CN106384092B - Online low-rank anomalous video event detecting method towards monitoring scene - Google Patents
Online low-rank anomalous video event detecting method towards monitoring scene Download PDFInfo
- Publication number
- CN106384092B CN106384092B CN201610814106.4A CN201610814106A CN106384092B CN 106384092 B CN106384092 B CN 106384092B CN 201610814106 A CN201610814106 A CN 201610814106A CN 106384092 B CN106384092 B CN 106384092B
- Authority
- CN
- China
- Prior art keywords
- frame
- video
- matrix
- anomalous
- low
- 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.)
- Active
Links
- 230000002547 anomalous effect Effects 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012544 monitoring process Methods 0.000 title claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims abstract description 95
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000012360 testing method Methods 0.000 claims abstract description 27
- 238000001514 detection method Methods 0.000 claims abstract description 21
- 238000013507 mapping Methods 0.000 claims abstract description 14
- 238000000605 extraction Methods 0.000 claims abstract description 13
- 238000010606 normalization Methods 0.000 claims abstract description 10
- 230000000007 visual effect Effects 0.000 claims abstract description 10
- 238000005457 optimization Methods 0.000 claims abstract description 9
- 238000010276 construction Methods 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 241000669244 Unaspis euonymi Species 0.000 claims description 3
- 238000002813 epsilometer test Methods 0.000 claims description 3
- 229940050561 matrix product Drugs 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 2
- 238000000354 decomposition reaction Methods 0.000 abstract description 3
- 230000005856 abnormality Effects 0.000 abstract 1
- 230000007123 defense Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010408 sweeping Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Abstract
The invention discloses a kind of online low-rank anomalous video event detecting method towards monitoring scene.The present invention proceeds as follows the monitor video set under given scenario: 1) monitor video being divided into training video and test video two parts, to each video frame extraction low-level visual feature, forming corresponding vectorization is indicated;2) by online sparse low-rank representation method, training video is learnt frame by frame with iterative gradient mapping ruler and random optimization criterion, the weight matrix and sparse coefficient matrix of update is obtained, constructs anomalous video event detection model;3) normalization reconstructed error is calculated to test video frame, if error is greater than given threshold, judging the frame, there are anomalous events, judge frame by frame like this, until video terminates.The present invention carries out frame coding and reconstruct to the video under monitoring scene from the angle of low-rank decomposition and rarefaction representation, can judge anomalous video event online frame by frame, improve the efficiency and precision of monitor video abnormality detection.
Description
Technical field
The invention belongs to Video Analysis Technology fields, the online low-rank anomalous video event inspection especially towards monitoring scene
Survey method.
Background technique
In recent years, the data that especially monitoring device obtains exponentially increase by all kinds of video capture devices,
These data are widely used in ensureing residential security, the work safety, traffic safety of general public, and the important of government aspect sets
The protection of the maintenance, emphasis institutional settings applied or even the defense military monitoring safety of State-level etc..Therefore, to different monitoring
Video data analysis and excavation under scene, the detection of especially anomalous event seem very urgent, academia and industry pair
This has carried out many researchs.Due to the light of monitoring scene often changes, target trajectory is different and external disturbance etc. because
Element, so that monitor video not only has unstructured, sweeping feature, but also there are motion features unapparent, all kinds of to make an uproar
The features such as acoustic jamming is more bring very big difficulty to the detection of anomalous video event.
There are some problems for traditional anomalous video event detecting method, such as only solve specific abnormal under some scene
Type (traffic lane change, parking offense, article lose), can not real-time detection anomalous event etc..Therefore, occur being suitable for some
The method of a variety of accident detections under scene, such as simple and effective frame difference method, sparse reconstructing method, but these methods are simultaneously
Not in view of low-rank characteristic existing for video interframe, thus it is special in the structure of lower dimensional space to characterize video data well
Sign.
Summary of the invention
In order to be detected in real time to the anomalous video event under monitoring scene, from the angle of low-rank decomposition and rarefaction representation
It spends and frame coding and reconstruct is carried out to the video under monitoring scene, the invention proposes a kind of, and the online low-rank towards monitoring scene is different
Normal Video Events detection method, method includes the following steps:
1, it after obtaining the monitor video set under given scenario, proceeds as follows:
1) monitor video is divided into training video and test video two parts, wherein training video is by selected normal video
Composition, and to each video frame extraction low-level visual feature, forming corresponding vectorization indicates.
