CN109712106A - A method of it removes and detects for monitor video object - Google Patents
A method of it removes and detects for monitor video object Download PDFInfo
- Publication number
- CN109712106A CN109712106A CN201711018163.2A CN201711018163A CN109712106A CN 109712106 A CN109712106 A CN 109712106A CN 201711018163 A CN201711018163 A CN 201711018163A CN 109712106 A CN109712106 A CN 109712106A
- Authority
- CN
- China
- Prior art keywords
- background
- background model
- update
- region
- fast
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of methods for removing detection for monitor video object, comprising the following steps: (1) reads in monitor video, create two background models, two background models are respectively fast to update background model and slow update background model；(2) background model is updated according to fast update background model and slowly, the region of steady change occurs in real-time detection image；It is exported using the background model of previous step as a result, obtaining the different contexts segmentation result of two renewal rates: fast to update background result and update background result slowly；(3) whether determinating area there is object loss, if so, saving the associated video frame of object loss, sounding an alarm, goes to step (4)；Otherwise (2) are gone to step；(4) double-background model is updated.
Description
Technical field
The present invention relates to intelligent video monitorings, computer vision field, and in particular to one kind is moved for monitor video object
Except the method for detection.
Background technique
It is an important application in security protection intellectual monitoring early warning system that object, which removes detection, can be applied to subway, airport,
To the intellectual monitoring of important objects and protection under the public arenas such as stadium.Currently, object removes detection, there are mainly two types of modes:
The first is that the variation of direct comparison topography determines whether target object is lost, such method need to be directed to specific monitored picture
Specified monitoring position, and judged by accident vulnerable to temporary block of the mobile objects such as pedestrian, vehicle；Another kind is bonding machine
The method of device study carries out target detection, then carries out real-time tracking therefore, it is determined that whether removing, this side to interesting target
Formula then needs huge operand, it is difficult to meet the requirement of real-time.So how under more complex environment to guarantee monitor video
Middle object removes the accuracy of detection, real-time is problem to be solved.
Summary of the invention
In order to solve the above technical problems, it is an object of the invention to objects in accurate, real-time detection monitor video to remove,
Based on fusion background model method, a kind of method for removing detection for monitor video object is proposed.
Technical scheme is as follows:
A method of it removes and detects for monitor video object, it is characterised in that the following steps are included:
(1) monitor video is read in, creates two background models, two background models are respectively the fast background model and slow of updating
Update background model；The image data of monitor video is obtained using camera, and using first frame image as two models of beginning
Background image, then use subsequent N frame image update so that model adapts to current illumination variation；
(2) background model is updated according to fast update background model and slowly, the area of steady change occurs in real-time detection image
Domain；It is exported using the background model of previous step as a result, obtaining the different contexts segmentation result of two renewal rates: fast to update
Background result and slow update background result；
(3) whether determinating area there is object loss, if so, saving the associated video frame of object loss, sounding an alarm, turns
Step (4)；Otherwise (2) are gone to step；
(4) double-background model is updated.
In the above method, the method for two background models of creation in step (1) are as follows:
Two background models are respectively fast update background model and update background model slowly, use identical creation machine
System --- mixed Gaussian background modeling, but renewal rate is different；The fast background model that updates uses larger learning rate, detects the short time
Interior contexts；The slow background model that updates uses lesser learning rate, detects the contexts in the long period.
In the above method, occur the method in the region of steady change in step (2) in detection image are as follows:
Fastly, the segmentation result that background model difference output valve is 0,1 slowly, 0 indicates background, and 1 indicates prospect, then for prison
The pixel of control picture specific position indicates that the position is currently static when fast background model is 0, slow background model is 1
State, but be movement state, as candidate region of variation within past a period of time；Fastly, background model value is respectively 1,1 slowly
When, then it represents that movement state；To reduce noise jamming, setting finite state machine is changed region screening, fast, slow when pixel
Background model value is switched to 10 states from 11 states, and when the 10 states a certain fixed frame number of maintenance, then it is assumed that the position is steady change
Region, lasting statistics, which meets the pixel number required in this way, can determine and stablize if number of pixels remains stable substantially
The region of variation.
In the above method, whether determinating area there is the method for object loss in step (3) are as follows:
The characteristics of according to monitor video, background of the general position that steady change occurs in video pictures is ground, wall
Face removes if object occurs, and performance in video is: from the picture with obvious contour of object, it is equal for changing in the region
Even, continuous background frame, gradient can reduce, and according to this feature, carry out gradient value meter to the region for being determined as steady change
It calculates, is compared with the corresponding position of preceding N frame.If gradient value is decreasing trend, it is determined as that object is lost.
In the above method, the update method of double-background model in step (4) are as follows:
Fast background model variation in time, is judged to substantially to impact subsequent judgement after object removes, and slow
Original Pixel Information can be retained in region of variation in a long time by updating background model, subsequent that object occurs in same position
Body will receive interference when removing, and therefore, only update slow background model, update method are as follows: the position that object removes position occurs for construction
Mask resets to original state to the mixture Gaussian background model relevant parameter of mask position, and other positions remain unchanged, and obtain
New slow update background image.
