CN106127125A - Distributed DTW human body behavior intension recognizing method based on human body behavior characteristics - Google Patents
Distributed DTW human body behavior intension recognizing method based on human body behavior characteristics Download PDFInfo
- Publication number
- CN106127125A CN106127125A CN201610436818.7A CN201610436818A CN106127125A CN 106127125 A CN106127125 A CN 106127125A CN 201610436818 A CN201610436818 A CN 201610436818A CN 106127125 A CN106127125 A CN 106127125A
- Authority
- CN
- China
- Prior art keywords
- human body
- body behavior
- information
- distributed
- dtw
- 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.)
- Granted
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
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
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)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses distributed DTW human body behavior intension recognizing method based on human body behavior characteristics, distributed DTW human body behavior intension recognizing method based on human body behavior characteristics, comprises the following steps: obtains present frame human body behavior characteristics and is added in human body behavior characteristics time series;Utilize dynamic time warping algorithm DTW that Current observation human body behavior characteristics sequence is mated with the specific behavior sequence learnt in database template, calculate optimal matching similarity based on card side's distance between the two;This similarity is taken inverse, after normalization, obtains matching probability;The action pattern probit of each sensor being exchanged as consistency information amount with adjacent sensors data again, through concordance interative computation, the recognition result of final adjacent sensors node is reached an agreement.The present invention is to realize accurate human body specific behavior identification.
Description
Technical field
The present invention relates to areas of information technology, be specifically related to distributed DTW human body behavior based on human body behavior characteristics meaning
Figure recognition methods.
Background technology
Human bodys' response based on multiple RGBD cameras, by the extensive concern of researcher, are applied to operating room, work
Human body behavioral value under the environment such as factory workshop, automobile, Indoor Video, efficiently solves human body occlusion issue and may send out
Raw man-robot collision problem, has important using value.
It is currently based on the human body behavior perception of multiple RGBD sensor also in the centralized stage, needs one or more number
Three-dimensional data, the fusion of human skeleton articulare data, computing capability and the Shandong to data fusion center is carried out according to fusion center
Rod requires higher, and more weak to the unstability resistance of network, expansible degree is low.
Along with the development of RGBD sensor technology, its usage quantity and coverage can increase therewith, centralized RGBD
Sensor network is required to be processed and the flow of information meeting explosive growth of transmission, and its bottleneck in real world applications can more be invented
Aobvious.
When scene exists multiple human body target, need the identification problem solving to be intended to about human body behavior.At present, close
It is inadequate that identification technology in human body behavior intention also exists accuracy of identification, it is impossible to meets existing demand.
Summary of the invention
For solving the deficiency that prior art exists, the invention discloses distributed DTW human body based on human body behavior characteristics
Behavior intension recognizing method, to realize accurate human body specific behavior identification.DTW Chinese is dynamic time warping algorithm.
For achieving the above object, the concrete scheme of the present invention is as follows:
Distributed DTW human body behavior intension recognizing method based on human body behavior characteristics, comprises the following steps:
Obtain present frame human body behavior characteristics and be added in human body behavior characteristics time series;
Utilize dynamic time warping algorithm DTW that Current observation human body behavior characteristics sequence is learnt in database template
Specific behavior sequence mate, calculate optimal matching similarity based on card side's distance between the two;
This similarity is taken inverse, after normalization, obtains matching probability;
The action pattern probit of each sensor is handed over as consistency information amount with adjacent sensors data again
Changing, through concordance interative computation, the recognition result of final adjacent sensors node is reached an agreement.
Further, when calculating matching probability, for merging ambient sensors recognition result, between adjacent connection sensor
Exchange matching probability information.
Further, between adjacent connection sensor exchange matching probability information mode particularly as follows:
I-th sensor is by matching probability corresponding for k moment action pattern αWith adjacent sensors node j calculating
Matching probabilityCarry out data exchange, and according to the transition probability matrix M between action pattern, determine the general of action pattern α
Rate value
Wherein η (k) is normalization factor, and M (β, α) is the transition probability value between action pattern β and action pattern α.
Further, after exchanging matching probability information between adjacent connection sensor, if finalMore than certain threshold
Value, then the match is successful with action pattern α to assert Current observation subject performance sequence.