2) quasi- with iterative gradient mapping ruler and random optimization to training video by online sparse low-rank representation method
Then learnt frame by frame, obtain the weight matrix and sparse coefficient matrix of update, constructs anomalous video event detection model.
3) normalization reconstructed error is calculated to test video frame and judges that there are different for the frame if error is greater than given threshold
Ordinary affair part, judges frame by frame like this, until video terminates.
Further, in the step 1) to each video frame extraction low-level visual feature, form corresponding vectorization table
Show, specifically:
1.1) to three kinds of low-level visual features of each video frame extraction, i.e. Scale invariant features transform (SIFT), part two
Value mode (LBP), histogram of gradients (HOG), portray the global and local structure in frame image, and three kinds of features of extraction are merged
The column vector tieed up at m
1.2) vectorization representing matrix is formed to n frame training video imageS frame is tested
Video image forms vectorization representing matrixWherein the representing matrix X of training video frame is as different
The dictionary of ordinary affair part detection model.
Further, pass through online sparse low-rank representation method in the step 2), to training video with iteration ladder
Degree mapping ruler and random optimization criterion are learnt frame by frame, obtain the weight matrix and sparse coefficient matrix of update, are constructed different
Normal Video Events detection model, specifically:
2.1) the representing matrix X of training video is decomposed into two parts, i.e. X=XWV+E: front is divided by dictionary X and power
The p Wiki matrix that weight matrix W structure linear reconstruction obtainsWith low-dimensional coefficient matrixProduct matrix,
Rear portion is divided into noise matrixThe dimension of wherein p < < min (m, n), i.e. basic matrix and coefficient matrix are much smaller than m and n
Minimum value, the corresponding p of each m dimension training frames x maintains number vector and is expressed as v, and corresponding m dimension noise vector is e.
2.2) online coefficient low-rank representation method, which refers to, adds Frobenius norm to weight matrix W | | | |F, to being
Matrix number V adds Frobenius norm simultaneously | | | |FAnd l1Norm | | | |1Sparse characteristic is made it have, to noise matrix E
Add l1Norm, is minimizing the mean square error between X and (XWV+E) under the constraint condition of above-mentioned norm, objective function is
Wherein, constant λ1> 0, λ2>0。
2.3) training video is learnt frame by frame with iterative gradient mapping ruler and random optimization criterion, is updated
Weight matrix W and sparse coefficient matrix V refer to t (t=0,1 ..., n) take turns iteration in the case of, first initialize weight square
Battle array W0, sparse coefficient matrix V0, noise matrix E0For full 0 matrix, and introduce companion matrixWithAlso initial
Full 0 matrix is turned to, subscript represents iteration wheel number, and specific step is as follows for iteration:
A) a frame x is selected at random from training video frame settCarry out model construction.
B) take turns iteration in t, fixed weight matrix and noise vector, by solve following formula obtain the coefficient of the wheel iteration to
Measure vt, i.e.,
Wherein symbol | | | |2Indicate the l of vector2Norm.
C) take turns iteration in t, fixed weight matrix and coefficient vector, by solve following formula obtain the noise of the wheel iteration to
Measure et, i.e.,
D) two companion matrixs are updated by following formula, i.e.,
Wherein symbol ()TIndicate the transposition of vector or matrix.
E) weight matrix W is updated using iterative gradient mapping ruler, i.e., gradient is calculated to matrix column vector, and by reflecting
It penetrates to obtain the new expression of each column vector.
F) it repeats the above steps, until training video frame set becomes empty set, and obtains the weight updated by n times iteration
Matrix W and sparse coefficient matrix V.
2.4) building anomalous video event detection model refers to video frame x to be detectedtestIt carries out in step 2.3)
B) after and c) operating, corresponding coefficient vector v can be obtainedtestWith noise vector etest, then calculate two reconstructed error value fvWith
fe, i.e.,
So far anomalous video event detection model construction is completed.
Normalization reconstructed error is calculated to test video frame in the step 3), if error is greater than given threshold,
Judging the frame, there are anomalous events, judge frame by frame like this, specifically:
3.1) normalization reconstructed error is calculated to test video frameIt is set if reconstructed error err is greater than
Fixed normal number threshold value Θ, then judge that there are anomalous events for the video frame.
3.2) step 3.1) is successively repeated to all test video frames frame by frame, that is, can determine whether different in all test videos
Ordinary affair part.
