25 上海美慧软件有限公司 Mobile phone user travel mode recognition method based on C4.5 decision tree
CN105447504A (en) * 2015-11-06 2016-03-30 中国科学院计算技术研究所 Traffic mode behavior recognition method and corresponding recognition model construction method
US20160129913A1 (en) * 2014-11-06 2016-05-12 Lars Boesen System and Method for the Automatic Persona Identification of a Person Post Motion

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130138436A1 (en) * 2011-11-26 2013-05-30 Microsoft Corporation Discriminative pretraining of deep neural networks
US20160129913A1 (en) * 2014-11-06 2016-05-12 Lars Boesen System and Method for the Automatic Persona Identification of a Person Post Motion
CN105101092A (en) * 2015-09-01 2015-11-25 上海美慧软件有限公司 Mobile phone user travel mode recognition method based on C4.5 decision tree
CN105447504A (en) * 2015-11-06 2016-03-30 中国科学院计算技术研究所 Traffic mode behavior recognition method and corresponding recognition model construction method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
中国汽车技术研究中心 等: "《节能与新能源汽车年鉴2014》", 31 December 2014, 北京：中国经济出版社 *
宓超 等: "《装卸机器视觉及其应用》", 31 January 2016, 上海：上海科学技术出版社 *

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107146445A (en) * 2017-06-30 2017-09-08 北京奇虎科技有限公司 Mobile terminal traffic information reminding method, device and mobile terminal
CN107277154A (en) * 2017-06-30 2017-10-20 北京奇虎科技有限公司 Information-pushing method, device and mobile terminal
CN107341226A (en) * 2017-06-30 2017-11-10 北京奇虎科技有限公司 Information displaying method, device and mobile terminal
CN107368850A (en) * 2017-06-30 2017-11-21 北京奇虎科技有限公司 User Activity state switching method, device and mobile terminal based on mobile terminal
CN107391605A (en) * 2017-06-30 2017-11-24 北京奇虎科技有限公司 Information-pushing method, device and mobile terminal based on geographical position
CN107396306A (en) * 2017-06-30 2017-11-24 北京奇虎科技有限公司 User Activity state identification method, device and mobile terminal based on mobile terminal
CN107391604A (en) * 2017-06-30 2017-11-24 北京奇虎科技有限公司 Map styles display methods, device and mobile terminal based on active state
CN107341226B (en) * 2017-06-30 2021-06-22 北京奇虎科技有限公司 Information display method and device and mobile terminal
WO2019056680A1 (en) * 2017-09-19 2019-03-28 平安科技（深圳）有限公司 Travel mode identification method and apparatus, terminal device, and storage medium
CN107657277B (en) * 2017-09-22 2022-02-01 金言 Human body abnormal behavior detection and judgment method and system based on big data
CN107657277A (en) * 2017-09-22 2018-02-02 上海斐讯数据通信技术有限公司 A kind of human body unusual checking based on big data and decision method and system
CN107662613A (en) * 2017-09-27 2018-02-06 西北工业大学 A kind of extreme driving behavior recognition methods and system based on mobile intelligent perception
CN108319960A (en) * 2018-01-09 2018-07-24 和芯星通（上海）科技有限公司 Activity recognition system and method, equipment and storage medium based on probability graph model
CN108334833A (en) * 2018-01-26 2018-07-27 和芯星通（上海）科技有限公司 Activity recognition method and system, equipment and storage medium based on FFT model
CN109086704A (en) * 2018-07-23 2018-12-25 杭州电子科技大学 A kind of physical activity recognition methods classified based on rarefaction representation and Softmax
US11403553B2 (en) 2018-11-16 2022-08-02 Beijing Didi Infinity Technology And Development Co., Ltd. Artificial intelligent systems and methods for identifying a drunk passenger by a car hailing order
WO2020097968A1 (en) * 2018-11-16 2020-05-22 Beijing Didi Infinity Technology And Development Co., Ltd. Artificial intelligent systems and methods for identifying drunk passenger by car hailing order
CN109816043B (en) * 2019-02-02 2021-01-01 拉扎斯网络科技（上海）有限公司 Method and device for determining user identification model, electronic equipment and storage medium
CN109816043A (en) * 2019-02-02 2019-05-28 拉扎斯网络科技（上海）有限公司 Method and device for determining user identification model, electronic equipment and storage medium
CN111723835A (en) * 2019-03-21 2020-09-29 北京嘀嘀无限科技发展有限公司 Vehicle movement track distinguishing method and device and electronic equipment
CN111814523A (en) * 2019-04-12 2020-10-23 北京京东尚科信息技术有限公司 Human body activity recognition method and device
CN112026686B (en) * 2019-06-04 2022-04-12 上海汽车集团股份有限公司 Method and device for automatically adjusting position of vehicle seat
CN112026686A (en) * 2019-06-04 2020-12-04 上海汽车集团股份有限公司 Method and device for automatically adjusting position of vehicle seat
CN110321940B (en) * 2019-06-24 2021-07-09 清华大学 Aircraft telemetry data feature extraction and classification method and device
CN110321940A (en) * 2019-06-24 2019-10-11 清华大学 The feature extraction of aircraft telemetry and classification method and device
CN110519681A (en) * 2019-07-18 2019-11-29 北京无线体育俱乐部有限公司 Campaign item identifying system, method, equipment and storage medium
CN111626222A (en) * 2020-05-28 2020-09-04 深圳市商汤科技有限公司 Pet detection method, device, equipment and storage medium
CN113361451A (en) * 2021-06-24 2021-09-07 福建万福信息技术有限公司 Ecological environment target identification method based on multi-level model and preset point automatic adjustment
CN113361451B (en) * 2021-06-24 2024-04-30 福建万福信息技术有限公司 Ecological environment target identification method based on multistage model and preset point automatic adjustment
CN115527373A (en) * 2022-01-05 2022-12-27 荣耀终端有限公司 Riding tool identification method and device
CN116369876A (en) * 2023-04-13 2023-07-04 中国人民解放军空军特色医学中心 Pilot distress call device

