CN106570482A - Method and device for identifying body motion - Google Patents
Method and device for identifying body motion Download PDFInfo
- Publication number
- CN106570482A CN106570482A CN201610973939.5A CN201610973939A CN106570482A CN 106570482 A CN106570482 A CN 106570482A CN 201610973939 A CN201610973939 A CN 201610973939A CN 106570482 A CN106570482 A CN 106570482A
- Authority
- CN
- China
- Prior art keywords
- acceleration signal
- feature
- obtains
- eigenvector
- image sequence
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention, which is applicable to the field of pattern recognition technology, provides a method and an apparatus for identifying a body motion. The method comprises: obtaining a depth image sequence and an acceleration signal of a body motion; extracting a first feature vector from the depth image sequence and extracting a second feature vector from the acceleration signal; obtaining a feature descriptor by combining the first feature vector and the second feature vector; and according to the feature descriptor, identifying the body motion. According to the method and device provided by the invention, with one or two fixed depth cameras and a plurality of acceleration sensors worn at key joint parts of the human body, a problem of low body motion identification accuracy due to shielding or error accumulation according to the existing body motion identification method can be solved and thus the accuracy of body motion identification can be improved.
Description
Technical field
The invention belongs to mode identification technology, more particularly to a kind of human motion recognition method and device.
Background technology
With the continuous development of sensor technology, the communication technology and data analysis technique, ambient intelligence is answered many
Achieved successfully with scene, and human action identification is exactly the key issue for realizing ambient intelligence.Human action identification is related to
To from automatic detection, analysis human action in the information collected by different modalities sensor.
Existing human motion recognition method mainly includes the human motion recognition method based on computer vision and is based on
Two kinds of the human motion recognition method of inertial sensor.Based on the human motion recognition method of computer vision, by computer
The original image or image sequence data of camera acquisition are processed and analyzed, is learnt and understood the wherein action of people and row
For.Based on the human motion recognition method of inertial sensor, by being fixed on the inertial sensor of human body privileged site people is gathered
The action message of body, and computer is sent to by wireless transport module, so data are pre-processed, feature extraction and choosing
Select, the classification of motion.
And at present the human action recognizer of most of view-based access control models is all completed under given conditions, such as high score
Resolution, fixed viewpoint, fixed background, fixed video camera and unobstructed etc., substantially also not can solve the problem that sternly without effective ways
Action recognition problem block again under.And for the human motion recognition method based on inertial sensor, sensor can be due to peace
The insecure of dress causes error, and error to accumulate during human motion.
The content of the invention
In view of this, a kind of human motion recognition method and device are embodiments provided, to solve prior art
Present in block or error accumulation caused by human action identification the relatively low problem of accuracy rate.
A kind of first aspect of the embodiment of the present invention, there is provided human motion recognition method, including：
Obtain the range image sequence and acceleration signal of human action；The range image sequence and the acceleration are believed
Number synchronization；
Extract first eigenvector from the range image sequence, from the acceleration signal extract second feature to
Amount；
With reference to the first eigenvector and second feature vector, Feature Descriptor is drawn；
According to the Feature Descriptor, human action is identified.
A kind of second aspect of the embodiment of the present invention, there is provided human action identifying device, including：
Acquisition module, for obtaining the range image sequence and acceleration signal of human action；The range image sequence
With the acceleration signal synchronization；
Extraction module, for extracting first eigenvector from the range image sequence, from the acceleration signal
Extract second feature vector；
Feature interpretation submodule, for reference to the first eigenvector and second feature vector, showing that feature is retouched
State son；
Identification module, for according to the Feature Descriptor, being identified to human action.
The embodiment of the present invention at least has the advantages that relative to prior art：The embodiment of the present invention gathers first people
The range image sequence and acceleration signal of body action, then extracts first eigenvector from the range image sequence, with
And second feature vector is extracted from the acceleration signal, finally by the first eigenvector and second feature vector
With reference to drawing Feature Descriptor, and according to the Feature Descriptor, human action is identified, it is only necessary to one to two positions
Fixed depth camera and several acceleration transducers for being worn on human body major joint position, it is possible to solve existing human body
Action identification method exist to block or error accumulation caused by the low problem of human action recognition accuracy, improve human body and move
The accuracy rate that work is recognized.
