CN103763515B - A kind of video abnormality detection method based on machine learning - Google Patents
A kind of video abnormality detection method based on machine learning Download PDFInfo
- Publication number
- CN103763515B CN103763515B CN201310722563.7A CN201310722563A CN103763515B CN 103763515 B CN103763515 B CN 103763515B CN 201310722563 A CN201310722563 A CN 201310722563A CN 103763515 B CN103763515 B CN 103763515B
- Authority
- CN
- China
- Prior art keywords
- image
- video
- judged
- frame
- color
- 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
Abstract
A kind of video abnormality detection method based on machine learning, comprises the following steps:1)Video file is read in:Video file is read in bmp view data one by one;2)Video abnormality detection:Process is as follows to be detected to the image decomposited:a)Picture is excessively bright, picture is excessively dark;b)Gain is disorderly;c)Obscure, be blocked;d)Ribbon interference, roll screen;e)Snowflake is disturbed;f)Shake;g)Colour cast;h)Freeze;i)Blue, blank screen;3)Machine learning:Abnormal video is will be deemed as, the video that the same type according to present in abnormality code and database is judged by accident carries out similarity-rough set, judges whether current video belongs to erroneous judgement situation.Present invention detection is comprehensive, higher with self-teaching improvement ability, the degree of accuracy.
Description
Technical field
The present invention relates to technical fields such as image procossing, video Similarity Measure and machine learning, main contents are one
Plant the video abnormality detection method based on machine learning.
Background technology
With the development and the rise of electronic applications of Chinese society, video monitoring system widely be applied to it is each
In each industry of row, the special dimensions such as public security, finance, bank, traffic, army and port in the past are no longer limited to, it is daily at us
Life tentacle energy and community, office building, hotel, public place, factory, market, cell, or even family has all been mounted with to regard
Frequency monitoring system.It is artificial real-time but with being continuously increased for application scenario CCTV camera quantity, the time of monitoring constantly extends
Maintenance to monitoring software is increasingly difficult to, so the intellectuality for monitoring software is also more and more urgent.
Existing monitor video detecting system is mostly based on hardware device detection, although the degree of accuracy is high, but does not possess flat
Platform transplantability, and functional module expands upgrading limitation greatly, while high expensive.And what existing Video Detection Algorithm can be detected
Exception Type is mostly not comprehensively, and for the in the case of of judging by accident does not possess the ability that self-teaching is remembered.
Video abnormality detection method proposed by the present invention based on machine learning can not only be realized different to most video
The detection of normal type, is also equipped with the ability to erroneous judgement situation self-teaching memory, it is to avoid the generation that next time equally judges by accident, and function
Autgmentability is good.Experiment proves that this method has very high accuracy and practicality.
The content of the invention
For existing video abnormality detection method it is still not comprehensive enough and lack self-teaching improve ability, the present invention propose
A kind of video abnormality detection method based on machine learning, this method reads video file, video is resolved into a frame first
Two field picture, is then detected to the view data decomposited;, can be according to exception class if user judges by accident in use
Type is stored as learning and memory by visual classification is judged by accident, is compared when occurring to call out when same type is abnormal again
It is right, prevent that identical erroneous judgement occurs again.
The technical scheme is that:
A kind of video abnormality detection method based on machine learning, it is characterised in that:The video abnormality detection method bag
Include following steps:
1) video file is read in:Resolve into a frame frame image data;
2) video abnormality detection:Process is as follows to be detected to the image decomposited:
A) picture is excessively bright, picture is excessively dark;
B) gain is disorderly;
C) obscure, be blocked;
D) ribbon interference, roll screen;
E) snowflake is disturbed;
F) shake;
G) colour cast;
H) freeze;
I) blue, blank screen;
3) machine learning:Abnormal video is will be deemed as, the same type according to present in abnormality code and database is judged by accident
Video carry out similarity-rough set, judge whether current video belongs to erroneous judgement situation.
Further, the method for detecting abnormality is further comprising the steps of:4) algorithm optimization:Set conflicting exception class
Type, need not repeat to detect for conflicting Exception Type.
The present invention technical concept be:This method reads video file first, video is resolved into a frame two field picture, then
The view data decomposited is detected;If judged by accident in use, visual classification can will be judged by accident according to Exception Type
Learning and memory is stored as, is compared when occurring to call out when same type is abnormal again, prevents again
Identical erroneous judgement.