Further, distributed DTW human body behavior intension recognizing method based on human body behavior characteristics is obtaining present frame
Also include before the step of human body behavior characteristics building and there is translation invariant and scale the step of constant organization of human body feature.
Further, build there is translation invariant and scaling constant organization of human body feature time, it is contemplated that build
Body difference, definition meets translation invariant and the joint vector angle scaling constant condition and upper limb partial joint vector mould ratio
24 dimension human body behavior characteristicss are collectively constituted as human body behavior characteristics, joint vector angle and vector mould ratio.
Further, upper limb partial joint vector mould ratio:
WhereinIt is the trunk center vectorial mould to head,WithBe respectively head point to right-hand man vectorial mould,
WithIt is the trunk center vectorial mould to right-hand man respectively.
Further, the acquisition of the joint vector angle information that translation invariant is constant with scaling: detect people by OPENNI
15 articulare distributed informations of body, through calculating, can obtain joint vector angle information, specifically include upper limb part 10
Group, lower extremities 4 groups, middle interconnecting piece divides 6 groups.
In the application, the articulare distributed information to the human body obtained carries out conforming estimation, specifically includes:
Skeleton joint point position initialization;
Local sensor is to articulare estimation: build motion model and the observation model of human joint points, it is achieved right
Effective estimation of articulare state;
Between sensor, the consistency on messaging of target joint point is estimated: information corresponding to definition human synovial dotted state to
Amount, information matrix and information contribution thereof and model probability are as the exchange capacity of consistency on messaging algorithm;
Each sensor is by the information contribution of articulare information vector, information matrix and correspondence thereof self estimated, model
Probability is sent to adjacent communication sensor node, and accepts the information of ambient sensors, utilizes consistency on messaging algorithm, merges
The estimated result of ambient sensors, subsequent iteration is for several times, it is achieved algorithm and the convergence of estimated result.
Also needed to build dynamic distributed sensor network before skeleton joint point position initialization.
Based on the dynamic distributed sensor network built, the human skeleton articulare information gathered is transmitted extremely by sensor
Information processing centre.
During skeleton joint point position initialization, by articulare depth information learning training in advance, it is achieved to every frame people
The detection of body articulare, or utilize existing instrument OPENNI NITE or Microsoft SDK extracting directly articulare.
When skeleton joint point position initialization, for removing invalid articulare, set up human joint points motion model physics
Constraint, rejects the human joint points being unsatisfactory for the human joint points anglec of rotation and length constraint.
The relevant parameter of human joint points motion model physical constraint, including length between human elbow and shoulder joint,
Can be according to detection data adaptive upgrading.
Realize based on Bayesian filter when realizing the effectively estimation to articulare state.
Build motion model and the observation model of human joint points, wherein, to linear model, utilize linear information wave filter
Estimating, and for nonlinear model, utilize nonlinear filter to estimate, nonlinear filter includes Extended information filter device
With based on centered difference information filter.
When consistency on messaging is estimated, the step specifically included is:
(1) parameter initialization, detection draws the initial position of human joint points, and the variance of position is according to articulare identification
Confidence level determines, and sets the motion model transition probability in joint according to joint motions feature;
(2) multi-model is mutual, i.e. according to the mixing probability between model probability and Model transition probability computation model, then depends on
Hybrid mean value and the mixing variance of each model is drawn according to mixing probability calculation;
(3) information filter, with hybrid mean value and mixing variance for input, calculate its information vector and information matrix, to line
Property and nonlinear motion model be respectively adopted linear information wave filter and centered difference information filter and estimate, according to current depth
The articulare position upgrading wave filter articulare state of image detection and model probability;
(4) information fusion based on distributed information consistency algorithm, exchanges articulare and estimates letter between the most each sensor
Breath, including articulare status information vector, information matrix and model probability, realizes each sensing by consistency algorithm weighted iteration
The concordance of device estimated state.
It addition, each sensor node mixing based on model probability weighted sum output, i.e. utilize model probability to each mould
The estimated result of type is weighted summation, as the estimated result of each sensor information processing system current time.