The invention proposes the online low-rank anomalous video event detecting methods towards monitoring scene, and the advantage is that can
Coding video frames and data reconstruction are carried out from the angle of low-rank decomposition and rarefaction representation to the video under monitoring scene, it can be frame by frame
Judge whether anomalous event occur in video in real time, improve the accuracy and efficiency of the judgement of monitor video anomalous event,
The special scenes such as safety monitoring for public transport, emphasis institutional settings, defense military strategic point provide good technical support.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention.
Specific embodiment
Below in conjunction with attached drawing, the invention will be further described.
The present invention proposes a kind of online anomalous video event detecting method towards monitoring scene, not only considers video data
Sparsity structure, it is also contemplated that the low-rank structure of data.Main thought is to introduce reconstructed error mechanism, is changed with normal video data
Generation training dictionary and basic matrix, then detect reconstructed error whether in reasonable value range to new video frame frame by frame.Pass through this
Kind mode can carry out accident detection to endlessly video data in real time, referring to attached drawing 1, further illustrate:
1, it after obtaining the monitor video set under given scenario, performs the following operation:
1) monitor video is divided into training video and test video two parts, wherein training video is by selected normal video
Composition, and to each video frame extraction low-level visual feature, forming corresponding vectorization indicates.
2) quasi- with iterative gradient mapping ruler and random optimization to training video by online sparse low-rank representation method
Then learnt frame by frame, obtain the weight matrix and sparse coefficient matrix of update, constructs anomalous video event detection model.
3) normalization reconstructed error is calculated to test video frame and judges that there are different for the frame if error is greater than given threshold
Ordinary affair part, judges frame by frame like this, until video terminates.
To each video frame extraction low-level visual feature described in step 1), forming corresponding vectorization is indicated, specifically:
1.1) to three kinds of low-level visual features of each video frame extraction, i.e. Scale invariant features transform (SIFT), part two
Value mode (LBP), histogram of gradients (HOG), portray the global and local structure in frame image, and three kinds of features of extraction are merged
The column vector tieed up at m
1.2) vectorization representing matrix is formed to n frame training video imageS frame is tested
Video image forms vectorization representing matrixWherein the representing matrix X of training video frame is as different
The dictionary of ordinary affair part detection model.
In step 2) by online sparse low-rank representation method, to training video with iterative gradient mapping ruler and with
Machine Optimality Criteria is learnt frame by frame, obtains the weight matrix and sparse coefficient matrix of update, constructs anomalous video event detection
Model, specifically:
2.1) the representing matrix X of training video is decomposed into two parts, i.e. X=XWV+E: front is divided by dictionary X and power
The p Wiki matrix that weight matrix W structure linear reconstruction obtainsWith low-dimensional coefficient matrixProduct moment
Battle array, rear portion is divided into noise matrixThe dimension of wherein p < < min (m, n), i.e. basic matrix and coefficient matrix are much smaller than m
With the minimum value of n, the corresponding p of each m dimension training frames x maintains number vector and is expressed as v, and corresponding m dimension noise vector is e.
2.2) online coefficient low-rank representation method, which refers to, adds Frobenius norm to weight matrix W | | | |F, to being
Matrix number V adds Frobenius norm simultaneously | | | |FAnd l1Norm | | | |1Sparse characteristic is made it have, to noise matrix E
Add l1Norm, is minimizing the mean square error between X and (XWV+E) under the constraint condition of above-mentioned norm, objective function is
Wherein, constant λ1> 0, λ2>0;
2.3) training video is learnt frame by frame with iterative gradient mapping ruler and random optimization criterion, is updated
Weight matrix W and sparse coefficient matrix V refer to t (t=0,1 ..., n) take turns iteration in the case of, first initialize weight square
Battle array W0, sparse coefficient matrix V0, noise matrix E0For full 0 matrix, and introduce companion matrixWithAlso initial
Full 0 matrix is turned to, subscript represents iteration wheel number, and specific step is as follows for iteration:
A) a frame x is selected at random from training video frame settCarry out model construction.
B) take turns iteration in t, fixed weight matrix and noise vector, by solve following formula obtain the coefficient of the wheel iteration to
Measure vt, i.e.,
Wherein symbol | | | |2Indicate the l of vector2Norm.
C) take turns iteration in t, fixed weight matrix and coefficient vector, by solve following formula obtain the noise of the wheel iteration to
Measure et, i.e.,
D) two companion matrixs are updated by following formula, i.e.,
Wherein symbol ()TIndicate the transposition of vector or matrix.