Description of the drawings
Technical scheme in order to be illustrated more clearly that the embodiment of the present invention, below will be to embodiment or description of the prior art
Needed for the accompanying drawing to be used be briefly described, it should be apparent that, drawings in the following description be only the present invention some
Embodiment, for those of ordinary skill in the art, without having to pay creative labor, can be with according to these
Accompanying drawing obtains other accompanying drawings.
Fig. 1 is the human motion recognition method flow chart that the embodiment of the present invention one is provided；
Fig. 2 is the flow process that first eigenvector is extracted from the range image sequence that the embodiment of the present invention two is provided
Figure；
Fig. 3 is the flow chart that second feature vector is extracted from the acceleration signal that the embodiment of the present invention three is provided；
Fig. 4 is the analogous diagram of the characteristic vector pickup of the 3-axis acceleration signal that the embodiment of the present invention three is provided；
Fig. 5 is the structured flowchart of the human action identifying device that the embodiment of the present invention four is provided；
Fig. 6 is the structured flowchart of the extraction module that the embodiment of the present invention four is provided；
Fig. 7 is the structured flowchart of the extraction module that the embodiment of the present invention four is provided.
Specific embodiment
In below describing, in order to illustrate rather than in order to limit, it is proposed that the such as tool of particular system structure, technology etc
Body details, thoroughly to understand the embodiment of the present invention.However, it will be clear to one skilled in the art that concrete without these
The present invention can also be realized in the other embodiments of details.In other situations, omit to well-known system, device, electricity
Road and the detailed description of method, in case unnecessary details hinders description of the invention.
In order to illustrate technical solutions according to the invention, illustrate below by specific embodiment.
Embodiment one：
Fig. 1 shows that the human motion recognition method that the embodiment of the present invention one is provided realizes flow process, and details are as follows：
In step S101, the range image sequence and acceleration signal of human action, the range image sequence are obtained
With the acceleration signal synchronization.
In the present embodiment, the range image sequence of human action can be obtained by depth camera.Can be by accelerating
Degree sensor obtains the acceleration signal of human action.For example, the depth camera that can be fixed by one to two positions is obtained
Take the range image sequence of human action.Again for example, at least one acceleration for being worn on human body major joint position can be passed through
Degree sensor obtains the acceleration signal of human action.In view of the accuracy of the acceleration signal of the human action for getting,
The acceleration signal that multiple acceleration transducers for being worn on human body major joint position obtain human action can be passed through.Its
In, the acceleration signal of the human action for getting can characterize acceleration situation of the human body on multiple axial directions or direction, i.e.,
Motion conditions of the human body on multiple axial directions or direction.
It should be noted that the range image sequence got in this step and acceleration signal are after elapsed time synchronization
Range image sequence and acceleration signal.So, range image sequence and acceleration signal can be corresponded in time, with reference to
Reflect the action of human body together such that it is able to improve the accuracy of identification to human action.
Further, since the space that human body is located is three dimensions, therefore corresponding acceleration signal can accelerate for three axles
Degree signal.Wherein, corresponding three axial directions of 3-axis acceleration signal are vertical each other.
In step s 102, first eigenvector is extracted from the range image sequence, from the acceleration signal
Extract second feature vector.
Specifically, line translation can be entered to the range image sequence, obtains corresponding Depth Motion mapping and to described
Depth Motion mapping is extracted characteristic vector and obtains first eigenvector.Simultaneously time slip-window side is adopted to the acceleration signal
Method extracts second feature vector.Implement process to be described in embodiment two and embodiment three, here is no longer excessively retouched
State.
Wherein, first eigenvector includes but is not limited to HOG (Histogram of Oriented Gradient, direction
Histogram of gradients) feature.Second feature vector including but not limited to FFT (Fast Fourier Transformation, quickly
Fourier transformation) coefficient.
In step s 103, with reference to the first eigenvector and second feature vector, Feature Descriptor is drawn.