Carry out after the application carries out conforming estimation to human joint points distributed information based on coloured image with deep
The joint probabilistic data association of degree image multiple features, comprises the following steps:
Associating data probabilistic correlation algorithm is used to realize local sensor node to following the tracks of to enter between target and target observation
Row data association for the first time;
Hungary Algorithm based on mahalanobis distance realizes between sensor node the second time data association following the tracks of target;
Exchange articulare estimated information between each sensor, realize each sensor by consistency algorithm weighted iteration and estimate
The concordance of state;
Wherein, when first time data association, target observation Candidate Set Regulation mechanism based on multiple features, utilize articulare
Position detection information z, coloured image gradient orientation histogram feature hcWith depth image gradient orientation histogram feature hdBuild three
Individual threshold value thresholding (γz,γc,γd) to limit observation collection size.
Also include utilizing model probability to each model after information fusion based on distributed information consistency algorithm
Estimated result is weighted summation, as the step of the estimated result of each sensor information processing system current time.
Distributed information consistency algorithm makes local sensor obtain after local data associated estimation result, can be with neighbour
Nearly node switching data association result, to realize the fusion of each sensing data association results.
Utilize articulare position detection information z, coloured image gradient orientation histogram feature hcWith depth image gradient side
To histogram feature hdBuild three threshold value thresholding (γz,γc,γd) to limit observation collection size:
Wherein,Being the current joint point position according to the prediction of previous frame articulare estimated value, S is the variance of z,With
Being the HOG feature centered by articulare position from nearest history keyword frame learning and HOD feature respectively, d is Nogata
Figure Chi-square card side distance measure.
The mahalanobis distance of articulare is defined as follows:
(xi-xj)T(Pi+Pj)-1(xi-xj),
(xi,xj) and (Pi,Pj) it is human joint points estimated state respectively on sensor i and sensor j and variance.
Beneficial effects of the present invention:
By building distributed RGBD sensor network, utilize consistency on messaging algorithm, it is achieved that to human joint points
Distributed fusion, no data fusion center in network, the system that improves, to nodal information mistake and invalid robustness, is easier to
Realize the extension to sensor network.
Sensor node is only connected node communication with neighbouring around, exchanges information vector, information matrix and information contribution, phase
The RGBD data original compared with transmission, greatly reduce data volume.
Consistency algorithm achieves and the effective integration of sensor node in network, indirectly achieves the multi-angle to target
Observation, decreases and blocks or the angle impact on human synovial point estimation, expand sensing range.
Propose human joint points distributed information Uniform estimates method based on Interactive Multiple-Model, to tackle human body not
Motor pattern with articulare time-varying.
Propose joint probabilistic data association method based on coloured image and depth image multiple features, estimate improving JPDA algorithm
Meter precision and execution efficiency.
Propose based on translation invariant and the distributed DTW human body behavior intention assessment side of the constant human body behavior characteristics of scaling
Method, to realize accurate human body specific behavior identification.
Accompanying drawing explanation
The distributed schematic diagram based on dynamic 3 D RGBD sensor network of Fig. 1 present invention;
The Distributed Three-dimensional sensor network of Fig. 2 present invention multiple-model estimator flow chart to human joint points;
Distributed many human body targets articulare track algorithm flow process of Fig. 3 present invention;
Fig. 4 human joint points schematic diagram (15 articulares).
Detailed description of the invention:
The present invention is described in detail below in conjunction with the accompanying drawings:
As it is shown in figure 1, human joint points distributed information Uniform estimates method based on Interactive Multiple-Model, by building
Dynamic distributed RGBD sensor network, it is achieved the distributed treatment to data and the distributed fusion to information, nothing in network
Centralized information processes and fusion center, and sensor node only exchanges with adjacent node information, by limited number of time concordance iteration,
Realize network interior consistent to the estimation of perception dbjective state.
Sensor network realizes the transmission of information by radio communication.Each sensor is connected to native processor, permissible
It is microcomputer or ARM development board.After native processor is to information processing, carry out network data friendship by wireless with adjacent node
Change.Dynamically refer to that network is made up of sensor and the position-movable sensor of position static state.Wherein, position be moved through by
Sensor is placed in mobile robot realization.The Distributed Calculation and the fusion that are distributed across information realize.
