CN106296677A - A kind of remnant object detection method of double mask context updates based on double-background model - Google Patents
A kind of remnant object detection method of double mask context updates based on double-background model Download PDFInfo
- Publication number
- CN106296677A CN106296677A CN201610638539.9A CN201610638539A CN106296677A CN 106296677 A CN106296677 A CN 106296677A CN 201610638539 A CN201610638539 A CN 201610638539A CN 106296677 A CN106296677 A CN 106296677A
- Authority
- CN
- China
- Prior art keywords
- image
- background
- background model
- double
- mask
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The invention discloses the remnant object detection method of a kind of double mask context updates based on double-background model, relate to intelligent monitoring, computer vision field.This method comprises the following steps: S1: reads in monitor video, and creates two background models；S2: detection static foreground object；S3: the stationary object detected is screened；The renewal of S4: double-background model；S5: the legacy detected labelling in addition is exported video monitoring.The present invention contrast by double-background model, detect stationary object roughly, obtain legacy accurately through the most accurate screening, the update method in two background models further ensures the accuracy of legacy detection, also assures that the real-time of detection simultaneously.And also can well adapt to various public arena and the interference effectively avoiding environmental change (such as light, the swing of wind object) to produce.
Description
Technical field
The present invention relates to intelligent video monitoring, computer vision field, be specifically related to a kind of based on double-background model double
The remnant object detection method of mask context update.
Background technology
Video monitoring system is obtained in all trades and professions and is widely applied, particularly in residential quarter, bank, supermarket,
There is video monitoring equipment in subway, airport, museum etc..Generally, above monitoring system is mainly by traditional closed circuit electricity
Constitute depending on CCTV monitoring, monitoring scene can be recorded and store, but can not be timely to criminal behaviors such as public safeties
Send warning, and need substantial amounts of staff constantly to monitor monitored picture.Therefore the video monitoring system based on CCTV
Cannot meet modern people's demand to safety precaution, thus intelligent video monitoring has arisen at the historic moment, and will take the most comprehensively
For the former.
Legacy detection is an important branch of intelligent video monitoring in security protection early warning field, on airport, subway, physical culture
The suspicious legacy detection of the public arenas such as field, waiting room and exhibition center is the indispensable content of intelligent video monitoring system.
Current existing remnant object detection method is primarily present following problem:
The most directly use mixed Gauss model that two backgrounds are modeled simultaneously, need the amount of calculation run the hugest,
It is difficult to meet the requirement of real-time.The long-time stop of legacy is easy to be updated to inside background at present, thus causes nothing
Method detects legacy；Legacy detection at present does not carry out strict rejecting to chaff interference, causes legacy false drop rate pole
High.
So how improving in video monitoring system, legacy detects under complex environment accuracy, real-time is to need
The problem solved.
Summary of the invention
In order to solve above-mentioned technical problem, the purpose of the present invention is that the accuracy improving legacy detection is with in real time
Property, it is provided that the remnant object detection method of a kind of double mask context updates based on double-background model.
The technical solution adopted in the present invention is:
S10, reads in monitor video, and creates two background models, updates using the first two field picture of monitor video as fast
Background model and the slow background image updating background model；
S20, detects static foreground object；
S30, screens the stationary object detected；
S40, the renewal of double-background model；
S50, exports video monitoring by the legacy detected labelling in addition；
Above-mentioned technical scheme, the most preferably, creates two background models and comprises the steps: in described step S10
S11, two background models are respectively fast renewal background model and update background model slowly, both not only backgrounds
Renewal rate is different, and their context update mechanism is the most different, and the fast background model that updates is to enter based on pixel relative method
Row context update, the slow background model that updates carries out context update based on mixed Gaussian background modeling.
Described step S20 detects in static foreground object and comprises the steps:
S21, carries out difference by the fast background model that updates with the slow background model that updates, indicates by two-value, if difference is more than threshold
Value, then be set to 255 at bianry image corresponding point position, and 255 signs differ, and i.e. indicate resting；Otherwise, then at bianry image pair
Should be set to 0 in the place of putting, 0 sign is identical, i.e. indicates nonstatic thing.