Specifically, can be by the first eigenvector of two different modalities in step S102 and second feature vector splicing
Into a characteristic vector, then dropped using PCA (Principal Component Analysis, principal component analysis) method
Dimension, finally gives the Feature Descriptor.
It should be noted that being not limited to carry out dimensionality reduction using PCA methods in the present embodiment, it would however also be possible to employ other dimensionality reduction sides
Method, and finally give the Feature Descriptor.
In step S104, according to the Feature Descriptor, human action is identified.
In the present embodiment, Linear SVM (support vector machine, SVMs) method can be adopted, be led to
The Feature Descriptor is crossed, classification, but not limited to this are identified to human action.Those skilled in the art can be according to need
Will, using additive method, by the Feature Descriptor, human action is identified.
Embodiment two：、
Referring to Fig. 2, in the present embodiment, first eigenvector bag is extracted from the range image sequence in step S102
Include：
Step S201, obtains the i-th frame depth image in the range image sequence, and by the i-th frame depth image
Project on each orthogonal plane in the same coordinate system, obtain the corresponding with the orthogonal plane of the i-th frame depth image
Each visual angle projected image.
Wherein, i is the integer more than or equal to 1, and i is less than or equal to the frame number of depth image in range image sequence.
Illustrated as a example by there are three orthogonal planes in the same coordinate system below, but be not limited to this.In the present embodiment, institute is obtained
After stating any i-th frame depth image in range image sequence, the i-th frame depth image can be projected to cartesian coordinate
On three orthogonal planes of system, facing angular projection image, side-looking angular projection image and bowing for the i-th frame depth image is obtained
Visual angle projected image.
Step S202, calculate i+1 frame depth image each visual angle projected image and the i-th frame depth image each regard
The absolute difference of the projected image at angle.
In the present embodiment, calculate i+1 frame depth image faces angular projection image, side-looking angular projection image and vertical view
The absolute difference of angular projection image visual angle projected image corresponding with the i-th frame depth image.That is, i+1 frame depth image is calculated
The absolute difference for facing angular projection image of angular projection image and the i-th frame depth image is faced, i+1 frame depth image is calculated
The absolute difference of the side-looking angular projection image of side-looking angular projection image and the i-th frame depth image, and calculate i+1 frame depth map
The absolute difference of the depression angle projected image of the depression angle projected image of picture and the i-th frame depth image.
Step S203, is superimposed the corresponding absolute difference of each frame depth image, obtains the range image sequence corresponding
Depth Motion maps.
In the present embodiment, each frame depth image in the range image sequence is traveled through, be superimposed each frame depth image
Corresponding absolute difference, obtains the corresponding Depth Motion mapping of the range image sequence.
Step S204, to the Depth Motion mapping feature extraction is carried out, and obtains the first eigenvector.
Preferably, the Depth Motion mapping to obtaining can extract feature from each visual angle respectively, obtain the depth fortune
The dynamic characteristic vector for mapping each visual angle, the subcharacter vector for connecting each visual angle obtains described first eigenvector.
For example, to the Depth Motion mapping that obtains respectively from facing, side-looking and overlook three visual angles and extract features, obtain institute
The characteristic vector that Depth Motion maps three visual angles is stated, the characteristic vector for connecting three visual angles obtains the first eigenvector.
Assume a given range image sequence { I comprising N frames1, I2..., IN}.It is any in for N frame depth images
One frame, for the ease of description, is designated as here the i-th frame, and the i-th frame depth image is projected to three under cartesian coordinate system
Orthogonal plane, obtains the projected image at three visual anglesThe v ∈ { f, s, t }, wherein, f represents positive visual angle, and s represents side-looking
Angle, t represents depression angle.The corresponding projected image of each frame depth image in the N frames range image sequence is obtainedAfterwards,
According to formulaIt is calculated three visual angle components of Depth Motion mapping.For each visual angle component,
Advantage distillation HOG features obtain characteristic vector HOG at each visual anglev, connect three visual angles characteristic vector obtain [HOGf,
HOGs, HOGt] it is the characteristic vector of the Depth Motion mapping, i.e. first eigenvector.