E) weight matrix W is updated using iterative gradient mapping ruler, i.e., gradient is calculated to matrix column vector, and by reflecting
It penetrates to obtain the new expression of each column vector.
F) it repeats the above steps, until training video frame set becomes empty set, and obtains the weight updated by n times iteration
Matrix W and sparse coefficient matrix V.
2.4) building anomalous video event detection model refers to video frame x to be detectedtestIt carries out in step 2.3)
B) after and c) operating, corresponding coefficient vector v can be obtainedtestWith noise vector etest, then calculate two reconstructed error value fvWith
fe, i.e.,
So far anomalous video event detection model construction is completed.
Normalization reconstructed error is calculated to test video frame in step 3), if error is greater than given threshold, judgement should
There are anomalous events for frame, judge frame by frame like this, specifically:
3.1) normalization reconstructed error is calculated to test video frameIt is set if reconstructed error err is greater than
Fixed normal number threshold value Θ, then judge that there are anomalous events for the video frame.
3.2) step 3.1) is successively repeated to all test video frames frame by frame, that is, can determine whether different in all test videos
Ordinary affair part.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention
Range should not be construed as being limited to the specific forms stated in the embodiments, and protection scope of the present invention is also and in this field skill
Art personnel conceive according to the present invention it is conceivable that equivalent technologies mean.
Claims (2)
1. the online low-rank anomalous video event detecting method towards monitoring scene, it is characterised in that the monitoring under given scenario
Video collection proceeds as follows:
1) monitor video is divided into training video and test video two parts, wherein training video is by selected normal video group
At, and to each video frame extraction low-level visual feature, forming corresponding vectorization indicates;
2) by online sparse low-rank representation method, to training video with iterative gradient mapping ruler and random optimization criterion into
Row learns frame by frame, obtains the weight matrix and sparse coefficient matrix of update, constructs anomalous video event detection model;
3) normalization reconstructed error is calculated to test video frame, if error is greater than given threshold, judges that the frame has abnormal thing
Part judges frame by frame like this, until video terminates;
In the step 1) to each video frame extraction low-level visual feature, forming corresponding vectorization indicates, specifically:
1.1) to three kinds of low-level visual features of each video frame extraction, i.e. Scale invariant features transform, local binary patterns and gradient
Histogram portrays the global and local structure in frame image, and three kinds of features of extraction are merged into the column vector of m dimension
1.2) vectorization representing matrix is formed to n frame training video imageTo s frame test video
Image forms vectorization representing matrixWherein the representing matrix X of training video frame is as abnormal thing
The dictionary of part detection model;
In the step 2) by online sparse low-rank representation method, to training video with iterative gradient mapping ruler and
Random optimization criterion is learnt frame by frame, obtains the weight matrix and sparse coefficient matrix of update, building anomalous video event inspection
Model is surveyed, specifically:
2.1) the representing matrix X of training video is decomposed into two parts, i.e. X=XWV+E: front is divided by dictionary X and weight square
The p Wiki matrix that battle array W structure linear reconstruction obtainsWith low-dimensional coefficient matrixProduct matrix, rear portion
It is divided into noise matrixThe dimension of wherein p < < min (m, n), i.e. basic matrix and coefficient matrix are most much smaller than m and n
Small value, the corresponding p of each m dimension training frames x maintain number vector and are expressed as v, and corresponding m dimension noise vector is e;
2.2) online coefficient low-rank representation method, which refers to, adds Frobenius norm to weight matrix W | | | |F, to coefficient matrix
V adds Frobenius norm simultaneously | | | |FAnd l1Norm | | | |1Sparse characteristic is made it have, l is added to noise matrix E1
Norm, is minimizing the mean square error between X and (XWV+E) under the constraint condition of above-mentioned norm, objective function is
Wherein, constant λ1> 0, λ2>0;
2.3) training video is learnt frame by frame with iterative gradient mapping ruler and random optimization criterion, obtains the power of update
Weight matrix W and sparse coefficient matrix V refer in the case of t (t=0,1 ..., n) takes turns iteration, first initialize weight matrix W0、
Sparse coefficient matrix V0, noise matrix E0For full 0 matrix, and introduce companion matrixWithAlso it is initialized as
Full 0 matrix, subscript represent iteration wheel number, and specific step is as follows for iteration:
A) a frame x is selected at random from training video frame settCarry out model construction;
B) iteration, fixed weight matrix and noise vector are taken turns in t, the coefficient vector v of the wheel iteration is obtained by solving following formulat,
I.e.