Video abnormality detection method proposed by the present invention based on machine learning can not only be realized different to most video
The detection of normal type, is also equipped with the ability to erroneous judgement situation self-teaching memory, it is to avoid the generation that next time equally judges by accident, and function
Autgmentability is good.
Video abnormality detection method is the video file that monitoring system is passed back can be handled, detected, it, which can allow, makes
User is rapid, intuitively grasp the operation conditions of current monitor system, is easy to maintenance of the user to monitoring system.Deposited at present
Video abnormality detection method detect that Exception Type is not comprehensive mostly, and be unable to self-perfection when judging by accident, present invention fortune
The detection of most of Exception Types and the ability of self-teaching of method are realized with image procossing, video measuring similarity, is passed through
Lot of experiment validation, with very high accuracy rate and real-time.
Beneficial effects of the present invention are essentially consisted in:On the one hand the abnormality detection class that video abnormality detection method can be detected is made
Type is more fully;On the other hand machine learning ability is added in video abnormality detection method, will be deemed as abnormal video,
Similarity-rough set is carried out according to the video that abnormality code and same type are judged by accident, to reduce the situation of current video erroneous judgement, inspection is improved
Survey accuracy rate.
Brief description of the drawings
Fig. 1 is a kind of video abnormality detection method flow chart based on machine learning.
Fig. 2 is that colour cast judges coordinate diagram.
Embodiment
The invention will be further described below in conjunction with the accompanying drawings.
Referring to Figures 1 and 2, a kind of video abnormality detection method based on machine learning, the described method comprises the following steps:
1) video file is read in:Video file is read in into method with bmp view data one by one;
2) video abnormality detection:The image decomposited is detected, detectable Exception Type has
A) picture is excessively bright, picture is excessively dark:Image frame overall bright, darkness depend on the camera lens gray scale of image.It is right first
Image carries out gray processing processing f (x, y):
F (x, y)=0.114*B (x, y)+0.587*G (x, y)+0.299*R (x, y),
Wherein B (x, y), G (x, y), R (x, y) distinguish blueness of the correspondence image in (x, y) this pixel, green and red
Colouring component value.Luminance threshold light, darkness threshold values dark are set up, as f (x, y) > light, the point was designated as bright spot, worked as f
During (x, y) < dark, the point was designated as dim spot.The mistake bright spot in statistics video lens region, excessively dim spot number NUMlight,
NUMdark.Set up threshold values TαAnd TβIf, NUMlight > Tα, then present frame was bright frame;If NUMdark > Tβ, then when
Previous frame was spacer.
If a number of excessively bright frame continuously occur, crossing spacer, the video was judged as bright, excessively dark.
B) gain is disorderly:Color of image is disorderly, is judged using YUV color spaces.
Y=0.114*B+0.587*G+0.299*R
U=0.436*B-0.147*R-0.289*G
V=0.615*R-0.515*G-0.100*B
Y represents gray scale, and U components and V component determine color partially blue or partially red colourity in itself.The base of YUV color spaces
Present principles are to set up two-dimensional coordinate, using U as abscissa, and V is ordinate, and color is divided into 4 Ge Se areas, and the origin of coordinates represents image
Gray scale, the distance with originBigger color is deeper, sets up threshold values TDIf, d > TD, then (U, V) is that color is deep
Point.2*2 subregions are carried out to image, when the points of total color point deeply are more than image areaAnd each color area, picture portion
Color point deeply exist and when number is uniform, then present frame is the disorderly frame of gain.
If continuously there is the disorderly frame of a number of gain, the video is judged as that gain is disorderly.
C) obscure, be blocked:Two kinds of exceptions all show unintelligible for image, and gradient is less than normal, and difference is that fuzzy is overall
Property, it is local to be blocked, and the partial content that is blocked is partially dark.Convolution is carried out to image pixel f (x, y) and Sobel operators
Ask for the gradient map f in image x and y directionsx(x, y) and fy(x, y), then tries to achieve f (x, y) gradient | f (x, y) |:
| f (x, y) |=| fx(x,y)|+|fy(x,y)|
If threshold values TγIf, | f (x, y) | > Tγ, then it is designated as marginal point.Subregion is carried out to image, threshold values T is set upAIf,
The marginal point number of each subregion is both less than TA, and the average gray F of image meets Tβ< F < Tα, then it is judged as fuzzy frame.Such as
The marginal point number in fruit part region is less than TA, and part average gray F meets F < Tβ, other edges of regions point numbers are more than
TA, then it is judged as the frame that is blocked.
If continuously there is a number of fuzzy frame, be blocked frame, the video is judged as obscuring, is blocked.