S22, is tracked timing, each point of scanning stationary object image, finds pixel value the most static object
Being the point of 255, add 1 at accumulative image corresponding point counting, then each point of the accumulative image of scanning, finds counting more than setting threshold
The pixel of value, forms the bianry image of new stationary object；
S23, carries out the expansive working of morphologic filtering to new bianry image.
In described step S30, stationary object carries out screening and comprises the steps:
S31, first filters out the stationary object of a suitable size, filters out excessive or too small stationary object.
S32, the length-width ratio first finding out boundary rectangle meets the stationary object that body shape requires, then utilizes HOG+SVM
Pedestrian detection algorithm detects whether these objects are static pedestrians.
S33, with the bianry image of moving object and the bianry image of stationary object with obtain new bianry image, at this
The interference of the swing of the wind objects such as medium filtering carrys out leaves outside filter chamber, Semen Salicis babylonicae cum pilus waves is used on bianry image.
S34 is static less than filtering out of the ratio set by the contour area of stationary object and boundary rectangle area ratio
Bicycle and the interference of electromobile.
Above-mentioned technical scheme, wherein, the renewal following steps of double-background model in described step S40:
S41, the fast background model that updates utilizes the every frame data read in and current background to compare on each pixel
Gray value size is carried out updating decision and is updated background model.
S42, first the slow background model that updates constructs two masks, along with in video monitoring, situation with or without legacy is continuous
Change, utilizes the two mask can switch the mixed Gaussian background in the slow overall situation/local background updating background model at any time
Modeling updates.
Preferably, described fast renewal background model utilizes the every frame video image data read in carry out each with current background
The updating decision that relatively comes of the gray value size on pixel updates background model, if the gray value of current video image ratio is current
The corresponding point position of background image is big, then the corresponding point gray value of current background image adds 1, whereas if current video image
Gray value is less than current background image corresponding point position, then current background image corresponding point position subtracts 1.
Preferably, the update method of described slow renewal background model is:
1) two masks of structure, first mask be gray value be all the image of 255, second mask is first mask
Result images of the inverted；
2) if monitor video does not occur legacy, update background model the most slowly and carry out overall situation mixed Gaussian background more
Newly, and two mask image information keep constant；Legacy once detected, then legacy position storage on image
Get off, then on the corresponding position in first mask image, the gray value of this part is all set to 0, covers for second
First mask image is re-started and negates renewal by film image；
3) first mask image and the current slow background area updating background image and obtain needing renewal, second
The background area that need not update that individual mask image is corresponding with obtaining legacy with current slow renewal background image；
4) first mask image and current video image with obtain needing the mixed Gaussian background in context update region
The input data image updated；
5) finally the background area image updated with need not update background area image phase or obtain complete
Completely newly update background image slowly.
The following several good effects of remnant object detection method generation of the present invention:
Effect 1: the present invention uses pixel relative method that the fast background model that updates is carried out context update, thus replaces mixing height
The background update method of this background modeling, reduces the amount of calculation that full HD video image analysis processes, and improves detection algorithm
Real-time.
Effect 2: the method that the present invention filters static pedestrian with HOG+SVM human body detecting method after using first screening, very
Decrease the amount of calculation of HOG+SVM human body detecting method in big degree, improve the real-time of detection algorithm.
Effect 3: the present invention uses double mask that the slow background model that updates is carried out local updating, is not the most leaving over
Switch freely between background local updating when background overall situation when thing occurs updates and has legacy to occur, thus avoid losing
The leaving over overlong time of thing is stayed to be dissolved into the slow phenomenon updated in background model.
Effect 4: the present invention uses strict screening technique, eliminates jamming target (noise, static pedestrian, wind object
Swing, stationary bike) impact on legacy, improve the accuracy of algorithm, reduce missing inspection and false drop rate as far as possible.
Accompanying drawing explanation
Fig. 1 is the block diagram of this method；
Fig. 2 is the flow chart calculating stationary object；
Fig. 3 is the flow chart that stationary object follows the tracks of timing；
Fig. 4 is the flow chart filtering the interference that wind object swings；
Fig. 5 is the context update flow chart updating background model soon；
Fig. 6 is the double mask context update flow charts updating background model slowly.