Similar Documents

Publication Publication Date Title
CN106778652A (en) Physical activity recognition methods and device
CN105678222B (en) A kind of mobile device-based Human bodys' response method
CN105447504B (en) A kind of travel pattern Activity recognition method and corresponding identification model construction method
Moral-Muñoz et al. Analyzing highly cited papers in intelligent transportation systems
CN108537117A (en) A kind of occupant detection method and system based on deep learning
CN105956625B (en) A kind of motion state of automobile recognition methods and system based on given physical model
Alamri et al. An effective bio-signal-based driver behavior monitoring system using a generalized deep learning approach
CN109002189B (en) Motion recognition method, device, equipment and computer storage medium
CN107861957B (en) Data analysis method and device
CN107396306A (en) User Activity state identification method, device and mobile terminal based on mobile terminal
CN117157678A (en) Method and system for graph-based panorama segmentation
CN107277154A (en) Information-pushing method, device and mobile terminal
CN104537356A (en) Pedestrian re-identification method and device for carrying out gait recognition through integral scheduling
CN105608444A (en) Wild animal image identification method used for automatic driving
CN111971708A (en) Guidance system
CN108108703A (en) Deceleration strip missing detection method, device and electronic equipment
Çetinkaya et al. Driver activity recognition using deep learning and human pose estimation
CN113420059A (en) Method and device for actively treating citizen hot line problem
CN106095915A (en) The processing method and processing device of user identity
KR102470153B1 (en) Risks detecting system and method for using pattern analysis of layered tags in user log data
CN109635863B (en) Method and device for intelligently judging riding of user
CN115240402B (en) Sightseeing vehicle scheduling method and system
TW202016772A (en) Analysis system
CN109300061A (en) A kind of individual information recommendation plateform system based on deep learning tourism
Grigoropoulos et al. Detection and classification of bicyclist group behavior for automated vehicle applications

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: 20170531