As in figure 2 it is shown, skeleton joint point position initialization: by articulare depth information learning training in advance, it is achieved
Detection to every frame human joint points, it is possible to utilize existing instrument OPENNI NITE or Microsoft SDK extracting directly articulare.For
Remove invalid articulare, set up human joint points motion model physical constraint, reject be unsatisfactory for the human joint points anglec of rotation and
The human joint points of length constraint.The relevant parameter of physical constraint model, such as length between human elbow and shoulder joint, can depend on
According to detection data adaptive upgrading.
Wherein, articulare depth information is to obtain scene by Microsoft's Kinect kit or OpenNI drive software of increasing income
RGB image and depth image.
The purpose of learning training is to build articulare feature database in advance, thus realizes the joint in image to be detected
The classification of point and identification.
RGBD sensor provides scene color and depth image.Articulare detection module extracts human synovial from image
Point.
Human joint points motion model physical constraint correlation technique content refers to paper Model-Based
Reinforcement of Kinect Depth Data for Human Motion Capture Applications。
Local RGBD sensor is to articulare estimation: build motion model and observation model, the base of human joint points
In Bayesian filter, it is achieved the effective estimation to articulare state (position, speed and acceleration).The motion of human joint points
Exist static, at the uniform velocity, accelerate multi-model attribute alternately, single movement model is not enough to describe articulare behavioral characteristics,
Therefore design Bayesian Estimation method based on Interactive Multiple-Model, the time-varying state of human joint points is effectively followed the tracks of and estimates
Meter.To linear model, available linear information wave filter is estimated, and for nonlinear model, available Extended information filter device
Estimate with based on nonlinear filters such as centered difference information filters.
Observation model refers to the relationship model between filter system state and sensor observation.Here, system mode
Refer to articulare three-dimensional position, speed and acceleration, and sensor observation is articulare three-dimensional position.
The specific algorithm effectively estimated refers to paper Central Difference Information Filter
with Interacting Multiple Model for Robust Maneuvering Object Tracking。
Between RGBD sensor, the consistency on messaging to target joint point is estimated: the letter that definition human synovial dotted state is corresponding
Ceasing vector, information matrix and information contribution thereof and the model probability exchange capacity as consistency on messaging algorithm, each sensor will
The information contribution of articulare information vector, information matrix and correspondence thereof that self estimates, model probability are sent to adjacent communication
Sensor node, and accept the information of ambient sensors, utilize consistency on messaging algorithm, merge the estimation knot of ambient sensors
Really, subsequent iteration is for several times, it is achieved algorithm and the convergence of estimated result.Specifically comprise the following steps that
The first step is that systematic parameter initializes, and wherein the initial position of human joint points can be by OPENNI directly from RGBD
The depth image detection of camera draws, the variance of position can determine according to the confidence level of the articulare identification that OPENNI returns,
And the motion model transition probability in joint is set according to joint motions feature.
Second step is that multi-model is mutual, i.e. general according to the mixing between model probability and Model transition probability computation model
Rate, then hybrid mean value and the mixing variance of each model is drawn according to mixing probability calculation.
3rd step is information filter, with hybrid mean value and mixing variance for input, calculate its information vector and information matrix,
Linear processes motion model is respectively adopted linear information wave filter and centered difference information filter is estimated, according to currently
Articulare position upgrading wave filter articulare state (information vector and information matrix) of depth image detection and model probability.
4th step is information fusion based on distributed information consistency algorithm, exchanges articulare and estimate between the most each sensor
Meter information, including articulare status information vector, information matrix and model probability, realizes each by consistency algorithm weighted iteration
The concordance of sensor estimated state, as sensor node i and sensor node j is adjacent communication node, between the two
Metroplis weight is εi,j, then the r time its information vector of iterationInformation matrixAnd model probabilityCan by its institute
The corresponding information weighted sum having adjacent node j calculates:
5th step is each sensor node mixing based on model probability weighted sum output, i.e. utilizes model probability to each
The estimated result of model is weighted summation, as the estimated result of each sensor information processing system current time.