Wherein symbol | | | |2Indicate the l of vector2Norm;
C) iteration, fixed weight matrix and coefficient vector are taken turns in t, the noise vector e of the wheel iteration is obtained by solving following formulat,
I.e.
D) two companion matrixs are updated by following formula, i.e.,
Wherein symbol ()TIndicate the transposition of vector or matrix;
E) weight matrix W is updated using iterative gradient mapping ruler, i.e., gradient is calculated to matrix column vector, and by mapping
To the new expression of each column vector;
F) it repeats the above steps, until training video frame set becomes empty set, and obtains the weight matrix updated by n times iteration
W and sparse coefficient matrix V;
2.4) building anomalous video event detection model refers to video frame x to be detectedtestCarry out the b in step 2.3)) and c)
After operation, corresponding coefficient vector v is obtainedtestWith noise vector etest, then calculate two reconstructed error value fvAnd fe, i.e.,
So far anomalous video event detection model construction is completed.
2. the online low-rank anomalous video event detecting method towards monitoring scene as described in claim 1, it is characterised in that:
Normalization reconstructed error is calculated to test video frame in the step 3), if error judges the frame greater than given threshold
There are anomalous events, judge frame by frame like this, specifically:
3.1) normalization reconstructed error is calculated to test video frameIf reconstructed error err is greater than setting
Normal number threshold value Θ then judges that there are anomalous events for the video frame;
3.2) step 3.1) is successively repeated to all test video frames frame by frame, that is, can determine whether the abnormal thing in all test videos
Part.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610814106.4A CN106384092B (en) | 2016-09-11 | 2016-09-11 | Online low-rank anomalous video event detecting method towards monitoring scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610814106.4A CN106384092B (en) | 2016-09-11 | 2016-09-11 | Online low-rank anomalous video event detecting method towards monitoring scene |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106384092A CN106384092A (en) | 2017-02-08 |
CN106384092B true CN106384092B (en) | 2019-04-26 |
Family
ID=57936378
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610814106.4A Active CN106384092B (en) | 2016-09-11 | 2016-09-11 | Online low-rank anomalous video event detecting method towards monitoring scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106384092B (en) |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107392100B (en) * | 2017-06-17 | 2020-07-07 | 复旦大学 | Detection method for automatically detecting local abnormality in monitoring video |
CN107967440B (en) * | 2017-09-19 | 2021-03-30 | 北京工业大学 | Monitoring video abnormity detection method based on multi-region variable-scale 3D-HOF |
CN109753851B (en) * | 2017-11-03 | 2022-08-09 | 郑州大学 | Anomaly detection method and system |
CN108304802B (en) * | 2018-01-30 | 2020-05-19 | 华中科技大学 | Rapid filtering system for large-scale video analysis |
CN108549857B (en) * | 2018-03-30 | 2021-04-23 | 国信优易数据股份有限公司 | Event detection model training method and device and event detection method |
CN108805002B (en) * | 2018-04-11 | 2022-03-01 | 杭州电子科技大学 | Monitoring video abnormal event detection method based on deep learning and dynamic clustering |
CN109117774B (en) * | 2018-08-01 | 2021-09-28 | 广东工业大学 | Multi-view video anomaly detection method based on sparse coding |
CN109145841A (en) * | 2018-08-29 | 2019-01-04 | 武汉大学 | A kind of detection method and device of the anomalous event based on video monitoring |
CN109272035A (en) * | 2018-09-12 | 2019-01-25 | 深圳市唯特视科技有限公司 | A kind of video rapid inference method based on circulation residual error module |
CN109447263B (en) * | 2018-11-07 | 2021-07-30 | 任元 | Space abnormal event detection method based on generation of countermeasure network |
CN110414598A (en) * | 2019-07-26 | 2019-11-05 | 国家消防工程技术研究中心 | Smog detection method, device, computer and storage medium |
CN110674790B (en) * | 2019-10-15 | 2021-11-23 | 山东建筑大学 | Abnormal scene processing method and system in video monitoring |
CN113836976A (en) * | 2020-06-23 | 2021-12-24 | 江苏翼视智能科技有限公司 | Method for detecting global abnormal event in surveillance video |
CN112380915A (en) * | 2020-10-21 | 2021-02-19 | 杭州未名信科科技有限公司 | Method, system, equipment and storage medium for detecting video monitoring abnormal event |
