CN106228164A - A kind of construction method in video dynamic primitive storehouse - Google Patents

A kind of construction method in video dynamic primitive storehouse Download PDF

Info

Publication number
CN106228164A
CN106228164A CN201610596607.XA CN201610596607A CN106228164A CN 106228164 A CN106228164 A CN 106228164A CN 201610596607 A CN201610596607 A CN 201610596607A CN 106228164 A CN106228164 A CN 106228164A
Authority
CN
China
Prior art keywords
video
matrix
distance
metric
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201610596607.XA
Other languages
Chinese (zh)
Inventor
罗冠
胡卫明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Automation of Chinese Academy of Science
Original Assignee
Institute of Automation of Chinese Academy of Science
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Institute of Automation of Chinese Academy of Science filed Critical Institute of Automation of Chinese Academy of Science
Priority to CN201610596607.XA priority Critical patent/CN106228164A/en
Publication of CN106228164A publication Critical patent/CN106228164A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/71—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Library & Information Science (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses the construction method in a kind of video dynamic primitive storehouse.The method comprises the following steps: step 1: collects video data file and constitutes video sample collection；Step 2: for each video data file of video sample collection, extracts three-dimensional space-time local feature；Step 3: for each three-dimensional space-time local feature, uses linear dynamic system as description of described three-dimensional space-time local feature；Step 4: calculate all three-dimensional space-time local features distance metric between any two, constitutes a distance matrix metric；Step 5: cluster described distance matrix metric, cluster centre is as the dynamic primitive storehouse of described video sample collection.The present invention can apply in internet video content retrieval, sensitive video frequency detection and the business such as filtration and intelligent video monitoring.