As it is shown on figure 3, joint probabilistic data association method based on coloured image and depth image multiple features, the method is for many
Target following, including:
Systematic parameter initializes;
Multi-model is mutual;
JPDA based on linear information wave filter and JPDA based on centered difference information filter;
Send local information to approaching sensor node;
Receive approaching sensor nodal information;
Data association based on mahalanobis distance;
Distributed information consistency algorithm realizes multi-model result and merges.
Distributed human body behavior intention assessment includes: builds and has translation invariant and scale constant organization of human body feature,
Set up human body behavior intent data library template, in conjunction with dynamic time warping algorithm, it is achieved the calculating to specific behavior matching probability.
Visible, distributed human body behavior intention assessment, comprise two parts: be first the fusion of matching probability, i.e. merge net
Other sensors matching probability to the behavior in network, improves local recognition accuracy;Next to that the concordance of recognition result is estimated
Meter, it is ensured that recognition result globally consistent.
Detailed process is as follows: in view of the joint vector that the individual variation of build, definable translation invariant and scaling are constant
Angle and upper limb partial joint vector mould ratio are as human body behavior characteristics.OPENNI can detect 15 articulares of human body, as
Shown in Fig. 4, through calculating, joint vector angle information can be obtained, specifically include upper limb part 10 groups, lower extremities 4 groups, centre
6 groups, coupling part.Wherein, the calculating of joint vector and angle thereof refers to paper: apply the Human bodys' response side of Kinect
Method research designs with system.
It addition, be the detail section information further describing upper limb behavior, the mould between definition upper limb partial joint vector
Ratio is as feature:
WhereinIt is the trunk center vectorial mould to head,WithBe respectively head point to right-hand man vectorial mould,
WithIt is the trunk center vectorial mould to right-hand man respectively.Joint vector angle and vector mould ratio collectively constitute 24 dimension human body row
It is characterized, meets translation invariant and scale constant condition.
After obtaining present frame human body behavior characteristics, it is added in human body behavior characteristics time series, utilizes dynamically
Current observation behavior characteristics sequence is learnt in data base by time wrapping algorithm (Dynamic Time Warping, DTW)
Specific behavior sequence template mate, calculate optimal matching similarity based on card side's distance between the two.Similar to this
Degree takes inverse, obtains matching probability after normalization.For merging ambient sensors recognition result, exchange between adjacent connection sensor
Matching probability information, if i-th sensor is by matching probability corresponding for k moment action pattern αWith adjacent sensors node
The matching probability that j calculatesCarry out data exchange, and according to the transition probability matrix M between action pattern, determine action mould
The probit of formula α
Wherein η (k) is normalization factor, and M (β, α) is the transition probability value between action pattern β and action pattern α.
Finally, using the action pattern probit of each sensor as consistency information amount and the another number of times of adjacent sensors
According to exchange, through concordance interative computation, the recognition result of final adjacent sensors node is reached an agreement, this just construct based on
The distributed behavior intention assessment algorithm of DTW.
If it is finalMore than certain threshold value, then can assert that Current observation subject performance sequence is mated with action pattern α
Success.
Although the detailed description of the invention of the present invention is described by the above-mentioned accompanying drawing that combines, but not the present invention is protected model
The restriction enclosed, one of ordinary skill in the art should be understood that on the basis of technical scheme, and those skilled in the art are not
Need to pay various amendments or deformation that creative work can make still within protection scope of the present invention.
Claims (10)
1. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics, is characterized in that, comprise the following steps:
Obtain present frame human body behavior characteristics and be added in human body behavior characteristics time series;
Utilize the dynamic time warping algorithm DTW spy to having learnt in Current observation human body behavior characteristics sequence and database template
Determine behavior sequence to mate, calculate optimal matching similarity based on card side's distance between the two;
This similarity is taken inverse, after normalization, obtains matching probability;
The action pattern probit of each sensor is exchanged as consistency information amount with adjacent sensors data again, warp
Crossing concordance interative computation, the recognition result of final adjacent sensors node is reached an agreement.
2. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 1, its feature
It is, when calculating matching probability, for merging ambient sensors recognition result, between adjacent connection sensor, to exchange matching probability letter
Breath.
3. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 2, its feature
Be, between adjacent connection sensor exchange matching probability information mode particularly as follows:
I-th sensor is by matching probability corresponding for k moment action pattern αWith mating that adjacent sensors node j calculates
ProbabilityCarry out data exchange, and according to the transition probability matrix M between action pattern, determine the probit of action pattern α
Wherein η (k) is normalization factor, and M (β, α) is the transition probability value between action pattern β and action pattern α.
4. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 3, its feature
It is, after exchanging matching probability information between adjacent connection sensor, if finalMore than certain threshold value, then assert current
The match is successful with action pattern α for observed object action sequence.
5. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 1, its feature
It is that distributed DTW human body behavior intension recognizing method based on human body behavior characteristics is obtaining present frame human body behavior characteristics
Also include before step building and there is translation invariant and scale the step of constant organization of human body feature.
6. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 5, its feature
It is that, when structure has translation invariant and scales constant organization of human body feature, it is contemplated that the individual variation of build, definition meets
Joint vector angle and the upper limb partial joint vector mould ratio of the condition that translation invariant is constant with scaling are special as human body behavior
Levying, joint vector angle and vector mould ratio collectively constitute 24 dimension human body behavior characteristicss.
7. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 6, its feature
It is, upper limb partial joint vector mould ratio:
WhereinIt is the trunk center vectorial mould to head,WithBe respectively head point to right-hand man vectorial mould,With
It is the trunk center vectorial mould to right-hand man respectively.
8. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 6, its feature
It is, the acquisition of the joint vector angle information that translation invariant is constant with scaling: by 15 articulares of OPENNI detection human body
Distributed information, through calculating, can obtain joint vector angle information, specifically include upper limb part 10 groups, lower extremities 4 groups,
Middle interconnecting piece divides 6 groups.
9. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 8, its feature
It is that in the application, the articulare distributed information to the human body obtained carries out conforming estimation, specifically includes:
Skeleton joint point position initialization;
Local sensor is to articulare estimation: build motion model and the observation model of human joint points, it is achieved to joint
Effective estimation of dotted state;
Between sensor, the consistency on messaging of target joint point is estimated: information vector corresponding to definition human synovial dotted state,
Information matrix and information contribution and model probability thereof are as the exchange capacity of consistency on messaging algorithm;
Each sensor is by the information contribution of articulare information vector, information matrix and correspondence thereof self estimated, model probability
It is sent to adjacent communication sensor node, and accepts the information of ambient sensors, utilize consistency on messaging algorithm, merge around
The estimated result of sensor, subsequent iteration is for several times, it is achieved algorithm and the convergence of estimated result.
10. distributed DTW human body behavior intension recognizing method based on human body behavior characteristics as claimed in claim 9, it is special
Levy and be, carry out based on coloured image and depth map after the application carries out conforming estimation to human joint points distributed information
As the joint probabilistic data association of multiple features, comprise the following steps:
Associating data probabilistic correlation algorithm is used to realize local sensor node to carrying out the between tracking target and target observation
Data association;
Hungary Algorithm based on mahalanobis distance realizes between sensor node the second time data association following the tracks of target;
Exchange articulare estimated information between each sensor, realize each sensor estimated state by consistency algorithm weighted iteration
Concordance;