Detailed description of the invention
Below in conjunction with accompanying drawing and be embodied as describing the present invention, but not as a limitation of the invention.
Such as Fig. 1, this method to realize step as follows:
A reads monitor video, and creates two background models
Photographic head is utilized to obtain monitor video view data, and using the first two field picture as at the beginning two background models
Background image.
B detects static foreground object
In previous step, we utilize vedio data, establish two background models, the difference of two background models it
Place is context update speed and background update method, is designated as respectively updating background soon and updating background slowly, updates the back of the body the soonest
The renewal rate of scape model is 0.5 second/time, and the slow renewal rate updating background model is 25 seconds/time.Fast update background model with
The renewal rate of slow renewal background model is proper between 1:40 to 1:80, and this ratio can be according to concrete actual rings
Border sets.
As in figure 2 it is shown, two background models are carried out difference, and indicate by two-value, if difference is more than threshold value 60, then exist
Bianry image corresponding point position is set to 255, and 255 signs differ, and i.e. indicate resting；Otherwise, then at bianry image corresponding point position
Being set to 0,0 sign is identical, i.e. indicates nonstatic thing.
Next the quiescent time of stationary object is followed the tracks of and carry out timing.As it is shown on figure 3, scanning stationary object image
Point that is each, that find pixel value to be 255, adds 1 at accumulative image corresponding point counting.Then each point of the accumulative image of scanning,
Finding more than threshold value is the pixel of 80, forms the bianry image of new stationary object.New bianry image is carried out morphology
Expansion process in filtering, expansive working uses the kernel of 3*3.
The stationary object detected is screened by C
First the bianry image of stationary object being carried out connected component labeling, the profile next using legacy connected domain is special
Property is screened as follows:
First connected domain size resting between 500 to 10000 is filtered out.
The resting that the length-width ratio of the boundary rectangle first filtering out connected domain on the basis of the most superincumbent is more than 1.5, then profit
Whether the resting screened with the detection of HOG+SVM pedestrian detection algorithm is static pedestrian, detecting the company for pedestrian
Logical territory is got rid of.
The most as shown in Figure 4, first calculating current video image and the current slow difference updating background image corresponding point, difference is big
It is set to 255 in the corresponding point position result that threshold value is 60, otherwise, corresponding point position result is set to 0.0 sign is identical, is background, and 255
Sign differs, and is sport foreground.The bianry image obtaining moving object calculated above, then moving object image with work as
Front stationary object image and the bianry image obtaining new stationary object.
4. find connected domain area little with the boundary rectangle area ratio of this connected domain on the image of new stationary object
In the connected domain of 0.5, filtering this out, last remaining connected domain part is exactly final legacy.
The renewal of D double-background model
The renewal of double-background model uses two different background update methods, the fast renewal side updating background model
Method is as it is shown in figure 5, the vedio data utilizing every frame to read in is big with the gray value that current background image compares corresponding point
Little.If the gray value of current video image is bigger than the corresponding point position of current background image, then the corresponding point of current background image
Gray value adds 1, whereas if the gray value of current video image is less than current background image corresponding point position, then and current background figure
As corresponding point position subtracts 1.The fast background model that updates is exactly the current background image that the most each frame goes to update it.
As shown in Figure 6, first two masks of structure, first mask is gray value to the update method of slow renewal background model
Being all the image of 255, second mask is first mask result images of the inverted.At the beginning, if monitor video does not has
Occur legacy, update background model the most slowly and carry out overall situation mixed Gaussian context update, and two mask image information are protected
Hold constant.Once find legacy to be detected, then legacy position on image is stored, then at first mask
On corresponding position on image, the gray value of this part is all set to 0.Second mask image is to first mask figure
Renewal is negated as re-starting.Then first mask image is updated background image slowly with current and obtain needing renewal
Background area, the back of the body that need not update that second mask image is corresponding with obtaining legacy with current slow renewal background image
Scene area.First mask image also will with current video image with obtain needing the mixed Gaussian background in context update region
The input data image updated.Finally the background area image updated with need not update background area image phase or
Completely newly background image is updated slowly to complete.