The present invention is compared with the prior art, and has following good result:
1, the present invention by merge background model remnant object detection method be applied to object remove detection, instead of by pair
Object than topography's variation removes detection method, has better robustness and anti-interference.
2, the present invention combines the practical application scene of monitor video, by the method for edge detection distinguish object leave and
It removes, while guaranteeing high accuracy, the method compared to real-time tracking improves detection system real-time.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention；
Fig. 2 is the finite state machine diagram that steady change occurs for determinating area.
Specific embodiment
Below in conjunction with attached drawing and specific implementation, the present invention will be described in detail, but not as a limitation of the invention.
Such as Fig. 1, steps are as follows for the realization of this method:
(1) monitor video is read in, two background models are created:
The image data of monitor video is obtained using camera, and using first frame image as the background for starting two models
Then image uses subsequent N frame image update, so that model adapts to current illumination variation.After N takes 300 frames or so may make
Continuous detection obtains better effects.
(2) occurs the region of steady change in detection image.
Using the background model output of previous step as a result, obtaining the different contexts segmentation knot of two renewal rates
Fruit --- it is fast to update background result and slow update background result.Wherein, the renewal rate for updating background result fastly is 28 seconds/time,
The slow renewal rate for updating background result is 0.5 second/time.The fast renewal rate for updating background model and updating background model slowly exists
Proper between 40:1 to 80:1, this ratio can be set according to specific actual environment.To given input video,
Fastly, the contexts segmentation result that background model provides respectively slowly, two there are the differences on certain time: when the fortune in video
Dynamic object puts down package, static in a short time due to wrapping up, the people then only moved in fast background model；And from length
From the point of view of phase, wrap up it is static be it is temporary, in regular hour threshold value, it is still movement state, then in slow background model
Both people is contained, also includes package.
The background segment result of two models can indicate that 0 indicates that segmentation result is background, and 1 is expressed as prospect with 01 matrix,
There are four types of a certain pixel of then fast, slow background model is total in the state of particular moment:
00: indicating that the pixel is background；
01: indicating the steady change region that may be removed；
10: indicating capped static foreground；
11: indicating that the pixel is moving object；
It can determine that the region of steady change by the state conversion of pixel, as shown in Figure 2.When pixel in video frame picture
State by movement state A (11) be transformed into may target B (10), the number that continuously occurs of statistics B, when B state frequency of occurrence reaches
When to given threshold n, pixel status is changed into C, that is, thinks that the pixel is the group pixel in steady change region.
According to monitor video the characteristics of changing over time, the sum of different moments state C pixel is counted, when C-state picture
Plain sum reach stablize and overlay area within the threshold range of setting when, then it is assumed that C-state pixel region be steady change
Region.
(3) whether determinating area there is object removal:
The judgement that detection carries out object removal can be changed by gradient value.If there is object removal, leave, is variation
Uniformly, continuously background frame, gradient can reduce.Present frame picture and its preceding k frame images are saved, institute in this period is counted
Some frame images are considered that object removes in the gradient value of corresponding position if there is gradient downward trend.
(4) update of double-background model:
Fast background model variation in time, is judged to substantially to impact subsequent judgement after object removes.And it is slow
Original Pixel Information can be retained in region of variation in a long time by updating background model, subsequent that object occurs in same position
Body will receive interference when removing.Therefore, slow background model, update method are only updated are as follows: the position that object removes position occurs for construction
Mask resets to original state to the mixture Gaussian background model relevant parameter of mask position, and other positions remain unchanged, and obtain
New slow update background image.
Claims (5)
1. a kind of method for removing detection for monitor video object, it is characterised in that the following steps are included:
(1) monitor video is read in, two background models are created, two background models are respectively that the fast background model that updates is updated with slow
Background model；The image data of monitor video is obtained using camera, and using first frame image as the back for starting two models
Then scape image uses subsequent N frame image update, so that model adapts to current illumination variation；
(2) background model is updated according to fast update background model and slowly, the region of steady change occurs in real-time detection image；Make
It is exported with the background model of previous step as a result, obtaining the different contexts segmentation result of two renewal rates: updating background fastly
As a result background result and is slowly updated；
(3) whether determinating area there is object loss, if so, saving the associated video frame of object loss, sounding an alarm, goes to step
(4)；Otherwise (2) are gone to step；
(4) double-background model is updated.
2. the method according to claim 1 for removing detection for monitor video object, which is characterized in that in step (1)
The method for creating two background models are as follows:
Two background models are respectively fast update background model and update background model slowly, use identical set-up mechanism --- and it is mixed
Gaussian Background modeling is closed, but renewal rate is different；The fast background model that updates uses larger learning rate, detects the front and back in the short time
Background；The slow background model that updates uses lesser learning rate, detects the contexts in the long period.