Embodiment three：
Referring to Fig. 3, in the present embodiment, the second feature vector that extracts from the acceleration signal includes：
Step S301, is split using time slip-window to the acceleration signal, obtains at least one comprising acceleration
The data window of degree signal segment.
In the present embodiment, Duplication can be adopted the acceleration signal is carried out for the time slip-window of preset ratio point
Cut, and it is long using the optimal window of cross validation acquisition, obtain at least one data window comprising acceleration signal fragment.Wherein, in advance
If ratio can be numerical value between 40% to 60% with value.For example, preset ratio can be with value as 40%, 50%, 60% etc..
Preferably, in the present embodiment, adopt Duplication the acceleration signal is carried out for 50% time slip-window point
Cut, and it is long using the optimal window of cross validation acquisition, obtain one group of data window comprising 3-axis acceleration signal segment.
Step S302, in each described data window, extracts the feature of the acceleration signal fragment, obtains each
The characteristic vector of the data window.
In the present embodiment, process can be filtered to 3-axis acceleration signal segment in each described data window
And feature is extracted, and remove DC component part therein, obtain the characteristic vector of three axial directions of each data window.
Step S303, according to the characteristic vector of all data windows, draws the second feature vector.
In the present embodiment, the characteristic vector of the correspondence axial direction that can be connected in all data windows obtains three axles
To the characteristic vector of acceleration signal, connect three axial acceleration signals characteristic vector be obtained final feature to
Amount.It is the analogous diagram of the characteristic vector pickup of 3-axis acceleration signal referring to Fig. 4.
Above-mentioned human motion recognition method, gathers first the range image sequence and acceleration signal of human action, then
First eigenvector is extracted from the range image sequence, and second feature vector is extracted from the acceleration signal,
Finally the first eigenvector and second feature vector are combined and draw Feature Descriptor, and according to the feature interpretation
Son, is identified to human action.Above-mentioned human motion recognition method is different in two kinds of range image sequence and acceleration signal
Carry out Fusion Features under modal data to recognize human action, it is only necessary to the depth camera that one to two positions are fixed with it is several
Be worn on the acceleration transducer at human body major joint position, it is possible to solve that existing human motion recognition method is present to hiding
The low problem of human action recognition accuracy caused by gear or error accumulation, improves the accuracy rate of human action identification.
It should be understood that the size of the sequence number of each step is not meant to the priority of execution sequence, each process in above-described embodiment
Execution sequence should be determined with its function and internal logic, and any limit should not be constituted to the implementation process of the embodiment of the present invention
It is fixed.
Example IV：
Corresponding to the human motion recognition method described in foregoing embodiments, Fig. 5 shows people provided in an embodiment of the present invention
The structured flowchart of body action recognition device.For convenience of description, illustrate only part related to the present embodiment.
With reference to Fig. 5, the device includes：Acquisition module 401, extraction module 402, feature interpretation submodule 403 and identification mould
Block 404.
Wherein, acquisition module 401, for obtaining the range image sequence and acceleration signal of human action；The depth
Image sequence and the acceleration signal synchronization.
Extraction module 402, for extracting first eigenvector from the range image sequence, from the acceleration signal
Middle extraction second feature vector.
Feature interpretation submodule 403, for the first eigenvector and second feature vector to be combined, draws spy
Levy description.
Identification module 404, for according to the Feature Descriptor, being identified to human action.
Used as a kind of embodiment, referring to Fig. 6, the extraction module 402 can include：Processing unit 501, calculating are single
Unit 502, map unit 503 and first eigenvector acquiring unit 504.
Wherein, processing unit 501, for obtaining the range image sequence in the i-th frame depth image, and by described
I frame depth images are projected on each orthogonal plane in the same coordinate system, obtain the i-th frame depth image with it is described just
Hand over the projected image at corresponding each visual angle of plane.Wherein, i is the integer more than or equal to 1, and i is less than or equal to depth map
As the frame number of depth image in sequence.
Computing unit 502, for calculating the projected image and the i-th frame depth map at each visual angle of i+1 frame depth image
As the absolute difference of the projected image at each visual angle.
Map unit 503, for traveling through the range image sequence in each frame depth image, be superimposed each frame depth
The corresponding absolute difference of image, obtains the corresponding Depth Motion mapping of the range image sequence.
First eigenvector acquiring unit 504, for carrying out feature extraction to Depth Motion mapping, obtains described the
One characteristic vector.
Preferably, the first eigenvector acquiring unit specifically for：To the Depth Motion mapping that obtains respectively from each
Feature is extracted at individual visual angle, obtains the characteristic vector that the Depth Motion maps each visual angle, connect the subcharacter at each visual angle to
Measure described first eigenvector.
Used as a kind of embodiment, referring to Fig. 7, the extraction module 402 can include：Cutting unit 601, feature is carried
Take unit 602 and second feature vector acquiring unit 603.
Wherein, cutting unit 601, for being split to the acceleration signal using time slip-window, are obtained at least
One data window comprising acceleration signal fragment.
Feature extraction unit 602, in each described data window, extracting the spy of the acceleration signal fragment
Levy, obtain the characteristic vector of each data window.
Second feature vector acquiring unit 603, for according to the characteristic vector of all data windows, drawing described second
Characteristic vector.
Preferably, the cutting unit 601 specifically for：Adopt Duplication for preset ratio time slip-window to described
Acceleration signal is split, and long using the optimal window of cross validation acquisition, obtains at least one comprising acceleration signal fragment
Data window.
Above-mentioned human action identifying device, gathers first the range image sequence and acceleration signal of human action, then
First eigenvector is extracted from the range image sequence, and second feature vector is extracted from the acceleration signal,
Finally the first eigenvector and second feature vector are combined and draw Feature Descriptor, and according to the feature interpretation
Son, is identified to human action.Above-mentioned human action identifying device is different in two kinds of range image sequence and acceleration signal
Carry out Fusion Features under modal data to recognize human action, it is only necessary to the depth camera that one to two positions are fixed with it is several
Be worn on the acceleration transducer at human body major joint position, it is possible to solve that existing human motion recognition method is present to hiding
The low problem of human action recognition accuracy caused by gear or error accumulation, improves the accuracy rate of human action identification.
Those skilled in the art can be understood that, for convenience of description and succinctly, only with above-mentioned each work(
Energy unit, the division of module are illustrated, and in practical application, as desired can distribute above-mentioned functions by different
Functional unit, module are completed, will the internal structure of described device be divided into different functional unit or module, to complete the above
The all or part of function of description.Each functional unit, module in embodiment can be integrated in a processing unit, also may be used
Being that unit is individually physically present, it is also possible to which two or more units are integrated in a unit, above-mentioned integrated
Unit both can be realized in the form of hardware, it would however also be possible to employ the form of SFU software functional unit is realized.In addition, each function list
Unit, the specific name of module are also only to facilitate mutually differentiation, is not limited to the protection domain of the application.Said system
The specific work process of middle unit, module, may be referred to the corresponding process in preceding method embodiment, will not be described here.
Those of ordinary skill in the art are it is to be appreciated that the list of each example with reference to the embodiments described herein description
Unit and algorithm steps, being capable of being implemented in combination in electronic hardware or computer software and electronic hardware.These functions are actually
Performed with hardware or software mode, depending on the application-specific and design constraint of technical scheme.Professional and technical personnel
Each specific application can be used different methods to realize described function, but this realization it is not considered that exceeding
The scope of the present invention.
In embodiment provided by the present invention, it should be understood that disclosed apparatus and method, can pass through other
Mode is realized.For example, system embodiment described above is only schematic, for example, the division of the module or unit,
It is only a kind of division of logic function, there can be other dividing mode when actually realizing, such as multiple units or component can be with
With reference to or be desirably integrated into another system, or some features can be ignored, or not perform.It is another, it is shown or discussed
Coupling each other or direct-coupling or communication connection can be INDIRECT COUPLING by some interfaces, device or unit or
Communication connection, can be electrical, mechanical or other forms.