Wherein, when first time data association, target observation Candidate Set Regulation mechanism based on multiple features, utilize articulare position
Observation information z, coloured image gradient orientation histogram feature hcWith depth image gradient orientation histogram feature hdBuild three thresholds
Value thresholding (γz,γc,γd) to limit observation collection size.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610436818.7A CN106127125B (en) | 2016-06-16 | 2016-06-16 | Distributed DTW Human bodys' response method based on human body behavioural characteristic |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610436818.7A CN106127125B (en) | 2016-06-16 | 2016-06-16 | Distributed DTW Human bodys' response method based on human body behavioural characteristic |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106127125A true CN106127125A (en) | 2016-11-16 |
CN106127125B CN106127125B (en) | 2019-04-02 |
Family
ID=57470799
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610436818.7A Active CN106127125B (en) | 2016-06-16 | 2016-06-16 | Distributed DTW Human bodys' response method based on human body behavioural characteristic |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106127125B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107590476A (en) * | 2017-09-22 | 2018-01-16 | 郑州云海信息技术有限公司 | A kind of comparison method of action, device and computer-readable storage medium |
CN109362066A (en) * | 2018-11-01 | 2019-02-19 | 山东大学 | A kind of real-time Activity recognition system and its working method based on low-power consumption wide area network and capsule network |
CN109460702A (en) * | 2018-09-14 | 2019-03-12 | 华南理工大学 | Passenger's abnormal behaviour recognition methods based on human skeleton sequence |
CN110060182A (en) * | 2019-03-16 | 2019-07-26 | 深圳壹账通智能科技有限公司 | Tourist image design method for tracing, device, computer equipment and storage medium |
CN110163113A (en) * | 2019-04-25 | 2019-08-23 | 上海师范大学 | A kind of human body behavior similarity calculating method and device |
CN110210454A (en) * | 2019-06-17 | 2019-09-06 | 合肥工业大学 | A kind of human action pre-judging method based on data fusion |
CN112001335A (en) * | 2020-08-27 | 2020-11-27 | 武汉科技大学 | Interface system for evaluating unsafe behavior risk of subway passengers |
CN113353083A (en) * | 2021-08-10 | 2021-09-07 | 所托(杭州)汽车智能设备有限公司 | Vehicle behavior recognition method |
CN113591712A (en) * | 2021-07-30 | 2021-11-02 | 天津市府易科技股份有限公司 | AI identification-based user abnormal behavior identification method and device |
CN115019395A (en) * | 2022-06-10 | 2022-09-06 | 杭州电子科技大学 | Group action consistency detection method and system based on stacked hourglass network |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0686931A2 (en) * | 1994-06-06 | 1995-12-13 | Xerox Corporation | Searching and matching unrecognized handwriting |
US20050157908A1 (en) * | 2004-01-15 | 2005-07-21 | Canon Kabushiki Kaisha | Action recognition apparatus and method, moving-object recognition apparatus and method, device control apparatus and method, and program |
CN101477619A (en) * | 2008-12-26 | 2009-07-08 | 北京航空航天大学 | Movement data gesture classification process based on DTW curve |
CN104504731A (en) * | 2014-12-19 | 2015-04-08 | 西安理工大学 | Human motion synthesis method based on motion diagram |
CN104598889A (en) * | 2015-01-30 | 2015-05-06 | 北京信息科技大学 | Human action recognition method and device |
-
2016
- 2016-06-16 CN CN201610436818.7A patent/CN106127125B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0686931A2 (en) * | 1994-06-06 | 1995-12-13 | Xerox Corporation | Searching and matching unrecognized handwriting |
US20050157908A1 (en) * | 2004-01-15 | 2005-07-21 | Canon Kabushiki Kaisha | Action recognition apparatus and method, moving-object recognition apparatus and method, device control apparatus and method, and program |
CN101477619A (en) * | 2008-12-26 | 2009-07-08 | 北京航空航天大学 | Movement data gesture classification process based on DTW curve |
CN104504731A (en) * | 2014-12-19 | 2015-04-08 | 西安理工大学 | Human motion synthesis method based on motion diagram |
CN104598889A (en) * | 2015-01-30 | 2015-05-06 | 北京信息科技大学 | Human action recognition method and device |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107590476B (en) * | 2017-09-22 | 2020-10-23 | 苏州浪潮智能科技有限公司 | Action comparison method and device and computer storage medium |
CN107590476A (en) * | 2017-09-22 | 2018-01-16 | 郑州云海信息技术有限公司 | A kind of comparison method of action, device and computer-readable storage medium |
CN109460702A (en) * | 2018-09-14 | 2019-03-12 | 华南理工大学 | Passenger's abnormal behaviour recognition methods based on human skeleton sequence |
CN109460702B (en) * | 2018-09-14 | 2022-02-15 | 华南理工大学 | Passenger abnormal behavior identification method based on human body skeleton sequence |