The legacy detected labelling in addition is exported video monitoring by E
Calculate the boundary rectangle of the legacy connected domain in legacy bianry image and they are signed in video monitoring image
Upper corresponding position.
Claims (6)
1. the remnant object detection method of double mask context updates based on double-background model, it is characterised in that include walking as follows
Rapid:
A, reading monitor video, create fast renewal background model and update background model slowly, with the first two field picture of monitor video
Background model and the slow background image updating background model is updated as fast；
B, detection static foreground object；
C, the stationary object detected is screened；
D, the renewal of double-background model；
E, the legacy detected labelling in addition is exported video monitoring.
The remnant object detection method of a kind of double mask context updates based on double-background model the most according to claim 1,
It is characterized in that in described step A, the fast background update method updating background model is pixel relative method；Slowly background model is updated
Background update method be double mask mixed Gaussian background modelings, fast update background model and the slow renewal speed updating background model
Rate ratio is between 1:40 to 1:80.
The remnant object detection method of a kind of double mask context updates based on double-background model the most according to claim 1,
It is characterized in that described step B specifically includes following steps:
S21, carries out difference by the fast background model that updates with the slow background model that updates, indicates by two-value, if difference is more than threshold value, then
Being set to 255 at bianry image corresponding point position, 255 signs differ, and i.e. indicate resting；Otherwise, then in bianry image corresponding point
Place is set to 0, and 0 sign is identical, i.e. indicates nonstatic thing.
S22, is tracked timing to the most static object, and each point of scanning stationary object image, finding pixel value is 255
Point, add 1 at accumulative image corresponding point counting, then each point of the accumulative image of scanning, find counting more than setting threshold value
Pixel, forms the bianry image of new stationary object；
S23, carries out the expansive working of morphologic filtering to new bianry image.
The remnant object detection method of a kind of double mask context updates based on double-background model the most according to claim 1,
It is characterized in that: described step C comprises the steps:
C1, filter out connected domain size resting between 500 to 10000；
The resting that C2, the length-width ratio of the boundary rectangle filtering out connected domain are more than 1.5, recycling HOG+SVM pedestrian detection is calculated
Whether the resting that method detection screens is static pedestrian, gets rid of detecting the connected domain for static pedestrian；
C3, by current video image with current slow update background image difference, moving image difference more than threshold value be 60 right
The place's of putting result should be set to 255, otherwise, corresponding point position result is set to 0, the moving image obtained and current resting image phase with
Obtain new resting image, the resting image that this is new uses medium filtering；
C4, on new resting image, filter out the boundary rectangle area ratio of connected domain area and this connected domain less than 0.5
Connected domain.
The legacy detection side of a kind of double mask context updates based on double-background model the most according to claim 1 and 2
Method, it is characterised in that described fast renewal background model utilizes the every frame video image data read in and current background to carry out each picture
The updating decision that relatively comes of the gray value size on vegetarian refreshments updates background model, if the gray value of current video image is than the current back of the body
The corresponding point position of scape image is big, then the corresponding point gray value of current background image adds 1, whereas if the ash of current video image
Angle value is less than current background image corresponding point position, then current background image corresponding point position subtracts 1.