Description

A kind of construction method in video dynamic primitive storehouse
Technical field
The present invention relates to image procossing and Computer Applied Technology field, particularly to the structure in a kind of video dynamic primitive storehouse Construction method.
Background technology
Along with developing rapidly of social economy and science and technology, video has been widely used in various place, such as, The monitoring system for security protection is installed in the area such as bank, airport, resident living area, and every day produces substantial amounts of monitor video data； The most such as, store the video frequency program of magnanimity on the internet, and the most also grow at top speed constantly.How such as The video data of this magnanimity retrieves our desired content rapidly and accurately, is one and there is important research value with huge The practical problem of big using value, this problem effectively solve by the deep development of related industry is played important promotion and Impetus.
One important step of Video content analysis technique is to build primitive storehouse (also referred to as dictionary, the word of video sample collection Bag etc.).The effect in primitive storehouse is the representative feature vector found and can express video sample collection, reduces the complexity of representation of video shot Degree, improves the accuracy of representation of video shot, thus promotes the precision of video frequency searching and identification.The structure in primitive storehouse mainly includes two The content of aspect: first, calculates video sample collection local feature distance metric between any two；Second, local feature is carried out Cluster, determines cluster centre.According to video sample collection Local Feature Extraction and the difference of the method for description, finally obtained Primitive stock is in very big difference.
The most conventional method for describing local characteristic includes gradient orientation histogram (HOG), light stream rectangular histogram (HOF) and fortune Moving boundary rectangular histogram (MBH) etc..Gradient orientation histogram is believed by calculating the gradient direction of every two field picture in space-time local feature Breath describes local feature.This method focuses on the apparent information of static state of feature, but have ignored in feature the fortune between frame and frame Dynamic information.Light stream rectangular histogram considers emphatically the movable information in feature between frame and frame, but this method is difficult to process camera lens The situation of displacement.Moving boundaries rectangular histogram is on the basis of light stream is histogrammic, by calculating the gradient information of optical flow field, energy Effectively filter the information of camera lens displacement, therefore can preferably describe the movable information of interesting target in video.
In actual video content analysis system, a viewpoint with common recognition is: the apparent information of static state of feature and Movable information, in terms of describing video features, has status of equal importance.To this end, the present invention provides a kind of based on linear dynamic The video dynamic primitive base construction method of system.Linear dynamic system can portray simultaneously local feature the apparent information of static state and Dynamic motion information, therefore its differentiation discriminating power is better than above-mentioned single method.Owing to linear dynamic system cannot be expressed as Europe A point in formula space, in order to cluster linear dynamic system, the present invention is calculating linear dynamic system between any two Distance metric constitute after a distance matrix metric, directly above-mentioned distance matrix metric is clustered, cluster result is made Primitive storehouse for video sample collection.Described primitive storehouse can significantly improve video frequency searching and the performance of identification and accuracy.
Summary of the invention
It is an object of the invention to propose the construction method in a kind of video dynamic primitive storehouse, it is thus achieved that video can be portrayed simultaneously Static apparent information and the video dynamic primitive storehouse of dynamic motion information, this primitive storehouse can significantly improve video frequency searching and identification Performance and accuracy.
The construction method in a kind of video dynamic primitive storehouse that the present invention proposes, comprises the following steps:
Step 1: collect video data file and constitute video sample collection；
Step 2: each video data file concentrated for video sample, extracts three-dimensional space-time local feature, constitutes spy Collection；
Step 3: for each three-dimensional space-time local feature, uses linear dynamic system special as described three-dimensional space-time local Description levied；
Step 4: calculate three-dimensional space-time local feature distance metric between any two in feature set, constitute a distance metric Matrix；
Step 5: cluster described distance matrix metric, cluster centre is as the dynamic primitive of described video sample collection Storehouse.
Preferably, the size of described three dimensions local feature is N × N × L, and wherein, N × N is every frame video image On local pixel block size centered by point of interest, L is the frame number on time orientation.
Preferably, in step 3 calculate three-dimensional space-time local feature describe son method particularly as follows:
Step 31, makes three-dimensional space-time local feature Y meet linear dynamic system model, and linear dynamic system model is expressed as Such as following formula:
x t + 1 = A x t + v t y t = Cx t + w t
Wherein, Y={y1,…,yi,…,yL, wherein yiIt is that the i-th frame block of pixels converts the column vector obtained；With subscript t table Show the moment of frame of video discrete in Y, xtRepresent the state variable of linear dynamic system；ytRepresent the feature of linear dynamic system Variable；vt,wtThe noise variation of expression system；A, C represent the model parameter of linear dynamic system；
Step 32, obtains parameter A of described linear dynamic system model, C by singular value decomposition and method of least square；
Singular value decomposition decomposition is carried out: Y=U Σ V firstly for YT,
Obtain model parameter C and systematic state variable X: C=U；X=Σ VT
Model parameter A computing formula under least square meaning is as follows:
Wherein,Represent More-Pan Lusi (Moore-Penrose) generalized inverse matrix.
Preferably, step 4 specifically includes following steps:
Step 41, builds distance matrix metric D, element D in distance matrix metric DijRepresent feature set ith feature and the Distance between j feature, the model parameter of the linear dynamic system of two features is (A respectivelyi,Ci) and (Aj,Cj), structure The matrix in block form obtained is:
A = A i 0 0 A j , C = C i C j .
Step 42, sets up Lyapunov Equation based on described matrix in block form and solves, obtaining matrix in block formMatrixCharacteristic root be exactly square cosine cos of subspace angle theta2θ；
Wherein, described Lyapunov Equation is expressed as:
Q=ATQA+CTC；
Step 43, the distance between feature i and feature j is:
D i j = - l o g Π k = 1 n cos 2 θ k .
Preferably, the algorithm using metric matrix of can directly adjusting the distance to carry out clustering in step 5 clusters.
Preferably, the algorithm that described metric matrix of can directly adjusting the distance carries out clustering be K central point (K-medoid) or Specification cuts (Normalized cuts).
Preferably, the individual number interval of cluster centre that metric matrix of adjusting the distance carries out clustering is [500,4000].
Preferably, the size N × N × L of described three dimensions local feature chooses, and N is chosen for 32 or 16, and L takes Value scope is [15,20].