3. the method according to claim 1 for removing detection for monitor video object, which is characterized in that in step (1)
Occur the method in the region of steady change in detection image are as follows:
Fastly, the segmentation result that background model difference output valve is 0,1 slowly, 0 indicates background, and 1 indicates prospect, then for monitoring picture
The pixel of face specific position indicates that the position is currently resting state when fast background model is 0, slow background model is 1, but
It is movement state, as candidate region of variation within past a period of time；Fastly, when background model value is respectively 1,1 slowly, then table
Show movement state；To reduce noise jamming, setting finite state machine is changed region screening, when fast, the slow background mould of pixel
Offset is switched to 10 states from 11 states, and when the 10 states a certain fixed frame number of maintenance, then it is assumed that the position is the region of steady change, is held
Continuous statistics meets the pixel number required in this way, if number of pixels remains stable substantially, can determine and steady change occurs
Region.
4. the method according to claim 1 for removing detection for monitor video object, which is characterized in that in step (1)
The method whether determinating area object loss occurs are as follows:
The characteristics of according to monitor video, background of the general position that steady change occurs in video pictures are ground, metope, if
Be that object occurs to remove, performance in video is: the region changes from the picture with obvious contour of object as uniformly, even
Continuous background frame, gradient can reduce, and according to this feature, carry out gradient value calculating to the region for being determined as steady change, with
The corresponding position of preceding N frame is compared.If gradient value is decreasing trend, it is determined as that object is lost.
5. the method according to claim 1 for removing detection for monitor video object, which is characterized in that in step (1)
The update method of double-background model are as follows:
Fast background model variation in time, is judged to substantially to impact subsequent judgement after object removes, and updates slowly
Background model can retain original Pixel Information in region of variation in a long time, subsequent that object shifting occurs in same position
Except when will receive interference, therefore, only update slow background model, update method are as follows: construction occur object remove position position mask,
Original state is reset to the mixture Gaussian background model relevant parameter of mask position, other positions remain unchanged, and obtain new
It is slow to update background image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711018163.2A CN109712106A (en) | 2017-10-26 | 2017-10-26 | A method of it removes and detects for monitor video object |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711018163.2A CN109712106A (en) | 2017-10-26 | 2017-10-26 | A method of it removes and detects for monitor video object |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109712106A true CN109712106A (en) | 2019-05-03 |
Family
ID=66252915
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711018163.2A Pending CN109712106A (en) | 2017-10-26 | 2017-10-26 | A method of it removes and detects for monitor video object |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109712106A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160187A (en) * | 2019-12-20 | 2020-05-15 | 浙江大华技术股份有限公司 | Method, device and system for detecting left-behind object |
-
2017
- 2017-10-26 CN CN201711018163.2A patent/CN109712106A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160187A (en) * | 2019-12-20 | 2020-05-15 | 浙江大华技术股份有限公司 | Method, device and system for detecting left-behind object |
| CN111160187B (en) * | 2019-12-20 | 2023-05-02 | 浙江大华技术股份有限公司 | Method, device and system for detecting left-behind object |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9501701B2 (en) | Systems and methods for detecting and tracking objects in a video stream | |
| CN103729858B (en) | A kind of video monitoring system is left over the detection method of article | |
| JP2009533778A (en) | Video segmentation using statistical pixel modeling | |
| CN108376235A (en) | Image detecting method, device and computer readable storage medium | |
| CN103714325A (en) | Left object and lost object real-time detection method based on embedded system | |
| CN104063885A (en) | Improved movement target detecting and tracking method | |
| CN102194443A (en) | Display method and system for window of video picture in picture and video processing equipment | |
| CN109919053A (en) | A kind of deep learning vehicle parking detection method based on monitor video | |
| Wang et al. | A video-based traffic violation detection system | |
| Zhang et al. | Moving vehicles segmentation based on Bayesian framework for Gaussian motion model | |
| CN102930719A (en) | Video image foreground detection method for traffic intersection scene and based on network physical system | |
| CN110636281B (en) | Real-time monitoring camera shielding detection method based on background model | |
| CN106650638A (en) | Abandoned object detection method | |
| CN102496281A (en) | Vehicle red-light violation detection method based on combination of tracking and virtual loop | |
| CN109948455A (en) | One kind leaving object detecting method and device | |
| CN106327525A (en) | Machine room important place border-crossing behavior real-time monitoring method | |
| KR20170077545A (en) | Real-Time Object Tracking System and Method for in Lower Performance Video Devices | |
| CN112837323A (en) | Video processing method, system and storage medium based on portrait segmentation | |
| CN110858392A (en) | Monitoring target positioning method based on fusion background model | |
| CN104504378A (en) | Method and device for detecting image information | |
| CN103049748A (en) | Behavior-monitoring method and behavior-monitoring system | |
| CN108986145A (en) | Method of video image processing and device | |
| CN106780544B (en) | The method and apparatus that display foreground extracts | |
| CN103209321B (en) | A kind of video background Rapid Updating | |
| CN109712106A (en) | A method of it removes and detects for monitor video object |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190503 |
|
| WD01 | Invention patent application deemed withdrawn after publication |