The legacy detection side of a kind of double mask context updates based on double-background model the most according to claim 1 and 2
Method, it is characterised in that the update method of described slow renewal background model is:
1) two masks of structure, first mask be gray value be all the image of 255, second mask is that first mask negates
After result images；
2) if monitor video does not occur legacy, update background model the most slowly and carry out overall situation mixed Gaussian context update,
And two mask image information keep constant；Legacy once detected, then under legacy position storage on image
Come, then on the corresponding position in first mask image, the gray value of this part is all set to 0, second mask
First mask image is re-started and negates renewal by image；
3) first mask image and the current slow background area updating background image and obtain needing renewal, cover for second
The background area that need not update that film image is corresponding with obtaining legacy with current slow renewal background image；
4) first mask image and current video image with obtain needing the mixed Gaussian context update in context update region
Input data image；
5) finally the background area image updated with need not the background area image phase that updates or obtain complete brand-new
Slowly background image is updated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610638539.9A CN106296677B (en) | 2016-08-03 | 2016-08-03 | A kind of remnant object detection method of double exposure mask context updates based on double-background model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610638539.9A CN106296677B (en) | 2016-08-03 | 2016-08-03 | A kind of remnant object detection method of double exposure mask context updates based on double-background model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106296677A true CN106296677A (en) | 2017-01-04 |
| CN106296677B CN106296677B (en) | 2019-04-02 |
Family
ID=57665716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610638539.9A Active CN106296677B (en) | 2016-08-03 | 2016-08-03 | A kind of remnant object detection method of double exposure mask context updates based on double-background model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106296677B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107038702A (en) * | 2017-04-17 | 2017-08-11 | 兰州交通大学 | The railroad track foreign body intrusion detection method of triple difference based on three background modelings |
| CN107204006A (en) * | 2017-06-01 | 2017-09-26 | 大连海事大学 | A kind of static target detection method based on double background difference |
| CN109409315A (en) * | 2018-11-07 | 2019-03-01 | 浩云科技股份有限公司 | A kind of ATM machine panel zone remnant object detection method and system |
| CN110896453A (en) * | 2019-09-22 | 2020-03-20 | 孙春兰 | Feedback type electronic equipment driving platform and method |
| CN111968181A (en) * | 2020-08-21 | 2020-11-20 | 中科微至智能制造科技江苏股份有限公司 | Real-time parcel position detection method and system based on image processing |
| CN112070033A (en) * | 2020-09-10 | 2020-12-11 | 天津城建大学 | Video carry-over detection method based on finite-state machine analysis |
| CN112084957A (en) * | 2020-09-11 | 2020-12-15 | 广东联通通信建设有限公司 | Mobile target retention detection method and system |
| CN114022468A (en) * | 2021-11-12 | 2022-02-08 | 珠海安联锐视科技股份有限公司 | Method for detecting article leaving and losing in security monitoring |
| CN114694092A (en) * | 2022-03-15 | 2022-07-01 | 华南理工大学 | Expressway monitoring video object-throwing detection method based on mixed background model |
| CN118097112A (en) * | 2024-03-15 | 2024-05-28 | 珠海安联锐视科技股份有限公司 | Deep learning-based fusion double-background article carry-over loss detection method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102902960A (en) * | 2012-09-25 | 2013-01-30 | 武汉烽火众智数字技术有限责任公司 | Leave-behind object detection method based on Gaussian modelling and target contour |
| CN103729858A (en) * | 2013-12-13 | 2014-04-16 | 广州中国科学院先进技术研究所 | Method for detecting article left over in video monitoring system |
| CN104156942A (en) * | 2014-07-02 | 2014-11-19 | 华南理工大学 | Detection method for remnants in complex environment |
-
2016
- 2016-08-03 CN CN201610638539.9A patent/CN106296677B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102902960A (en) * | 2012-09-25 | 2013-01-30 | 武汉烽火众智数字技术有限责任公司 | Leave-behind object detection method based on Gaussian modelling and target contour |
| CN103729858A (en) * | 2013-12-13 | 2014-04-16 | 广州中国科学院先进技术研究所 | Method for detecting article left over in video monitoring system |
| CN104156942A (en) * | 2014-07-02 | 2014-11-19 | 华南理工大学 | Detection method for remnants in complex environment |
Non-Patent Citations (1)
| Title |
|---|