The construction method in a kind of video dynamic primitive storehouse that the present invention proposes, can be portrayed by linear dynamic system simultaneously The apparent information of static state of local feature and dynamic motion information, solve from a brand-new angle and catch videometer the most simultaneously Sight information and a difficult problem for movable information, it distinguishes discriminating power also superior to current existing method；It is linear dynamic that the present invention provides The method that the distance metric of state system calculates, it is possible to calculate the distance metric described between son of non-Euclidean space, is used for expressing line Similarity between property dynamical system；The method of the cluster calculation of the distance matrix metric that the present invention provides, it is possible to process non-Europe The clustering problem of scalar characteristic point；The present invention can be widely used in Video content retrieval, sensitive video frequency detection with filter with And in the business such as intelligent video monitoring.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of the present invention.
Detailed description of the invention
So that advantages of the present invention, technical scheme, goal of the invention will be apparent from clear, below in conjunction with example and attached Figure, further divides the present invention and elaborates.Wherein, the used herein example that is embodied as is used only for explaining this Bright, it is not intended to limit the present invention.
The schematic flow sheet of construction method in a kind of video dynamic primitive storehouse that the present invention provides, as it is shown in figure 1, include with Lower step:
Step 1: collect video data file and constitute video sample collection；
Step 2: each video data file concentrated for video sample, extracts three-dimensional space-time local feature, constitutes spy Collection；
The size of described three dimensions local feature is N × N × L, and wherein, N × N is with interest on every frame video image Local pixel block size centered by Dian, L is the frame number on time orientation.All such space-time local features constitute described The space-time local feature collection of video data file, described point of interest can be local sparse point of interest, it is also possible to be according to grade between Every the dense point of interest that sampling obtains.
Choosing of described three-dimensional space-time local feature size, N is typically chosen for 32 or 16, the most apparent excessive change play Strong, too small, it is not enough to portray the apparent information in local；The general span of L is [15,20], and track is long easily causes tracking mistake Lose, and linear system condition may be unsatisfactory for；The too short then dynamic characteristic of track is inconspicuous, also results in track too much simultaneously, meter Calculation amount sharply increases.
Step 3: for each three-dimensional space-time local feature, uses linear dynamic system special as described three-dimensional space-time local Description levied；
In step 3 calculate three-dimensional space-time local feature describe son method particularly as follows:
Step 31, makes three-dimensional space-time local feature Y meet linear dynamic system model, and linear dynamic system model is expressed as Formula (1):
x t + 1 = Ax t + v t y t = Cx t + w t - - - ( 1 )
Wherein, Y={y1,…,yi,…,yL, wherein yiIt is that the i-th frame block of pixels converts the column vector obtained；With subscript t table Show the moment of frame of video discrete in Y；xtRepresenting the state variable of linear dynamic system, its dimension is referred to as linear dynamic system Exponent number, generally this exponent number are far smaller than characteristic variable ytDimension, its span is [3,10]；ytRepresent linear dynamic system The characteristic variable of system is (as t represents the moment of frame of video i discrete in Y, then yt=yi)；vt,wtThe noise variation of expression system； A, C represent the model parameter of linear dynamic system.Model parameter A, C can be used to description as space-time local feature Y.
Step 32, obtains parameter A of described linear dynamic system model, C by singular value decomposition and method of least square.
The basic skills of singular value decomposition is: assume that M is m × n rank matrix, then there is a decomposition and make M=U Σ V*；Wherein U is m × m rank unitary matrice, and Σ is positive semidefinite m × n rank diagonal matrix, and V* is the conjugate transpose of V, and V* is n × n rank Unitary matrice.
In order to obtain model parameter A in the present embodiment, C, carry out singular value decomposition firstly for Y, as shown in formula (2):
Y=U Σ VT (2)
Obtain model parameter C and systematic state variable X, respectively as shown in formula (3), (4):
C=U (3)
X=Σ VT (4)
Shown in model parameter A computing formula under least square meaning such as formula (5):
Wherein,Represent More-Pan Lusi (Moore-Penrose) generalized inverse matrix.
So far, completing linear dynamic system model parameter A, C solves.
Step 4: calculate three-dimensional space-time local feature distance metric between any two in feature set, constitute a distance metric Matrix；
Described step 4 further includes steps of
Rapid 4 specifically include following steps:
Step 41, builds distance matrix metric D, element D in distance matrix metric DijRepresent feature set ith feature and the Distance between j feature, the model parameter of the linear dynamic system of two features is (A respectivelyi,Ci) and (Aj,Cj), structure The matrix in block form obtained is:
A = A i 0 0 A j
C=(Ci Cj)。
Step 42, sets up Lyapunov Equation based on described matrix in block form and solves, obtaining matrix in block form MatrixCharacteristic root be exactly square cosine cos of subspace angle theta2θ；
Wherein, described Lyapunov Equation is expressed as formula (6):
Q=ATQA+CTC (6)
Step 43, the distance between feature i and feature j is shown in formula (7):
D i j = - l o g Π k = 1 n cos 2 θ k - - - ( 7 )
Calculate all three-dimensional local feature distance metrics between any two according to above step, constitute distance matrix metric D.
Step 5: cluster described distance matrix metric, cluster centre is as the dynamic primitive of described video sample collection Storehouse.
Described step 5 is particularly as follows: use the algorithm that metric matrix of can directly adjusting the distance carries out clustering, such as K central point (K-medoid) algorithm or specification are cut (Normalized cuts) algorithm and are clustered it.The number of cluster centre according to The difference of problem, can select interval for [500,4000].These cluster centres just constitute the dynamic primitive storehouse of video sample collection.
The construction method in a kind of video dynamic primitive storehouse that the present invention proposes, can be portrayed by linear dynamic system simultaneously The apparent information of static state of local feature and dynamic motion information, solve from a brand-new angle and catch videometer the most simultaneously Sight information and a difficult problem for movable information, it distinguishes discriminating power also superior to current existing method；It is linear dynamic that the present invention provides The method that the distance metric of state system calculates, it is possible to calculate the distance metric described between son of non-Euclidean space, is used for expressing line Similarity between property dynamical system；The method of the cluster calculation of the distance matrix metric that the present invention provides, it is possible to process non-Europe The clustering problem of scalar characteristic point；The present invention can be widely used in Video content retrieval, sensitive video frequency detection with filter with And in the business such as intelligent video monitoring.
Particular embodiments described above, has been carried out the purpose of the present invention, technical scheme and beneficial effect the most in detail Describe in detail bright, be it should be understood that the specific embodiment that the foregoing is only the present invention, be not limited to the present invention, all Within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. done, should be included in the guarantor of the present invention Within the scope of protecting.