CN113011399B (en) * | 2021-04-28 | 2023-10-03 | 南通大学 | Video abnormal event detection method and system based on generation cooperative discrimination network |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103023927A (en) * | 2013-01-10 | 2013-04-03 | 西南大学 | Method and system for intrusion detection based on non-negative matrix factorization under sparse representation |
CN105335653A (en) * | 2014-07-21 | 2016-02-17 | 华为技术有限公司 | Abnormal data detection method and apparatus |
CN105427300A (en) * | 2015-12-21 | 2016-03-23 | 复旦大学 | Low-rank expression and learning dictionary-based hyperspectral image abnormity detection algorithm |
-
2016
- 2016-09-11 CN CN201610814106.4A patent/CN106384092B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103023927A (en) * | 2013-01-10 | 2013-04-03 | 西南大学 | Method and system for intrusion detection based on non-negative matrix factorization under sparse representation |
CN105335653A (en) * | 2014-07-21 | 2016-02-17 | 华为技术有限公司 | Abnormal data detection method and apparatus |
CN105427300A (en) * | 2015-12-21 | 2016-03-23 | 复旦大学 | Low-rank expression and learning dictionary-based hyperspectral image abnormity detection algorithm |
Non-Patent Citations (3)
Title |
---|
Anomaly Detection in Hyperspectral Images Based on Low-Rank and Sparse Representation;Yang Xu 等;《IEEE Transactions on Geoscience and Remote Sensing》;20151109;第54卷(第4期);第1900-2000页 * |
Towards robust subspace recovery via sparsity-constrained latent low-rank representation;Ping Li 等;《Journal of Visual Communication and Image Representation》;20150625;第37卷;第46-52页 * |
基于HOG3D描述器与稀疏编码的异常行为检测方法;何聪芹 等;《华东理工大学学报(自然科学版)》;20160229;第42卷(第1期);第110-118页 * |
Also Published As
Publication number | Publication date |
---|---|
CN106384092A (en) | 2017-02-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106384092B (en) | Online low-rank anomalous video event detecting method towards monitoring scene | |
Kim et al. | Towards a rigorous evaluation of time-series anomaly detection | |
CN109359519A (en) | A kind of video anomaly detection method based on deep learning | |
CN103761748B (en) | Anomaly detection method and device | |
CN104599292B (en) | A kind of anti-noise moving object detection algorithm decomposed based on low-rank matrix | |
CN107967440B (en) | Monitoring video abnormity detection method based on multi-region variable-scale 3D-HOF | |
CN101996327B (en) | Video anomaly detection method based on weighted tensor subspace background modeling | |
CN106663316A (en) | Block sparse compressive sensing-based infrared image reconstruction method and system thereof | |
CN105374026B (en) | A kind of detection method of marine infrared small target suitable for coast defence monitoring | |
CN110361778A (en) | A kind of Reconstruction of seismic data method based on generation confrontation network | |
CN108596108A (en) | Method for detecting change of remote sensing image of taking photo by plane based on the study of triple semantic relation | |
CN104751472A (en) | Fabric defect detection method based on B-spline wavelets and deep neural network | |
CN104899567A (en) | Small weak moving target tracking method based on sparse representation | |
CN104820824A (en) | Local abnormal behavior detection method based on optical flow and space-time gradient | |
CN105354542B (en) | A kind of video accident detection method under crowd scene | |
Lyudchik | Outlier detection using autoencoders | |
Zhang et al. | Automatic recognition of oil industry facilities based on deep learning | |
CN102867195A (en) | Method for detecting and identifying a plurality of types of objects in remote sensing image | |
CN105678047A (en) | Wind field characterization method with empirical mode decomposition noise reduction and complex network analysis combined | |
Li et al. | Cracked insulator detection based on R-FCN | |
CN104156979B (en) | Deviant Behavior online test method in a kind of video based on gauss hybrid models | |
CN103473559A (en) | SAR image change detection method based on NSCT domain synthetic kernels | |
CN103268484A (en) | Design method of classifier for high-precision face recognitio | |
Zhang et al. | I-MMCCN: Improved MMCCN for RGB-T crowd counting of drone images | |
CN103693532A (en) | Method of detecting violence in elevator car |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20231019 Address after: No. 508-2A, Baoli Tianji North Block, Qiandenghu, Guicheng Street, Nanhai District, Foshan City, Guangdong Province, 528200 Patentee after: Foshan Haixie Technology Co.,Ltd. Address before: 310018 No. 2 street, Xiasha Higher Education Zone, Hangzhou, Zhejiang Patentee before: HANGZHOU DIANZI University |