| FATIH PORIKLI 等: "Robust Abandoned Object Detection Using Dual Foregrounds", 《EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107038702A (en) * | 2017-04-17 | 2017-08-11 | 兰州交通大学 | The railroad track foreign body intrusion detection method of triple difference based on three background modelings |
| CN107204006B (en) * | 2017-06-01 | 2020-02-07 | 大连海事大学 | Static target detection method based on double background difference |
| CN107204006A (en) * | 2017-06-01 | 2017-09-26 | 大连海事大学 | A kind of static target detection method based on double background difference |
| CN109409315B (en) * | 2018-11-07 | 2022-01-11 | 浩云科技股份有限公司 | Method and system for detecting remnants in panel area of ATM (automatic Teller machine) |
| CN109409315A (en) * | 2018-11-07 | 2019-03-01 | 浩云科技股份有限公司 | A kind of ATM machine panel zone remnant object detection method and system |
| CN110896453A (en) * | 2019-09-22 | 2020-03-20 | 孙春兰 | Feedback type electronic equipment driving platform and method |
| CN110896453B (en) * | 2019-09-22 | 2021-11-16 | 合肥森岑汽车用品有限公司 | Feedback type electronic equipment driving platform and method |
| CN111968181A (en) * | 2020-08-21 | 2020-11-20 | 中科微至智能制造科技江苏股份有限公司 | Real-time parcel position detection method and system based on image processing |
| CN112070033A (en) * | 2020-09-10 | 2020-12-11 | 天津城建大学 | Video carry-over detection method based on finite-state machine analysis |
| CN112084957A (en) * | 2020-09-11 | 2020-12-15 | 广东联通通信建设有限公司 | Mobile target retention detection method and system |
| CN114022468A (en) * | 2021-11-12 | 2022-02-08 | 珠海安联锐视科技股份有限公司 | Method for detecting article leaving and losing in security monitoring |
| CN114694092A (en) * | 2022-03-15 | 2022-07-01 | 华南理工大学 | Expressway monitoring video object-throwing detection method based on mixed background model |
| CN118097112A (en) * | 2024-03-15 | 2024-05-28 | 珠海安联锐视科技股份有限公司 | Deep learning-based fusion double-background article carry-over loss detection method |
| CN118097112B (en) * | 2024-03-15 | 2024-08-16 | 珠海安联锐视科技股份有限公司 | Deep learning-based fusion double-background article carry-over loss detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106296677B (en) | 2019-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106296677A (en) | A kind of remnant object detection method of double mask context updates based on double-background model | |
| CN107943837B (en) | Key-framed video abstract generation method for foreground target | |
| Cuevas et al. | Labeled dataset for integral evaluation of moving object detection algorithms: LASIESTA | |
| CN102663743B (en) | Personage's method for tracing that in a kind of complex scene, many Kameras are collaborative | |
| CN104134222B (en) | Traffic flow monitoring image detecting and tracking system and method based on multi-feature fusion | |
| CN104123544A (en) | Video analysis based abnormal behavior detection method and system | |
| CN111881853B (en) | Method and device for identifying abnormal behaviors in oversized bridge and tunnel | |
| CN105893962A (en) | Method for counting passenger flow at airport security check counter | |
| CN105160297A (en) | Masked man event automatic detection method based on skin color characteristics | |
| CN102819764A (en) | Method for counting pedestrian flow from multiple views under complex scene of traffic junction | |
| CN109145736B (en) | A kind of detection method that the subway station pedestrian based on video analysis inversely walks | |
| CN101715070B (en) | Method for automatically updating background in specifically monitored video | |
| CN110415268A (en) | A kind of moving region foreground image algorithm combined based on background differential technique and frame difference method | |
| Karpagavalli et al. | Estimating the density of the people and counting the number of people in a crowd environment for human safety | |
| Xu et al. | Segmentation and tracking of multiple moving objects for intelligent video analysis | |
| Buch et al. | Vehicle localisation and classification in urban CCTV streams | |
| CN103049748B (en) | Behavior monitoring method and device | |
| CN103020611A (en) | Method for detecting fighting behaviors | |
| CN110659546A (en) | Illegal booth detection method and device | |
| Huang et al. | A real-time and color-based computer vision for traffic monitoring system | |
| Li et al. | Intelligent transportation video tracking technology based on computer and image processing technology | |
| CN107122732A (en) | The quick license plate locating method of high robust under a kind of monitoring scene | |
| Yuliandoko et al. | Automatic vehicle counting using Raspberry pi and background subtractions method in the sidoarjo toll road | |
| CN105095891A (en) | Human face capturing method, device and system | |
| CN110858392A (en) | Monitoring target positioning method based on fusion background model |
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 |