Claims (8)

1. the construction method in a video dynamic primitive storehouse, it is characterised in that comprise the following steps:
Step 1: collect video data file and constitute video sample collection；
Step 2: each video data file concentrated for video sample, extracts three-dimensional space-time local feature, constitutive characteristic collection；
Step 3: for each three-dimensional space-time local feature, uses linear dynamic system as described three-dimensional space-time local feature Son is described；
Step 4: calculate three-dimensional space-time local feature distance metric between any two in feature set, constitute a distance metric square Battle array；
Step 5: cluster described distance matrix metric, cluster centre is as the dynamic primitive storehouse of described video sample collection.
2. the method for claim 1, it is characterised in that the size of described three dimensions local feature is N × N × L, Wherein, N × N is the local pixel block size on every frame video image centered by point of interest, and L is the frame number on time orientation.
3. method as claimed in claim 2, it is characterised in that calculate description of three-dimensional space-time local feature in step 3 Method particularly as follows:
Step 31, makes three-dimensional space-time local feature Y meet linear dynamic system model, and linear dynamic system model is expressed as Formula:
x t + 1 = Ax t + v t y t = Cx t + w t
Wherein, Y={y1,…,yi,…,yL, wherein yiIt is that the i-th frame block of pixels converts the column vector obtained；Represent in Y by subscript t The moment of t frame of video, xtRepresent the state variable of linear dynamic system；ytRepresent the characteristic variable of linear dynamic system； vt,wtThe noise variation of expression system；A, C represent the model parameter of linear dynamic system；
Step 32, obtains parameter A of described linear dynamic system model, C by singular value decomposition and method of least square；
Singular value decomposition decomposition is carried out: Y=U Σ V firstly for YT,
Obtain model parameter C and systematic state variable X: C=U；X=Σ VT
Model parameter A computing formula under least square meaning is as follows:
Wherein,Represent More-Pan Lusi (Moore-Penrose) generalized inverse matrix.
4. method as claimed in claim 3, it is characterised in that step 4 specifically includes following steps:
Step 41, builds distance matrix metric D, element D in distance matrix metric DijRepresent feature set ith feature and jth Distance between feature, the model parameter of the linear dynamic system of two features is (A respectivelyi,Ci) and (Aj,Cj), structure obtains Matrix in block form be:
A = A i 0 0 A j , C = C i C j ;
Step 42, sets up Lyapunov Equation based on described matrix in block form and solves, obtaining matrix in block form MatrixCharacteristic root be exactly square cosine cos of subspace angle theta2θ；
Wherein, described Lyapunov Equation is expressed as:
Q=ATQA+CTC；
Step 43, the distance between feature i and feature j is:
D i j = - l o g Π k = 1 n cos 2 θ k .
5. method as claimed in claim 4, it is characterised in that use metric matrix of can directly adjusting the distance to carry out in step 5 The algorithm of cluster clusters.
6. method as claimed in claim 5, it is characterised in that the calculation that described metric matrix of can directly adjusting the distance carries out clustering Method is K central point (K-medoid) or specification cuts (Normalized cuts).
7. method as claimed in claim 6, it is characterised in that metric matrix of adjusting the distance carries out the number of the cluster centre clustered Interval is [500,4000].
8. method as claimed in claim 7, it is characterised in that the size N × N × L's of described three dimensions local feature Choosing, N is chosen for 32 or 16, and L span is [15,20].
CN201610596607.XA 2016-07-26 2016-07-26 A kind of construction method in video dynamic primitive storehouse Pending CN106228164A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610596607.XA CN106228164A (en) 2016-07-26 2016-07-26 A kind of construction method in video dynamic primitive storehouse