CN109362066B (en) * | 2018-11-01 | 2021-06-25 | 山东大学 | Real-time behavior recognition system based on low-power-consumption wide-area Internet of things and capsule network and working method thereof |
CN109362066A (en) * | 2018-11-01 | 2019-02-19 | 山东大学 | A kind of real-time Activity recognition system and its working method based on low-power consumption wide area network and capsule network |
CN110060182A (en) * | 2019-03-16 | 2019-07-26 | 深圳壹账通智能科技有限公司 | Tourist image design method for tracing, device, computer equipment and storage medium |
CN110163113A (en) * | 2019-04-25 | 2019-08-23 | 上海师范大学 | A kind of human body behavior similarity calculating method and device |
CN110163113B (en) * | 2019-04-25 | 2023-04-07 | 上海师范大学 | Human behavior similarity calculation method and device |
CN110210454B (en) * | 2019-06-17 | 2020-12-29 | 合肥工业大学 | Human body action pre-judging method based on data fusion |
CN110210454A (en) * | 2019-06-17 | 2019-09-06 | 合肥工业大学 | A kind of human action pre-judging method based on data fusion |
CN112001335A (en) * | 2020-08-27 | 2020-11-27 | 武汉科技大学 | Interface system for evaluating unsafe behavior risk of subway passengers |
CN113591712A (en) * | 2021-07-30 | 2021-11-02 | 天津市府易科技股份有限公司 | AI identification-based user abnormal behavior identification method and device |
CN113353083A (en) * | 2021-08-10 | 2021-09-07 | 所托(杭州)汽车智能设备有限公司 | Vehicle behavior recognition method |
CN113353083B (en) * | 2021-08-10 | 2021-10-29 | 所托(杭州)汽车智能设备有限公司 | Vehicle behavior recognition method |
CN115019395A (en) * | 2022-06-10 | 2022-09-06 | 杭州电子科技大学 | Group action consistency detection method and system based on stacked hourglass network |
CN115019395B (en) * | 2022-06-10 | 2022-12-06 | 杭州电子科技大学 | Group action consistency detection method and system based on stacked hourglass network |
Also Published As
Publication number | Publication date |
---|---|
CN106127125B (en) | 2019-04-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106127125A (en) | Distributed DTW human body behavior intension recognizing method based on human body behavior characteristics | |
Kim et al. | Simultaneous gesture segmentation and recognition based on forward spotting accumulative HMMs | |
JP4625129B2 (en) | Monocular tracking of 3D human motion using coordinated mixed factor analysis | |
CN104680559B (en) | The indoor pedestrian tracting method of various visual angles based on motor behavior pattern | |
CN105512621A (en) | Kinect-based badminton motion guidance system | |
CN111161315A (en) | Multi-target tracking method and system based on graph neural network | |
CN105279769A (en) | Hierarchical particle filtering tracking method combined with multiple features | |
CN105869181A (en) | Body joint distributed information consistency estimation method based on interacting multiple models | |
CN103985143A (en) | Discriminative online target tracking method based on videos in dictionary learning | |
CN106127119B (en) | Joint probabilistic data association method based on color image and depth image multiple features | |
Palconit et al. | Three-dimensional stereo vision tracking of multiple free-swimming fish for low frame rate video | |
Yu et al. | Accurate and robust visual localization system in large-scale appearance-changing environments | |
Ding et al. | Simultaneous body part and motion identification for human-following robots | |
Siddiq et al. | Integrated Internet of Things (IoT) technology device on smart home system with human posture recognition using kNN method | |
Batool et al. | Telemonitoring of daily activities based on multi-sensors data fusion | |
CN106127806B (en) | RGB-D video target tracking methods based on depth Boltzmann machine cross-module formula feature learning | |
CN115018999A (en) | Multi-robot-cooperation dense point cloud map construction method and device | |
CN106096565A (en) | Mobile robot based on sensing network and the task cooperative method of static sensor | |
CN103839280A (en) | Method for tracking human body posture based on visual information | |
Weinrich et al. | Appearance-based 3D upper-body pose estimation and person re-identification on mobile robots | |
Moon et al. | Holistic planimetric prediction to local volumetric prediction for 3d human pose estimation | |
CN113327269A (en) | Unmarked cervical vertebra movement detection method | |
Huang et al. | View-independent behavior analysis | |
Raza et al. | An integrative approach to robust hand detection using CPM-YOLOv3 and RGBD camera in real time | |
CN112507940A (en) | Skeleton action recognition method based on difference guidance representation learning network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |