CN108681700A - A kind of complex behavior recognition methods - Google Patents
A kind of complex behavior recognition methods Download PDFInfo
- Publication number
- CN108681700A CN108681700A CN201810421670.9A CN201810421670A CN108681700A CN 108681700 A CN108681700 A CN 108681700A CN 201810421670 A CN201810421670 A CN 201810421670A CN 108681700 A CN108681700 A CN 108681700A
- Authority
- CN
- China
- Prior art keywords
- joint point
- skeletal joint
- motion
- action
- artis
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a kind of complex behavior recognition methods, including：The three-dimensional skeletal joint point information of target movement is obtained using sensor；To artis information pre-processing, normalized coordinate system；Extract the movement locus of each artis, projection to three two dimensional surfaces；The motion vector and its length and deflection for extracting every two interframe are clustered to obtain movement primitive with k means algorithms, and statistics obtains histogram；The weight of each artis is calculated in conjunction with the value of each cluster of all histograms using time pyramid binding time information, forms descriptor；With svm classifier, action recognition is realized.The present invention can carry out action skeletal joint point information the extraction of feature and effectively indicate, improve the accuracy rate of action recognition；All movable informations are completely retained, and action reconstruction can be carried out；Everything class is clustered, human action's feature is captured from the overall situation；Using low-level feature, difficulty in computation is reduced, action recognition efficiency is improved, meets the requirement of real-time of system.
Description
Technical field
The present invention relates to a kind of complex behavior recognition methods, belong to image identification technical field.
Background technology
Action recognition is the hot spot of field of machine vision research, and action identification method has been widely used in man-machine friendship
Mutually, virtual reality, video frequency searching and safety monitoring etc..With the development of depth camera, skeleton artis information can
To be directly acquired, the action identification method based on skeleton character substantially increases the accuracy of action recognition.Although existing permitted
More correlative study and exciting as a result, effective description to human action is still a full of challenges task.
Many methods extract a variety of high-level features from bone information, and then they are combined shape with certain forms
At descriptor, but the descriptor of this integrated mode structure is not complete, and there is always the loss of movable information.Another party
Face, many methods are trained different action classes respectively, this causes the global characteristics of human action in each independent class
Do not move work description in produce deviation.Meanwhile there is also calculate the excessively high problem of cost using the method for high-level feature.Have
Necessity designs a kind of algorithm, using rudimentary another characteristic, reduces and calculates cost, improves efficiency of algorithm, while not having movable information
Loss, and from all categories action in extract human action global characteristics.
Therefore, in view of the above technical problems, it is necessary to propose a kind of complex behavior recognition methods.
Invention content
The goal of the invention of the present invention is to provide a kind of complex behavior recognition methods, is carried out to action skeletal joint point information special
The extraction of sign and effectively expression, improve the accuracy rate of action recognition, completely retain all movable informations, carry out action reconstruction；
Everything class is clustered, human action's feature is captured from the overall situation；Using low-level feature, difficulty in computation is reduced, is carried
High action recognition efficiency, meets the requirement of real-time of system.
To achieve the above object of the invention, the technical solution adopted by the present invention is：A kind of complex behavior recognition methods, including such as
Lower step：
(1), the three-dimensional skeletal joint point information of target movement is obtained using depth transducer, obtains the three-dimensional in each joint of human body
Coordinate；
(2), skeletal joint point information is pre-processed, normalized coordinate system；
(3), the movement locus of each skeletal joint point is extracted, adjacent interframe movement is defined as motion-let；
(4), each skeletal joint point three-dimensional track is projected to three two dimensional surfaces, obtains all skeletal joint point two dimensions
Motion-let collection；
(5), each vector length parameter and direction angular dimensions for representing motion-let are calculated；
(6), the institute's directed quantity for gathering single skeletal joint point everything class, using k-means algorithms to its length parameter and
Direction angular dimensions carries out two-dimentional cluster, obtains movement primitive；
(7), the motion-let quantity that statistics is indicated by each movement primitive obtains movement primitive histogram；
(8), the temporal information of time pyramid capturing motion is utilized；
(9), in conjunction with the value of each cluster of all histograms, the weight of each skeletal joint point is calculated, descriptor is ultimately formed；
(10), final descriptor is trained using SVM classifier, obtains the good division of more action classification descriptors, is realized dynamic
It identifies.
Preferably, step (2) includes：Using the left shoulder of action sequence first frame to right shoulder vector as horizontal axis, with hipbone to double
Shoulder midpoint vector is the longitudinal axis, and X-Y-Z coordinate systems are converted to X '-Y '-Z ' coordinate systems by normalized coordinate system.
Preferably, the movement locus of each skeletal joint point is extracted in step (3), it is specific as follows：
The action sequence S of n frames is expressed as：
S={ Γj| j ∈ [1, J] },
Γj={ pj(t) | t ∈ [1, n], j ∈ [1, J] },
Wherein ΓjIt is the three-dimensional track of artis j, J is artis sum, and t is frame number serial number, Pj(t) it is artis j in t frames
Position：
The movement of front and back two interframe skeletal joint point is defined as motion-let, and skeletal joint point j is in t frames between t+1 frames
Motion-let can be expressed as vector vj(t)：
The then three-dimensional track Γ of skeletal joint point jjIt can be expressed as a sequence vector：
Γj={ vj(t)|t∈[1,n-1]}。
Preferably, step (4) is specific as follows：
The motion-let of each 3D is projected into three two dimensional surfaces, is obtained：
Wherein,WithThe motion-let of 2D, calculation formula on respectively three two dimensional surfaces
It is as follows：
In conjunction with all motion-let, action sequence S is further represented as：
Preferably, step (5) is specific as follows：
The skeletal joint point j under x-y coordinate system, the calculation formula of parameter are as follows：
The skeletal joint point j under y-z coordinate systems, the calculation formula of parameter are as follows：
The skeletal joint point j under x-z coordinate systems, the calculation formula of parameter are as follows：
WhereinWithFor skeletal joint point j t to t+1 frames motion-let in three two dimensional surfaces
The direction angular dimensions of corresponding vector, value range are -180 °~180 °,WithFor skeletal joint point j
Vectorial length parameter is corresponded in three two dimensional surfaces in the motion-let of t to t+1 frames.
Preferably, step (6) is specific as follows：
Under x-y coordinate system, skeletal joint point j, cluster centre quantity is K, then all cluster centres can be expressed as gathering
U：
WhereinIt is the cluster centre of k-th of cluster,WithIt is the coordinate value of the cluster centre,
Each point cluster centre of cluster where it indicates that K cluster centre represents corresponding K movement primitive, k-th of cluster
Movement primitiveIt is expressed as：
Therefore, all movement primitive P of action sequence S are：
Preferably, all under each two dimensional surface of each skeletal joint point of motion sequence S in step (7)
Motion-let is indicated that the motion-let quantity that statistics is indicated by each movement primitive is formed by corresponding movement primitive
Histogram, therefore the value of all clusters can be expressed as H in the histogram under three coordinates of skeletal joint point jj：
Wherein,WithFor movement primitive k-th of column of histogram of three of skeletal joint point j two-dimentional tracks
Value.
Preferably, step (8) uses three layers of time pyramid pull-in time information, first layer to be directed to entire complete track
Movement primitive histogram is calculated, track is temporally divided equally into two parts by the second layer, respectively statistics movement primitive histogram, third
Layer does not divide this two parts further equally, the tracks the 3D descriptor D of final skeletal joint point jjFor：
Preferably, step (9) is specific as follows：
It calculates in entire action sequence, the length of the three-dimensional track of all artis, it is respective total to respectively obtain all artis
Amount of exercise mj：
The total amount of exercise of all artis for adding up the sample, obtains total amount of exercise F of sample action SS：
All training samples of every a kind of action are gathered, is arranged by sample serial number, then has set：
F={ F1,F2,...,Fe,
Wherein e is certain a kind of action training collection sample number,
For each action class, M is calculatedjWith the covariance of F, covariance set is obtained, covariance set specifically calculates step
It is as follows：
The total amount of exercise of all J artis and the total amount of exercise F composition J+1 n-dimensional random variable ns G of sample action：
G=(M1,M2,...,MJ,F)T,
Then matrix
For the covariance matrix of J+1 n-dimensional random variable ns G, covariance calculation formula is：
covij=E [Mi-E(Mi)]×E[Mj-E(Mj)],
Wherein F is as Mj+1It participates in calculating, E (Mj) it is MjMathematic expectaion, calculation formula is as follows：
Last row of Matrix C are the covariance set of Mj and F；If the covariance cov of Mj and Fj(J+1)Less than 0, then artis
The weight w of jjIt is 0；If the covariance cov of Mj and Fj(J+1)More than or equal to 0, then the weight w of artis jjFor：
Wherein, max { covi(J+1)| i ∈ [1, J+1] } refer to the maximum value that Matrix C J+1 is arranged；
Finally, the descriptor D of action sequence S is expressed as：
D={ Dj×wj|j∈[1,J]}。
Since above-mentioned technical proposal is used, the present invention has following advantages compared with prior art：
This method can carry out action skeletal joint point information the extraction of feature and effectively indicate, improve the standard of action recognition
True rate；All movable informations can completely be retained, and action reconstruction can be carried out；Everything class is clustered, from complete
Human action's feature is captured in office；Using low-level feature, difficulty in computation is reduced, improves action recognition efficiency, Ke Yiman
The requirement of real-time of pedal system.
Description of the drawings
Fig. 1 is the coordinate system normalization schematic diagram of the present invention.
Fig. 2 is of the invention projects skeletal joint point three-dimensional track to the schematic diagram of three two dimensional surfaces.
Fig. 3 is the length parameter and direction angular dimensions calculation schematic diagram of the present invention.
Fig. 4 is cluster knot of the vectorial two-dimensional parameter of the skeletal joint point everything classification of the present invention in three planes
Fruit schematic diagram.
Fig. 5 is histogram of the skeletal joint point of the present invention under tri- coordinate systems of x-y, y-z, x-z.
Fig. 6 is the time pyramid schematic diagram of the present invention.
Fig. 7 is the flow chart of the present invention.
Specific implementation mode
The invention will be further described with reference to the accompanying drawings and embodiments：
Embodiment one：Shown in Figure 7, a kind of complex behavior recognition methods includes the following steps：
(1) it utilizes depth transducer to obtain the three-dimensional skeletal joint point information of target movement, obtains the three-dimensional in each joint of human body
Coordinate；
(2) skeletal joint point information is pre-processed, normalized coordinate system is shown in Figure 1, is with left shoulder to right shoulder vector
Horizontal axis, using hipbone to both shoulders midpoint vector as the longitudinal axis, X-Y-Z coordinate systems are converted to X '-Y '-Z ' coordinates by normalized coordinate system
System；
(3) three-dimensional coordinate by each skeletal joint point in action sequence is connected in chronological order, obtains all skeletal joints
The three-dimensional track of point；
The present embodiment is expressed as using the action sequence S (both hands are flat to be waved) with 20 skeletal joint points of 60 frames：
S={ Γj| j ∈ [1,20] },
Γj={ pj(t) | t ∈ [1,60], j ∈ [1,20] },
Wherein ΓjIt is the three-dimensional track of artis j, J is artis sum, and t is frame number serial number, Pj(t) it is artis j in t frames
Position：
The movement of front and back two interframe skeletal joint point is defined as motion-let, skeletal joint point j t frames to t+1 frames it
Between motion-let can be expressed as vector vj(t)：
The then three-dimensional track Γ of skeletal joint point jjIt can be expressed as a sequence vector：
Γj={ vj(t)|t∈[1,n-1]}；
(4) for each skeletal joint point, its three-dimensional track is projected to three two dimensional surfaces, i.e. x-y, y-z and x-z-plane,
Three two-dimentional tracks are obtained, as shown in Figure 2.
The motion-let of each 3D has projected to three two dimensional surfaces, obtains：
Wherein,WithThe motion-let of 2D, calculation formula on respectively three two dimensional surfaces
It is as follows：
In conjunction with all motion-let, action sequence S can be further represented as：
(5) length and deflection for calculating the vector that each represents motion-let, as shown in figure 3, in x-y coordinate system
Lower skeletal joint point j, the calculation formula of the two parameters are as follows：
The skeletal joint point j under y-z coordinate systems, the calculation formula of parameter are as follows：
The skeletal joint point j under x-z coordinate systems, the calculation formula of parameter are as follows：
WhereinWithFor skeletal joint point j t to t+1 frames motion-let in three two dimensional surfaces
The direction angular dimensions of corresponding vector, value range are -180 °~180 °,WithFor skeletal joint point j
Vectorial length parameter is corresponded in three two dimensional surfaces in the motion-let of t to t+1 frames.
(6) in order to extract human motion feature from the overall situation, for each skeletal joint point, set all categories act
Institute's directed quantity of the skeletal joint point carries out two-dimentional cluster with k-means algorithms to its length parameter and direction angular dimensions.With it
In for a certain artis, cluster result is as shown in Figure 4.
By taking skeletal joint point j is in the cluster result under x-y coordinate system as an example, the value of cluster centre quantity K is 8, then institute
Some cluster centres can be expressed as set U：
WhereinIt is the cluster centre of k-th of cluster,WithIt is the coordinate value of the cluster centre.
Each point cluster centre of cluster where it indicates that 8 cluster centres just represent corresponding 8 movement bases
Member.Such as the movement primitive of k-th of clusterIt can be expressed as：
Therefore, all movement primitive P of action sequence S are：
(7) for motion sequence S, all motion-let under each two dimensional surface of each skeletal joint point are by right
Primitive should be moved to indicate, the motion-let quantity that statistics is indicated by each movement primitive forms histogram, as shown in Figure 5.
Cluster for bosom in cluster result and its corresponding movement primitive, are not involved in statistics and follow-up calculating, because most
The cluster at center contains the vector of all directions, does not have the meaning divided on deflection.In addition to this, the cluster in bosom
Vector length parameter all very littles, it is smaller to the composition contribution of movement, it can be ignored.
The value of all clusters can be expressed as H in three histograms of skeletal joint point jj：
Wherein,WithFor movement primitive k-th of column of histogram of three of skeletal joint point j two-dimentional tracks
Value.
(8) application time pyramid carrys out the temporal information of capturing motion, and by taking three layers of time pyramid as an example, first layer is directed to
Entire complete trajectory calculation moves primitive histogram, and track is temporally divided equally into two parts by the second layer, respectively statistics movement
Primitive histogram, this two parts are further bisected into respectively in third layer.Therefore the description of skeletal joint point j is divided into 7 parts,
As shown in Figure 6：First layerThe second layerWithThird layerWith
Therefore, the tracks the 3D descriptor D of skeletal joint point jjIt is the combination of above 7 parts, i.e.,：
(9) consider that different skeletal joint points is different for the identification percentage contribution of action, it is necessary to increase crucial bone
Importance of the bone artis descriptor to action recognition.By taking the descriptor of skeletal joint point j as an example, corresponding weight w is calculatedj。
It calculates in entire action sequence, the length of the three-dimensional track of all artis respectively obtains all artis respectively
Total amount of exercise mj：
The total amount of exercise of all artis for adding up the sample, obtains total amount of exercise F of sample action SS：
All training samples of every a kind of action are gathered, are arranged by sample serial number, certain one kind is dynamic in the present embodiment
There are 80 samples as training set, then has set：
F={ F1,F2,...,F80}。
For each action class, M is calculatedjWith the covariance of F, covariance set is obtained.Steps are as follows for specific calculating：Institute
21 n-dimensional random variable n G are formed by the total amount of exercise of 20 artis and the total amount of exercise F of sample action：
G=(M1,M2,...,M20,F)T,
Then matrix
For the covariance matrix of 21 n-dimensional random variable n G, covariance calculation formula is：
covij=E [Mi-E(Mi)]×E[Mj-E(Mj)],
Wherein F is as M21It participates in calculating, E (Mj) it is MjMathematic expectaion, calculation formula is as follows：
Last row of Matrix C are the covariance set of Mj and F.If the covariance cov of Mj and Fj(21)Less than 0, then close
The weight w of node jjIt is 0.If the covariance cov of Mj and Fj(21)More than or equal to 0, then the weight w of artis jjFor：
Wherein, max { covi(21)| i ∈ [1,21] } refer to the maximum value that Matrix C the 21st arranges.
Therefore, the descriptor D of action sequence S can be expressed as：
D={ Dj×wj|j∈[1,20]}。
(10) final descriptor is trained using SVM classifier, obtains the good division of more action classification descriptors, realized
Action recognition.
Claims (9)
1. a kind of complex behavior recognition methods, which is characterized in that include the following steps：
(1), the three-dimensional skeletal joint point information of target movement is obtained using depth transducer, obtains the three-dimensional in each joint of human body
Coordinate；
(2), skeletal joint point information is pre-processed, normalized coordinate system；
(3), the movement locus of each skeletal joint point is extracted, adjacent interframe movement is defined as motion-let；
(4), each skeletal joint point three-dimensional track is projected to three two dimensional surfaces, obtains all skeletal joint point two dimensions
Motion-let collection；
(5), each vector length parameter and direction angular dimensions for representing motion-let are calculated；
(6), the institute's directed quantity for gathering single skeletal joint point everything class, using k-means algorithms to its length parameter and
Direction angular dimensions carries out two-dimentional cluster, obtains movement primitive；
(7), the motion-let quantity that statistics is indicated by each movement primitive obtains movement primitive histogram；
(8), the temporal information of time pyramid capturing motion is utilized；
(9), in conjunction with the value of each cluster of all histograms, the weight of each skeletal joint point is calculated, descriptor is ultimately formed；
(10), final descriptor is trained using SVM classifier, obtains the good division of more action classification descriptors, is realized dynamic
It identifies.
2. according to the complex behavior recognition methods in claim 1, which is characterized in that step (2) includes：With action sequence first
The left shoulder of frame to right shoulder vector is horizontal axis, using hipbone to both shoulders midpoint vector as the longitudinal axis, normalized coordinate system, by X-Y-Z coordinates
System is converted to X '-Y '-Z ' coordinate systems.
3. according to the complex behavior recognition methods in claim 1, which is characterized in that each skeletal joint of extraction in step (3)
The movement locus of point, it is specific as follows：
The action sequence S of n frames is expressed as：
S={ Γj| j ∈ [1, J] },
Γj={ pj(t) | t ∈ [1, n], j ∈ [1, J] },
Wherein ΓjIt is the three-dimensional track of artis j, J is artis sum, and t is frame number serial number, Pj(t) it is artis j in t frames
Position：
The movement of front and back two interframe skeletal joint point is defined as motion-let, and skeletal joint point j is in t frames between t+1 frames
Motion-let can be expressed as vector vj(t)：
The then three-dimensional track Γ of skeletal joint point jjIt can be expressed as a sequence vector：
Γj={ vj(t)|t∈[1,n-1]}。
4. according to the complex behavior recognition methods in claim 1, which is characterized in that step (4) is specific as follows：
The motion-let of each 3D is projected into three two dimensional surfaces, is obtained：
Wherein,WithThe motion-let of 2D, calculation formula on respectively three two dimensional surfaces
It is as follows：
In conjunction with all motion-let, action sequence S is further represented as：
5. according to the complex behavior recognition methods in claim 1, which is characterized in that step (5) is specific as follows：
The skeletal joint point j under x-y coordinate system, the calculation formula of parameter are as follows：
The skeletal joint point j under y-z coordinate systems, the calculation formula of parameter are as follows：
The skeletal joint point j under x-z coordinate systems, the calculation formula of parameter are as follows：
WhereinWithFor skeletal joint point j t to t+1 frames motion-let in three two dimensional surfaces pair
It is -180 °~180 ° to answer the direction angular dimensions of vector, value range,WithIt is skeletal joint point j in t
Motion-let to t+1 frames corresponds to vectorial length parameter in three two dimensional surfaces.
6. according to the complex behavior recognition methods in claim 1, which is characterized in that step (6) is specific as follows：
Under x-y coordinate system, skeletal joint point j, cluster centre quantity is K, then all cluster centres can be expressed as gathering
U：
WhereinIt is the cluster centre of k-th of cluster,WithIt is the coordinate value of the cluster centre,
Each point cluster centre of cluster where it indicates that K cluster centre represents corresponding K movement primitive, k-th of cluster
Movement primitiveIt is expressed as：
Therefore, all movement primitive P of action sequence S are：
7. according to the complex behavior recognition methods in claim 1, which is characterized in that in step (7), motion sequence S's is every
All motion-let under each two dimensional surface of a skeletal joint point are indicated by corresponding movement primitive, are counted by each
The motion-let quantity that primitive indicates is moved, forms histogram, therefore in the histogram under three coordinates of skeletal joint point j
The value of all clusters can be expressed as Hj：
Wherein,WithFor movement primitive k-th of column of histogram of three of skeletal joint point j two-dimentional tracks
Value.
8. according to the complex behavior recognition methods in claim 1, which is characterized in that step (8) uses three layers of time pyramid
Pull-in time information, first layer move primitive histogram for entire complete trajectory calculation, and the second layer temporally puts down track
It is divided into two parts, statistics movement primitive histogram, third layer do not divide this two parts further equally respectively, final skeletal joint point
The tracks the 3D descriptor D of jjFor：
9. according to the complex behavior recognition methods in claim 1, which is characterized in that step (9) is specific as follows：
It calculates in entire action sequence, the length of the three-dimensional track of all artis, it is respective total to respectively obtain all artis
Amount of exercise mj：
The total amount of exercise of all artis for adding up the sample, obtains total amount of exercise F of sample action SS：
All training samples of every a kind of action are gathered, is arranged by sample serial number, then has set：
F={ F1,F2,...,Fe,
Wherein e is certain a kind of action training collection sample number,
For each action class, M is calculatedjWith the covariance of F, covariance set is obtained, covariance set specifically calculates step such as
Under：
The total amount of exercise of all J artis and the total amount of exercise F composition J+1 n-dimensional random variable ns G of sample action：
G=(M1,M2,...,MJ,F)T,
Then matrix
For the covariance matrix of J+1 n-dimensional random variable ns G, covariance calculation formula is：
covij=E [Mi-E(Mi)]×E[Mj-E(Mj)],
Wherein F is as Mj+1It participates in calculating, E (Mj) it is MjMathematic expectaion, calculation formula is as follows：
Last row of Matrix C are the covariance set of Mj and F；If the covariance cov of Mj and Fj(J+1)Less than 0, then artis
The weight w of jjIt is 0；If the covariance cov of Mj and Fj(J+1)More than or equal to 0, then the weight w of artis jjFor：
Wherein, max { covi(J+1)| i ∈ [1, J+1] } refer to the maximum value that Matrix C J+1 is arranged；
Finally, the descriptor D of action sequence S is expressed as：
D={ Dj×wj|j∈[1,J]}。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810421670.9A CN108681700B (en) | 2018-05-04 | 2018-05-04 | Complex behavior identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810421670.9A CN108681700B (en) | 2018-05-04 | 2018-05-04 | Complex behavior identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108681700A true CN108681700A (en) | 2018-10-19 |
| CN108681700B CN108681700B (en) | 2021-09-28 |
Family
ID=63801510
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810421670.9A Active CN108681700B (en) | 2018-05-04 | 2018-05-04 | Complex behavior identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108681700B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670401A (en) * | 2018-11-15 | 2019-04-23 | 天津大学 | A kind of action identification method based on skeleton motion figure |
| CN110070070A (en) * | 2019-04-30 | 2019-07-30 | 苏州大学 | A kind of action identification method |
| CN110084211A (en) * | 2019-04-30 | 2019-08-02 | 苏州大学 | A kind of action identification method |
| CN110119707A (en) * | 2019-05-10 | 2019-08-13 | 苏州大学 | A kind of human motion recognition method |
| CN110414316A (en) * | 2019-06-11 | 2019-11-05 | 中国科学院自动化研究所 | Data de-noising method, apparatus, computer equipment and storage medium |
| CN111028339A (en) * | 2019-12-06 | 2020-04-17 | 国网浙江省电力有限公司培训中心 | Behavior action modeling method and device, electronic equipment and storage medium |
| CN111310590A (en) * | 2020-01-20 | 2020-06-19 | 北京西米兄弟未来科技有限公司 | Action recognition method and electronic equipment |
| CN111914798A (en) * | 2020-08-17 | 2020-11-10 | 四川大学 | Human body behavior identification method based on skeletal joint point data |
| CN113011381A (en) * | 2021-04-09 | 2021-06-22 | 中国科学技术大学 | Double-person motion identification method based on skeleton joint data |
| US11625938B2 (en) | 2020-12-29 | 2023-04-11 | Industrial Technology Research Institute | Method and device for detecting human skeletons |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104298974A (en) * | 2014-10-10 | 2015-01-21 | 北京工业大学 | Human body behavior recognition method based on depth video sequence |
| CN104598890A (en) * | 2015-01-30 | 2015-05-06 | 南京邮电大学 | Human body behavior recognizing method based on RGB-D video |
| US20150279053A1 (en) * | 2014-03-31 | 2015-10-01 | Electronics And Telecommunications Research Institute | System and method for motion estimation |
| CN107194366A (en) * | 2017-06-06 | 2017-09-22 | 西安电子科技大学 | The Activity recognition method of son is described based on dense track covariance |
-
2018
- 2018-05-04 CN CN201810421670.9A patent/CN108681700B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150279053A1 (en) * | 2014-03-31 | 2015-10-01 | Electronics And Telecommunications Research Institute | System and method for motion estimation |
| CN104298974A (en) * | 2014-10-10 | 2015-01-21 | 北京工业大学 | Human body behavior recognition method based on depth video sequence |
| CN104598890A (en) * | 2015-01-30 | 2015-05-06 | 南京邮电大学 | Human body behavior recognizing method based on RGB-D video |
| CN107194366A (en) * | 2017-06-06 | 2017-09-22 | 西安电子科技大学 | The Activity recognition method of son is described based on dense track covariance |
Non-Patent Citations (2)
| Title |
|---|
| J.W. WENG ET AL.: "Spatio-Temporal Naive-Bayes Nearest-Neighbor (ST-NBNN) for Skeleton-Based Action Recognition", 《CVPR》 * |
| 丁毅 等: "基于BOF-Gist特征的手势识别算法研究", 《计算机工程与应用》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670401A (en) * | 2018-11-15 | 2019-04-23 | 天津大学 | A kind of action identification method based on skeleton motion figure |
| CN110070070A (en) * | 2019-04-30 | 2019-07-30 | 苏州大学 | A kind of action identification method |
| CN110084211A (en) * | 2019-04-30 | 2019-08-02 | 苏州大学 | A kind of action identification method |
| CN110119707A (en) * | 2019-05-10 | 2019-08-13 | 苏州大学 | A kind of human motion recognition method |
| CN110414316A (en) * | 2019-06-11 | 2019-11-05 | 中国科学院自动化研究所 | Data de-noising method, apparatus, computer equipment and storage medium |
| CN111028339A (en) * | 2019-12-06 | 2020-04-17 | 国网浙江省电力有限公司培训中心 | Behavior action modeling method and device, electronic equipment and storage medium |
| CN111028339B (en) * | 2019-12-06 | 2024-03-29 | 国网浙江省电力有限公司培训中心 | Behavior modeling method and device, electronic equipment and storage medium |
| CN111310590A (en) * | 2020-01-20 | 2020-06-19 | 北京西米兄弟未来科技有限公司 | Action recognition method and electronic equipment |
| CN111914798A (en) * | 2020-08-17 | 2020-11-10 | 四川大学 | Human body behavior identification method based on skeletal joint point data |
| CN111914798B (en) * | 2020-08-17 | 2022-06-07 | 四川大学 | Human body behavior identification method based on skeletal joint point data |
| US11625938B2 (en) | 2020-12-29 | 2023-04-11 | Industrial Technology Research Institute | Method and device for detecting human skeletons |
| CN113011381A (en) * | 2021-04-09 | 2021-06-22 | 中国科学技术大学 | Double-person motion identification method based on skeleton joint data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108681700B (en) | 2021-09-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108681700A (en) | A kind of complex behavior recognition methods | |
| EP3711024B1 (en) | Event camera-based deformable object tracking | |
| WO2021129064A9 (en) | Posture acquisition method and device, and key point coordinate positioning model training method and device | |
| CN102332095B (en) | Face motion tracking method, face motion tracking system and method for enhancing reality | |
| Ding et al. | STFC: Spatio-temporal feature chain for skeleton-based human action recognition | |
| Sincan et al. | Using motion history images with 3d convolutional networks in isolated sign language recognition | |
| CN108241849A (en) | Human body interactive action recognition methods based on video | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN108875586B (en) | Functional limb rehabilitation training detection method based on depth image and skeleton data multi-feature fusion | |
| CN101877143A (en) | Three-dimensional scene reconstruction method of two-dimensional image group | |
| CN107808376A (en) | A kind of detection method of raising one's hand based on deep learning | |
| CN108171133A (en) | A kind of dynamic gesture identification method of feature based covariance matrix | |
| CN101826155B (en) | Method for identifying act of shooting based on Haar characteristic and dynamic time sequence matching | |
| CN106327528A (en) | Moving object tracking method and operation method of unmanned aerial vehicle | |
| CN108280421A (en) | Human bodys' response method based on multiple features Depth Motion figure | |
| CN113378649A (en) | Identity, position and action recognition method, system, electronic equipment and storage medium | |
| CN114419732A (en) | HRNet human body posture identification method based on attention mechanism optimization | |
| CN110555383A (en) | Gesture recognition method based on convolutional neural network and 3D estimation | |
| CN112906520A (en) | Gesture coding-based action recognition method and device | |
| CN106529441A (en) | Fuzzy boundary fragmentation-based depth motion map human body action recognition method | |
| CN104794446A (en) | Human body action recognition method and system based on synthetic descriptors | |
| Batool et al. | Telemonitoring of daily activities based on multi-sensors data fusion | |
| Zhang | Analyzing body changes of high-level dance movements through biological image visualization technology by convolutional neural network | |
| Pismenskova et al. | Classification of a two-dimensional pose using a human skeleton | |
| CN111626212B (en) | Method and device for identifying object in picture, storage medium and electronic device |
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 |