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610596607.XA CN106228164A (en) 2016-07-26 2016-07-26 A kind of construction method in video dynamic primitive storehouse

Publications (1)

Publication Number Publication Date
CN106228164A true CN106228164A (en) 2016-12-14

Family

ID=57534314

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610596607.XA Pending CN106228164A (en) 2016-07-26 2016-07-26 A kind of construction method in video dynamic primitive storehouse

Country Status (1)

Country Link
CN (1) CN106228164A (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102682306A (en) * 2012-05-02 2012-09-19 武汉大学 Wavelet pyramid polarization texture primitive feature extracting method for synthetic aperture radar (SAR) images
CN103246891A (en) * 2013-05-28 2013-08-14 重庆邮电大学 Chinese sign language recognition method based on kinect
CN103854014A (en) * 2014-02-25 2014-06-11 中国科学院自动化研究所 Terror video identification method and device based on sparse representation of context
CN104199902A (en) * 2014-08-27 2014-12-10 中国科学院自动化研究所 Similarity measurement computing method of linear dynamical systems
CN104200235A (en) * 2014-07-28 2014-12-10 中国科学院自动化研究所 Time-space local feature extraction method based on linear dynamic system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102682306A (en) * 2012-05-02 2012-09-19 武汉大学 Wavelet pyramid polarization texture primitive feature extracting method for synthetic aperture radar (SAR) images
CN103246891A (en) * 2013-05-28 2013-08-14 重庆邮电大学 Chinese sign language recognition method based on kinect
CN103854014A (en) * 2014-02-25 2014-06-11 中国科学院自动化研究所 Terror video identification method and device based on sparse representation of context
CN104200235A (en) * 2014-07-28 2014-12-10 中国科学院自动化研究所 Time-space local feature extraction method based on linear dynamic system
CN104199902A (en) * 2014-08-27 2014-12-10 中国科学院自动化研究所 Similarity measurement computing method of linear dynamical systems

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
GUAN LUO等: "Learning Human Actions by Combining Global Dynamics and Local Appearance", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 *
HAORAN WANG等: "Action recognition using linear dynamic systems", 《PATTERN RECOGNITION》 *
王茹等: "面向中国手语合成的视频语义描述方法", 《北京工业大学学报》 *

Similar Documents

Publication Publication Date Title
CN113936339B (en) Fighting identification method and device based on double-channel cross attention mechanism
CN106897675B (en) Face living body detection method combining binocular vision depth characteristic and apparent characteristic
CN101894276B (en) Training method of human action recognition and recognition method
CN107122752B (en) Human body action comparison method and device
CN105005772B (en) A kind of video scene detection method
CN102136066B (en) Method for recognizing human motion in video sequence
CN106295564B (en) A kind of action identification method of neighborhood Gaussian structures and video features fusion
CN107909104A (en) The face cluster method, apparatus and storage medium of a kind of picture
CN106651915B (en) The method for tracking target of multi-scale expression based on convolutional neural networks
CN104966081B (en) Spine image-recognizing method
CN103065158B (en) The behavior recognition methods of the ISA model based on relative gradient
Yang et al. A vehicle real-time detection algorithm based on YOLOv2 framework
CN106097381B (en) A kind of method for tracking target differentiating Non-negative Matrix Factorization based on manifold
CN103218609A (en) Multi-pose face recognition method based on hidden least square regression and device thereof
CN104063721A (en) Human behavior recognition method based on automatic semantic feature study and screening
CN108280421A (en) Human bodys' response method based on multiple features Depth Motion figure
CN108229432A (en) Face calibration method and device
CN113470076A (en) Multi-target tracking method for yellow-feather chickens in flat-breeding henhouse
Li et al. Augmenting bag-of-words: a robust contextual representation of spatiotemporal interest points for action recognition
CN104200235A (en) Time-space local feature extraction method based on linear dynamic system
Abdellaoui et al. Deep multi-stations weather forecasting: explainable recurrent convolutional neural networks
Guo et al. Foreign object detection of transmission lines based on faster R-CNN
Quan et al. An instance segmentation-based method to obtain the leaf age and plant centre of weeds in complex field environments
CN110969101A (en) Face detection and tracking method based on HOG and feature descriptor
CN104699781A (en) Specific absorption rate image retrieval method based on double-layer anchor chart hash

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20161214

WD01 Invention patent application deemed withdrawn after publication