The unit as separating component explanation can be or may not be it is physically separate, it is aobvious as unit
The part for showing can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On NE.Some or all of unit therein can according to the actual needs be selected to realize the mesh of this embodiment scheme
's.
In addition, each functional unit in each embodiment of the invention can be integrated in a processing unit, it is also possible to
It is that unit is individually physically present, it is also possible to which two or more units are integrated in a unit.Above-mentioned integrated list
Unit both can be realized in the form of hardware, it would however also be possible to employ the form of SFU software functional unit is realized.
If the integrated unit is realized using in the form of SFU software functional unit and as independent production marketing or used
When, during a computer read/write memory medium can be stored in.Based on such understanding, the technical scheme of the embodiment of the present invention
The part for substantially contributing to prior art in other words or all or part of the technical scheme can be with software products
Form embody, the computer software product is stored in a storage medium, including some instructions use so that one
Computer equipment (can be personal computer, server, or network equipment etc.) or processor (processor) perform this
The all or part of step of bright embodiment each embodiment methods described.And aforesaid storage medium includes：USB flash disk, portable hard drive,
Read-only storage (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic
Dish or CD etc. are various can be with the medium of store program codes.
Embodiment described above only to illustrate technical scheme, rather than a limitation；Although with reference to aforementioned reality
Apply example to be described in detail the present invention, it will be understood by those within the art that：It still can be to aforementioned each
Technical scheme described in embodiment is modified, or carries out equivalent to which part technical characteristic；And these are changed
Or replace, the spirit and scope of the essence disengaging various embodiments of the present invention technical scheme of appropriate technical solution are not made, all should
It is included within protection scope of the present invention.
Claims (10)
1. a kind of human motion recognition method, it is characterised in that include：
Obtain the range image sequence and acceleration signal of human action；The range image sequence and the acceleration signal are same
Step；
First eigenvector is extracted from the range image sequence, second feature vector is extracted from the acceleration signal；
With reference to the first eigenvector and second feature vector, Feature Descriptor is drawn；
According to the Feature Descriptor, human action is identified.
2. human motion recognition method according to claim 1, it is characterised in that described from the range image sequence
Extracting first eigenvector includes：
The i-th frame depth image in the range image sequence is obtained, and the i-th frame depth image is projected to into same coordinate
On each orthogonal plane in system, the throwing at each visual angle corresponding with the orthogonal plane of the i-th frame depth image is obtained
Shadow image；Wherein, i is the integer more than or equal to 1；
Calculate the projected image at each visual angle of i+1 frame depth image and the projected image at i-th each visual angle of frame depth image
Absolute difference；
The each frame depth image in the range image sequence is traveled through, the corresponding absolute difference of each frame depth image is superimposed,
Obtain the corresponding Depth Motion mapping of the range image sequence；
Feature extraction is carried out to Depth Motion mapping, the first eigenvector is obtained.
3. human motion recognition method according to claim 2, it is characterised in that described that the Depth Motion is mapped into
Row feature extraction, obtains the first eigenvector and is specially：
Depth Motion mapping to obtaining extracts feature from each visual angle respectively, obtains the Depth Motion and maps each visual angle
Characteristic vector, the subcharacter vector for connecting each visual angle obtains described first eigenvector.
4. human motion recognition method according to claim 1, it is characterised in that described to carry from the acceleration signal
Taking second feature vector includes：
The acceleration signal is split using time slip-window, obtains at least one number comprising acceleration signal fragment
According to window；
In each described data window, the feature of the acceleration signal fragment is extracted, obtain each data window
Characteristic vector；
According to the characteristic vector of all data windows, the second feature vector is drawn.
5. human motion recognition method according to claim 4, it is characterised in that the employing time slip-window is to described
Acceleration signal is split, and obtains at least one data window comprising acceleration signal fragment and is specially：
Adopt Duplication the acceleration signal is split for the time slip-window of preset ratio, and obtained using cross validation
Take that optimal window is long, obtain at least one data window comprising acceleration signal fragment.