D) ribbon interference, roll screen:Two kinds of exceptions can all produce interfering line in the picture, distinguish and to be disturbed with ribbon
Lines noise is varied less, and position is changed with picture rolling.The gradient of image is tried to achieve with step c) | f (x, y) |, according to threshold values pair
Image pixel point is marked, respectively marginal point and non-edge point.Marginal point number { N of the statistics per a linei, i=1,
2 ..., H and each row marginal point number { Mj, j=1,2 ..., W }, wherein H and W represent the height and width of image respectively.
When at least row of NiDuring > 0.6*W, then it is judged as that strip disturbs frame;As at least one row MjDuring > 0.6*H, then it is judged as band
Shape disturbs frame.When there is Ni> 0.9*W or Mj> 0.9*H, record maximum:
Or
Image lines location maxi or maxj are once compared every 3 frames, roll screen is set up and rolls threshold values TNAnd TM, when
| maxi-maxi'| > TNOr | maxj-maxj'| > TM
Wherein maxi' frames corresponding with maxi are separated by 3 frames, then are judged as roll screen frame.
If continuously there is a number of strip interference frame, banding interference frame, roll screen frame, the video is judged as strip
Interference, banding interference, roll screen.
E) snowflake is disturbed:Consecutive frame image change very little, snowflake interference picture is uniformly distributed noise, and position becomes at random
Change.First to image gray processing processing, pixel difference threshold values T is set upηIf, inter-pixel difference absolute value | fk(x,y)-fk+1(x,y)
| > Tη, then snowflake noise is designated as, when noise number is more than image areaWhen, it is understood that there may be snow noise.Image is divided into 16*
16 fritter, when all there is noise for every piece, is then judged as snowflake frame.
If continuously there is a number of snowflake frame, the video is judged as snowflake.
F) shake:Consecutive frame image change very little, but position rocks, if by consecutive frame picture registration, then normal picture
Substantially it is consistent, and ghost image, content increase occurs in dither image.The gradient map of image is first tried to achieve, if | | fk(x,y)|-
|fk+2(x, y) | | > Tκ, then (x, y) is designated as image and rocks profile point.Count present image rocks profile point n2And image
In edge points n1If, | n2-n1| more than image areaThen it is judged as shaking frame.
If continuously there is a number of shake frame, the video is judged as shake.
G) colour cast:RGB color structure is unfavorable for the judgement of colour cast, can be judged by accident in some special screnes, such as in image
Green tree leaf accounts for vast scale, it is easy to normal picture is judged as partially green.RGB color is transformed into Lab space to be examined
Survey.Lab colour models are made up of three key elements, and L represents brightness, and a represents the scope from carmetta to green, and b is represented from yellow
To the scope of blueness.
Image RGB is first converted into Lab space:
X=0.412453*r+0.357580*g+0.180423*b x=X/97.31
Y=0.212671*r+0.715160*g+0.072169*b y=Y/100
Z=0.019334*r+0.119193*g+0.950227*b z=Z/60.19
If x>0.008856, x=x1/3, otherwise x=7.787*x+16/116;
If y>0.008856, y=y1/3, otherwise y=7.787*y+16/116;
If z>0.008856, z=z1/3, otherwise z=7.787*z+16/116.
L=116*y-16;A=500* (z-y);B=200* (y-z);
Obtain image averaging colourity D, image chroma centre-to-centre spacing M:
The colour cast factorSpecific colour cast criterion is as shown in Figure 2.
H) freeze:Video image picture continuously fixes constant
When the pixel of continuous certain amount frame is the same, fk(x, y)=fk+1(x, y), is judged as freezing.
I) blue, blank screen:Screen picture is lacked, in unified blueness or black, and in centre, display is carried according to different cameras
Show the character of picture missing (general camera is blue, blank screen can all show " no signal ").
Statistics and the pixel number N of pixel f (10,10) equally, whenWhen, judgement may be lacked.Work as picture
R (10,10) < 20 in plain f (10,10), B (10,10) < 20, during G (10,10) < 20, are judged as blank screen frame;When B (10,10)
> R (10,10) and during B (10,10) > G (10,10), are judged as blue screen frame.
If continuously there is a number of blank screen frame, blue screen frame, the video is judged as blank screen, blue screen.3) machine
Study:Accuracy rate to improve system, reduces erroneous judgement.Will be deemed as abnormal video, according to abnormality code with it is once similar
The video of type erroneous judgement carries out similarity-rough set, judges whether current video belongs to erroneous judgement situation.
With reference to《Video measuring similarity》, the video features of extraction have color and the class of texture two, introduce video lens feature
Vectorial k(1,2,...35), vectorial k preceding 32 dimension is the colouring information component obtained according to hsv color statistics with histogram, and latter three are
Roughness, contrast, three components of directionality, compare the phase between the distance between two video lens centroid vectors calculating video
Like degree.
If camera lens s is made up of m key frame, s={ k1,k2,...,km, each key frame kiBy the feature of 35 dimensions
Vector representation:
k1={ F11,F12,...,F1N},…,km={ Fm1,Fm2,...,FmN, Q=35
It is as follows that characteristic vector similarity compares formula:
Claims (3)
1. a kind of video abnormality detection method based on machine learning, it is characterised in that:The method for detecting abnormality includes following
Step:
1) video is read in:Video file is read in bmp view data one by one;
2) video abnormality detection:Process is as follows to be detected to the image decomposited:
A) picture is excessively bright, the excessively dark detection process of picture:Gray processing processing is carried out to image first, gray-scale map, f (x, y) table is obtained
Show in gray-scale map the pixel value on (x, y) point, set up luminance threshold light, darkness threshold values dark, as f (x, y) > light,
The point was designated as bright spot, as f (x, y) < dark, and the point was designated as dim spot, the mistake bright spot in statistics video lens region, excessively dark
Point number NUMlight, NUMdark sets up threshold values TαAnd TβIf, NUMlight > Tα, then present frame was bright frame;If
NUMdark > Tβ, then present frame was spacer;
B) the disorderly detection process of gain:Color of image is disorderly, is judged using YUV color spaces, Y represents gray scale, U components
Color partially blue or partially red colourity in itself is determined with V component;The general principle of YUV color spaces is to set up two-dimensional coordinate, with U
For abscissa, V is ordinate, and color is divided into 4 Ge Se areas, threshold values T is set upDIf, d > TD, then (U, V) is the deep point of color;It is right
Image carries out 2*2 subregions, when the points of total color point deeply are more than the 1/2 of image area, and each color area, the face of picture portion
When color depth point is present and number is uniform, then present frame is the disorderly frame of gain, distanceBigger color is deeper;
C) obscure, be blocked detection process:The gradient map that convolution asks for image x and y directions is carried out to f (x, y) and Sobel operators
fx(x, y) and fy(x, y), then tries to achieve f (x, y) gradient, if threshold values TγIf, | f (x, y) | > Tγ, then it is designated as marginal point;
Subregion is carried out to image, threshold values T is set upAIf the marginal point number of each subregion is both less than TA, and the average gray F of image is full
Sufficient Tβ< F < Tα, then it is judged as fuzzy frame;If the marginal point number of subregion is less than TA, and part average gray F is full
Sufficient F < Tβ, other edges of regions point numbers are more than TA, then it is judged as the frame that is blocked;
D) ribbon interference, the detection process of roll screen:The gradient of image is tried to achieve with step c) | f (x, y) |, according to threshold values to image
Pixel is marked, respectively marginal point and non-edge point;Marginal point number N of the statistics per a lineiAnd the edge of each row
Point number Mj, i=1,2 ..., H, j=1,2 ..., W, wherein H representative graphs image height, W are that image is wide;When at least row of Ni>
During 0.6*W, then it is judged as that strip disturbs frame;As at least one row MjDuring > 0.6*H, then it is judged as that banding disturbs frame;When there is Ni
> 0.9*W or Mj> 0.9*H, record position max i or the max j of maximum, every 3 frames to image lines location max i
Or max j are once compared, set up roll screen and roll threshold values TNAnd TM, when | max i-max i'| > TNOr | max j-max j'
| > TM, then it is judged as roll screen frame, wherein max i' frames corresponding with max i are separated by 3 frames;
E) detection process of snowflake interference:First to image gray processing processing, pixel difference threshold values T is set upηIf, inter-pixel difference
Absolute value | fk(x,y)-fk+1(x, y) | > Tη, then snowflake noise is designated as, when noise number is more than the 1/5 of image area, Ke Nengcun
In snow noise, image being divided into 16*16 fritter, when all there is noise for every piece, being then judged as snowflake frame;
F) detection process of shake:The gradient map of image is first tried to achieve, if tried to achieve | | fk(x,y)|-|fk+2(x, y) | | >
Tκ, then (x, y) is designated as image and rocks profile point, count present image rocks profile points n2And the marginal point in image
Number n1If, | n2-n1| more than the 1/10 of image area, then it is judged as shaking frame;
G) detection process of colour cast is as follows:RGB color structure is unfavorable for the judgement of colour cast, can be judged by accident in some special screnes,
Such as image Green leaf accounts for vast scale, it is easy to normal picture is judged as partially green, and RGB color is transformed into Lab
Space is detected that Lab colour models are made up of three key elements, and L represents brightness, and a represents the scope from carmetta to green, b
The scope from yellow to blueness is represented, image RGB is first converted into Lab space, obtained in image averaging colourity D, image chroma
The heart is judged away from M, colour cast factor K=D/M according to colour cast standard;
H) detection process is freezed as follows:When the pixel of continuous certain amount frame is the same, fk(x, y)=fk+1(x, y), judges
To freeze;
3) machine learning:Abnormal video is will be deemed as, according to abnormality code and regarding that same type present in database is judged by accident
Frequency carries out similarity-rough set, judges whether current video belongs to erroneous judgement situation.
2. a kind of video abnormality detection method based on machine learning as claimed in claim 1, it is characterised in that:The exception
Detection method is further comprising the steps of:
Algorithm optimization, by setting conflicting Exception Type, need not repeat to detect for conflicting Exception Type.
3. a kind of video abnormality detection method based on machine learning as claimed in claim 1 or 2, it is characterised in that:Step
3) in, similarity-rough set process:The video features of extraction have color and the class of texture two, introduce video lens characteristic vector, vector
Preceding 32 dimension be the colouring information components that are obtained according to hsv color statistics with histogram, it is rear it is three-dimensional be roughness, contrast, direction
Property three components, compare the distance between two video lens centroid vectors calculate video between similarity.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310722563.7A CN103763515B (en) | 2013-12-24 | 2013-12-24 | A kind of video abnormality detection method based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310722563.7A CN103763515B (en) | 2013-12-24 | 2013-12-24 | A kind of video abnormality detection method based on machine learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103763515A CN103763515A (en) | 2014-04-30 |
CN103763515B true CN103763515B (en) | 2017-08-11 |
Family
ID=50530666
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310722563.7A Active CN103763515B (en) | 2013-12-24 | 2013-12-24 | A kind of video abnormality detection method based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103763515B (en) |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104301712B (en) * | 2014-08-25 | 2016-05-18 | 浙江工业大学 | A kind of monitoring camera blur detecting method based on video analysis |
CN106296634B (en) * | 2015-05-28 | 2019-11-15 | 腾讯科技(深圳)有限公司 | A kind of method and apparatus detecting similar image |
CN104967840B (en) * | 2015-06-19 | 2017-09-15 | 中央电视台 | A kind of exceeded detection method and device of video image |
CN105554494B (en) * | 2015-12-09 | 2018-09-11 | 浙江省公众信息产业有限公司 | Snow point image detecting method and device and video quality detection device and system |
CN105955884B (en) * | 2016-04-27 | 2018-11-20 | 上海携程商务有限公司 | App page white screen inspection method and device |
CN106303512B (en) * | 2016-08-19 | 2018-11-20 | 云赛智联股份有限公司 | video snow noise detection method |
CN108229262B (en) * | 2016-12-22 | 2021-10-15 | 腾讯科技(深圳)有限公司 | Pornographic video detection method and device |
CN106658172A (en) * | 2016-12-30 | 2017-05-10 | 广州虎牙信息科技有限公司 | Video image abnormity judgment method and device and terminal device |
CN106682639B (en) * | 2017-01-03 | 2019-05-14 | 安徽农业大学 | A kind of crop leaf portion abnormal image extracting method based on video surveillance |
AU2017414991B2 (en) * | 2017-05-17 | 2021-10-21 | Inaho, Inc. | Agricultural work apparatus, agricultural work management system, and program |
JP6298562B1 (en) * | 2017-05-31 | 2018-03-20 | 伸和コントロールズ株式会社 | Status monitoring device, status monitoring method and program |
CN108337505B (en) * | 2018-01-30 | 2020-11-10 | 百度在线网络技术(北京)有限公司 | Information acquisition method and device |
CN110827364B (en) * | 2018-08-07 | 2023-01-13 | 阿里巴巴(中国)有限公司 | Method and device for detecting green screen image |
CN109413411B (en) * | 2018-09-06 | 2020-08-11 | 腾讯数码(天津)有限公司 | Black screen identification method and device of monitoring line and server |
CN109299687A (en) * | 2018-09-18 | 2019-02-01 | 成都网阔信息技术股份有限公司 | A kind of fuzzy anomalous video recognition methods based on CNN |
CN109325533A (en) * | 2018-09-18 | 2019-02-12 | 成都网阔信息技术股份有限公司 | A kind of artificial intelligence frame progress CNN repetitive exercise method |
CN109905670B (en) * | 2019-03-02 | 2020-11-24 | 湛江广通科技有限公司 | Multistage platform monitoring system |
CN110298869A (en) * | 2019-07-08 | 2019-10-01 | 威创集团股份有限公司 | A kind of video analysis method and system |
CN111951244B (en) * | 2020-08-11 | 2024-03-01 | 北京百度网讯科技有限公司 | Method and device for detecting single-color screen in video file |
CN116055876A (en) * | 2021-10-27 | 2023-05-02 | 北京字跳网络技术有限公司 | Video processing method and device, electronic equipment and storage medium |
CN114120579A (en) * | 2021-11-26 | 2022-03-01 | 广东电网有限责任公司 | Intelligent well lid monitoring system, method, equipment and storage medium |
CN115914673A (en) * | 2022-01-27 | 2023-04-04 | 北京中关村科金技术有限公司 | Compliance detection method and device based on streaming media service |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101625721A (en) * | 2009-08-06 | 2010-01-13 | 安霸半导体技术(上海)有限公司 | Face detection and tracking method based on statistic data |
CN103165094A (en) * | 2013-03-07 | 2013-06-19 | 京东方科技集团股份有限公司 | Method and device of liquid crystal display |
CN103455795A (en) * | 2013-08-27 | 2013-12-18 | 西北工业大学 | Method for determining area where traffic target is located based on traffic video data image |
-
2013
- 2013-12-24 CN CN201310722563.7A patent/CN103763515B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101625721A (en) * | 2009-08-06 | 2010-01-13 | 安霸半导体技术(上海)有限公司 | Face detection and tracking method based on statistic data |
CN103165094A (en) * | 2013-03-07 | 2013-06-19 | 京东方科技集团股份有限公司 | Method and device of liquid crystal display |
CN103455795A (en) * | 2013-08-27 | 2013-12-18 | 西北工业大学 | Method for determining area where traffic target is located based on traffic video data image |
Non-Patent Citations (2)
Title |
---|
Real time anomaly detection in H.264 compressed videos;S. Biswas等;《Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG)》;20131221;第1-4页 * |
视频监控图像异常检测技术在安防行业中的应用;高旭麟;《中国公共安全》;20130615(第12期);第141-143页 * |
Also Published As
Publication number | Publication date |
---|---|
CN103763515A (en) | 2014-04-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103763515B (en) | A kind of video abnormality detection method based on machine learning | |
CN110135269B (en) | Fire image detection method based on mixed color model and neural network | |
CN104834912B (en) | A kind of weather recognition methods and device based on image information detection | |
CN105761261B (en) | A method of detection camera suffers artificial malicious sabotage | |
CN108133215B (en) | Processing unit | |
US8305440B2 (en) | Stationary object detection using multi-mode background modelling | |
US20200250840A1 (en) | Shadow detection method and system for surveillance video image, and shadow removing method | |
CN109918971B (en) | Method and device for detecting number of people in monitoring video | |
CN103281559A (en) | Method and system for detecting quality of video | |
CN104318266A (en) | Image intelligent analysis processing early warning method | |
CN109087363B (en) | HSV color space-based sewage discharge detection method | |
JP3486229B2 (en) | Image change detection device | |
CN107563985A (en) | A kind of detection method of infrared image moving air target | |
CN111047624A (en) | Image dim target detection method, device, equipment and storage medium | |
CN110223344A (en) | A kind of infrared small target detection method based on morphology and vision noticing mechanism | |
Bullkich et al. | Moving shadow detection by nonlinear tone-mapping | |
CN109858394A (en) | A kind of remote sensing images water area extracting method based on conspicuousness detection | |
CN110807406B (en) | Foggy day detection method and device | |
CN104954627B (en) | A kind of information processing method and electronic equipment | |
CN113744326B (en) | Fire detection method based on seed region growth rule in YCRCB color space | |
CN110728212A (en) | Road well lid monitoring device and method based on computer vision | |
CN109815784A (en) | A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium | |
CN112364884A (en) | Method for detecting moving object | |
WO2024016632A1 (en) | Bright spot location method, bright spot location apparatus, electronic device and storage medium | |
CN107452008A (en) | Method for detecting image edge and device |
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 |