6. a kind of human action identifying device, it is characterised in that include：
Acquisition module, for obtaining the range image sequence and acceleration signal of human action；The range image sequence and institute
State acceleration signal synchronization；
Extraction module, for extracting first eigenvector from the range image sequence, extracts from the acceleration signal
Second feature vector；
Feature interpretation submodule, for reference to the first eigenvector and second feature vector, drawing Feature Descriptor；
Identification module, for according to the Feature Descriptor, being identified to human action.
7. human action identifying device according to claim 6, it is characterised in that the extraction module includes：
Processing unit, for obtaining the range image sequence in the i-th frame depth image, and by the i-th frame depth image
Project on each orthogonal plane in the same coordinate system, obtain the corresponding with the orthogonal plane of the i-th frame depth image
Each visual angle projected image；Wherein, i is the integer more than or equal to 1；
Computing unit, for calculate i+1 frame depth image each visual angle projected image and the i-th frame depth image each regard
The absolute difference of the projected image at angle；
Map unit, for traveling through the range image sequence in each frame depth image, be superimposed each frame depth image pair
The absolute difference answered, obtains the corresponding Depth Motion mapping of the range image sequence；
First eigenvector acquiring unit, for carrying out feature extraction to Depth Motion mapping, obtains the fisrt feature
Vector.
8. human action identifying device according to claim 7, it is characterised in that the first eigenvector acquiring unit
Specifically for：Depth Motion mapping to obtaining extracts feature from each visual angle respectively, obtains the Depth Motion and maps each
The characteristic vector at visual angle, the subcharacter vector for connecting each visual angle obtains described first eigenvector.
9. human action identifying device according to claim 6, it is characterised in that the extraction module includes：
Cutting unit, for splitting to the acceleration signal using time slip-window, obtains at least one comprising acceleration
The data window of degree signal segment；
Feature extraction unit, in each described data window, extracting the feature of the acceleration signal fragment, obtains every
The characteristic vector of one data window；
Second feature vector acquiring unit, for according to the characteristic vector of all data windows, draw the second feature to
Amount.
10. human action identifying device according to claim 9, it is characterised in that the cutting unit specifically for：Adopt
The acceleration signal is split with the time slip-window that Duplication is preset ratio, and obtains optimal using cross validation
Window is long, obtains at least one data window comprising acceleration signal fragment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610973939.5A CN106570482B (en) | 2016-11-03 | 2016-11-03 | Human motion recognition method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610973939.5A CN106570482B (en) | 2016-11-03 | 2016-11-03 | Human motion recognition method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106570482A true CN106570482A (en) | 2017-04-19 |
| CN106570482B CN106570482B (en) | 2019-12-03 |
Family
ID=58540016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610973939.5A Active CN106570482B (en) | 2016-11-03 | 2016-11-03 | Human motion recognition method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106570482B (en) |
Cited By (6)
| 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 |
| CN107704799A (en) * | 2017-08-10 | 2018-02-16 | 深圳市金立通信设备有限公司 | A kind of human motion recognition method and equipment, computer-readable recording medium |
| CN108535714A (en) * | 2018-05-25 | 2018-09-14 | 加驰（厦门）智能科技有限公司 | A kind of millimetre-wave radar detection open space blocks the method and device of object |
| CN108985355A (en) * | 2018-06-28 | 2018-12-11 | 中国空间技术研究院 | A kind of data fusion method based on the orthogonal local sensitivity Hash of grouping |
| CN109460734A (en) * | 2018-11-08 | 2019-03-12 | 山东大学 | The video behavior recognition methods and system shown based on level dynamic depth projection difference image table |
| CN115393964A (en) * | 2022-10-26 | 2022-11-25 | 天津科技大学 | Body-building action recognition method and device based on BlazePose |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100310157A1 (en) * | 2009-06-05 | 2010-12-09 | Samsung Electronics Co., Ltd. | Apparatus and method for video sensor-based human activity and facial expression modeling and recognition |
| CN103839040A (en) * | 2012-11-27 | 2014-06-04 | 株式会社理光 | Gesture identification method and device based on depth images |
| CN105608421A (en) * | 2015-12-18 | 2016-05-25 | 中国科学院深圳先进技术研究院 | Human movement recognition method and device |
-
2016
- 2016-11-03 CN CN201610973939.5A patent/CN106570482B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100310157A1 (en) * | 2009-06-05 | 2010-12-09 | Samsung Electronics Co., Ltd. | Apparatus and method for video sensor-based human activity and facial expression modeling and recognition |
| CN103839040A (en) * | 2012-11-27 | 2014-06-04 | 株式会社理光 | Gesture identification method and device based on depth images |
| CN105608421A (en) * | 2015-12-18 | 2016-05-25 | 中国科学院深圳先进技术研究院 | Human movement recognition method and device |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704799A (en) * | 2017-08-10 | 2018-02-16 | 深圳市金立通信设备有限公司 | A kind of human motion recognition method and equipment, computer-readable recording 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 |
| 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 |
| CN108535714A (en) * | 2018-05-25 | 2018-09-14 | 加驰（厦门）智能科技有限公司 | A kind of millimetre-wave radar detection open space blocks the method and device of object |
| CN108535714B (en) * | 2018-05-25 | 2021-10-22 | 厦门精益远达智能科技有限公司 | Method and device for detecting object sheltered in open space by millimeter wave radar |
| CN108985355A (en) * | 2018-06-28 | 2018-12-11 | 中国空间技术研究院 | A kind of data fusion method based on the orthogonal local sensitivity Hash of grouping |
| CN109460734A (en) * | 2018-11-08 | 2019-03-12 | 山东大学 | The video behavior recognition methods and system shown based on level dynamic depth projection difference image table |
| CN109460734B (en) * | 2018-11-08 | 2020-07-31 | 山东大学 | Video behavior identification method and system based on hierarchical dynamic depth projection difference image representation |
| CN115393964A (en) * | 2022-10-26 | 2022-11-25 | 天津科技大学 | Body-building action recognition method and device based on BlazePose |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106570482B (en) | 2019-12-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3961485A1 (en) | Image processing method, apparatus and device, and storage medium | |
| CN106570482A (en) | Method and device for identifying body motion | |
| US8442307B1 (en) | Appearance augmented 3-D point clouds for trajectory and camera localization | |
| CN110807451B (en) | Face key point detection method, device, equipment and storage medium | |
| CN108319901B (en) | Biopsy method, device, computer equipment and the readable medium of face | |
| CN113874870A (en) | Image-based localization | |
| JP7128708B2 (en) | Systems and methods using augmented reality for efficient collection of training data for machine learning | |
| US11328481B2 (en) | Multi-resolution voxel meshing | |
| EP3906527B1 (en) | Image bounding shape using 3d environment representation | |
| CN111243093A (en) | Three-dimensional face grid generation method, device, equipment and storage medium | |
| CN108734058B (en) | Obstacle type identification method, device, equipment and storage medium | |
| CN112927363B (en) | Voxel map construction method and device, computer readable medium and electronic equipment | |
| CN107924571A (en) | Three-dimensional reconstruction is carried out to human ear from a cloud | |
| US20220319146A1 (en) | Object detection method, object detection device, terminal device, and medium | |
| JP2008033958A (en) | Data processing system and method | |
| CN110363817A (en) | Object pose estimation method, electronic equipment and medium | |
| JP2022529367A (en) | Peripheral estimation from postural monocular video | |
| US20220301277A1 (en) | Target detection method, terminal device, and medium | |
| CN111191582A (en) | Three-dimensional target detection method, detection device, terminal device and computer-readable storage medium | |
| US11676361B2 (en) | Computer-readable recording medium having stored therein training program, training method, and information processing apparatus | |
| CN108734773A (en) | A kind of three-dimensional rebuilding method and system for mixing picture | |
| US20220301176A1 (en) | Object detection method, object detection device, terminal device, and medium | |
| CN113592015B (en) | Method and device for positioning and training feature matching network | |
| CN113570725A (en) | Three-dimensional surface reconstruction method and device based on clustering, server and storage medium | |
| WO2021120578A1 (en) | Forward calculation method and apparatus for neural network, and computer-readable storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |