CN101848369B - Method for detecting video stop event based on self-adapting double-background model - Google Patents
Method for detecting video stop event based on self-adapting double-background model Download PDFInfo
- Publication number
- CN101848369B CN101848369B CN200910216543A CN200910216543A CN101848369B CN 101848369 B CN101848369 B CN 101848369B CN 200910216543 A CN200910216543 A CN 200910216543A CN 200910216543 A CN200910216543 A CN 200910216543A CN 101848369 B CN101848369 B CN 101848369B
- Authority
- CN
- China
- Prior art keywords
- target
- background
- detection
- background model
- stop
- 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
Images
Abstract
The invention discloses a method for detecting a video stop event based on a self-adapting double-background model, which belongs to the field of the automatic detection and tracking of the video target of an intelligent type image monitoring technology (IVS). A self-adapting long/short-effect double-background model is designed and constructed, a quick target detection and tracking method is designed, the automatic stop detection of a stop event is realized by the long-effect tracking of a static/moving target, and an alarm is sent out at real time so as to overcome the defect of the prior art. The method comprises the following steps of: (1) initializing the long/short-effect double-background model; (2) carrying out BLOB target detection, stop target detection, ghost target detection and long/short-effect double-background update; (3) carrying out the quick BLOB target tracking of a heuristic method; (4) detecting the stop event and outputting the alarm; and (5) updating the selectivity of the long/short-effect double-background model. Relative to a traditional monitoring mode, the real-time property of the alarm is greatly enhanced, the intelligent type image monitoring reliability is higher, and the cost is lower.
Description
Technical field
The invention belongs to the automatic detection and tracking of intelligent image monitoring technique (IVS) video object field, particularly relate to a kind of method for detecting video stop event based on self-adapting double-background model.
Background technology
The operating personnel that traditional event detection monitor mode relies on Surveillance center fully carry out the picture monitoring.Have experimental data to show, after through monitoring in 22 minutes, the monitor staff possibly miss the picture institute event up to 95%.In the face of huge day by day monitoring content; Under the influence of various factorss such as monitor staff's sense of responsibility, operating state; The artificial supervisory control system of tradition often exists rate of failing to report height, response speed to reach problems such as poor reliability slowly, and the validity of whole video supervisory control system can't be guaranteed.Because supervisory control system is more and more huger, if all adopt artificial monitoring, its human cost also can be very high.And intelligent image monitoring technique (IVS) originates from computer vision technique (computer vision); It is analyzed image; Events of interest is found in information extraction from image, thereby can substitute artificial monitoring or assist artificial monitoring in some occasion.After 911 incidents, based on the situation of anti-terrorism, more and more urgent for the demand of IVS in the world, the video stop event detection has very and uses widely as one of technological important application of video intelligent event detection.For example the suspicious packages of boarding lounge, waiting room detects, and the airplane parking area is illegally placed quality testing and surveyed, and the taboo violating the regulations of field of traffic is stopped detection etc.Stop the ability that event detection possesses round-the-clock running, automatic time warning based on the taboo of video technique, huge economic and social value are arranged.General conventional video analytical technology mainly to as if moving target, be difficult to realize stopping comparatively accurately target detection and handle moving target and the situation that stops the target blending.Therefore, this research makes up self adaptation length and imitates double-background model according to level background model thought, designs target detection and tracking fast, through the long-acting tracking of quiet moving-target, realizes that the automatic time of prohibiting the incident of stopping detects and alarm.At home up to the present, do not retrieve similar techniques or related patent U.S. Patent No. report as yet.
Summary of the invention
The object of the invention provides a kind of method for detecting video stop event based on self-adapting double-background model; Design is imitated double-background model with structure self adaptation length; Design target detection and tracking fast; Through the long-acting tracking of quiet moving-target, the automatic detection that realizes prohibiting the incident of stopping is prohibited and is stopped and Real-time Alarm, to overcome the deficiency of conventional art.
The technical solution measure of realization the present invention's purpose is following:
A kind of method for detecting video stop event based on self-adapting double-background model carries out intellectuality to the situation in the scene and keeps watch on the basis of video signal collective, comprise that target detection, target following, event analysis is characterized in that comprising the steps:
(1) initialization of length effect background model is divided thought based on mixed Gauss model and background area, and the subelement that carries out length effect background model is divided, the subelement mixed Gauss model is initial;
(2) BLOB target detection, stop target, ghost target detection and length are imitated context update, stop the target area and update in the fugitive background, and long-acting background is carried out non-moving region selective updating, to adapt to the slow variation of illumination; Judge through ghost, find to stop target and leave original position, be used for realizing accurate renewal is done in corresponding fugitive background area;
(3) the BLOB fast target of heuristic is followed the tracks of, and for guaranteeing the adaptability of motion target tracking, designs heuristic fast, according to the related rule of the preferential geometric properties in position, follows the tracks of and judge position, the speed of target; Do you judge and support static target to follow the tracks of? If do not support, get into PBlob Tracker motion target tracking, if support, get into the target following of PBlob Tracker sound;
(4) prohibit and to stop event detection and outputting alarm, accomplish on the detection basis that stops target, stop target geometric properties, the time of staying, when dwell regions meets alarm conditions, then carry out the acousto-optic electric alarm, corresponding alarm event record warehouse-in;
(5) length is imitated the background model selective updating, and just static, just move and illumination variation to motion, selective updating length is imitated background model.
Described a kind of method for detecting video stop event based on self-adapting double-background model; It is characterized in that initialization, BLOB target detection that described length is imitated background model adopt the systemic decision method in all kinds of BLOB of design zone to realize the fast detecting of target.
Described a kind of method for detecting video stop event based on self-adapting double-background model is characterized in that, described stop target, ghost target detection and length are imitated context update; Comprise that the detection that stops target judges, stop between target following trial period, according to the geometric properties of this target; Accurately be updated in the fugitive background, need carry out the moving target association to attempting stopping target simultaneously, if do not reach the trial requirement; Then correct fugitive background, change motion target tracking simultaneously over to; The target of short, long-acting background subtraction output is for stopping target output result, and long-acting background is not carried out the renewal of moving region, and other zones are the same with fugitive background, carry out the mixed Gaussian background model and upgrade, to adapt to the slow variation of illumination.
Described a kind of method for detecting video stop event based on self-adapting double-background model; It is characterized in that the BLOB fast target of described heuristic is followed the tracks of, comprise that designing heuristic fast guarantees the real-time of following the tracks of; Which target detected BLOB zone belongs to; According to the nearest preferential matching principle in position, set up the Gaussian distribution likelihood function, calculate t target velocity constantly according to two continuous frames.
Described a kind of method for detecting video stop event based on self-adapting double-background model is characterized in that, describedly comprises that the detection that stops target judges, according to two factors: the movement velocity of target in successive frame is less than given threshold value; The target continuous proper motion seed that frame difference method is confirmed is counted less than given threshold value, and promptly stop the detection criteria of target: speed is less than given threshold value, and the proper motion seed number is less than threshold value.
It is emphasized that conventional hybrid Gauss or its modified model background model, mainly is to moving object detection, can not detect moving target and static target simultaneously.And this programme has original design on the adjacent frame seed penalty method filtration that combines to follow the tracks of, selective updating strategy.On double-background model foundation and adaptive updates flow process, done a large amount of creative works, further in embodiment, explain.
The present invention compares with traditional event detecting method of prior art, has following advantage and beneficial effect:
(1) based on existing video monitoring equipment, add this video intelligent detection module to expand its function, realize round-the-clock running.Can carry out 7 * 24 continual analyses to real-time video according to the detection threshold of user's setting.
(2) the great warning real-time that improved for traditional monitor mode.
(3) reliability of the present invention is higher, and cost is cheaper.
(4) significantly save human cost.When the supervisory control system scale was huger, traditional manual type need drop into more manpower, material resources and financial resources, and embodiment of the present invention then can significantly be saved the manpower cost.
Description of drawings
Fig. 1 is a method for detecting video stop event whole process sketch map of the present invention.
Fig. 1 .5 is that BLOB characteristic of the present invention is divided sketch map.
Fig. 2 is a method for detecting video stop event FB(flow block) sketch map of the present invention.
Fig. 3 is that conventional I VS taboo is stopped the event detection schematic flow sheet.
Fig. 4 is that background model of the present invention is set up and adaptive updates stream FB(flow block) sketch map.
Fig. 5 figure (containing a, b, c, d, the little figure of e) method for detecting video stop event embodiment 1 design sketch of the present invention.
Fig. 6 (containing a, b, c, d, e, the little figure of f) is a method for detecting video stop event of the present invention
Embodiment 2 design sketchs.
Embodiment
Can know referring to Fig. 1 Fig. 1 .5 Fig. 2 Fig. 3; Fig. 1 is that the video stop based on double-background model of the present invention stays the incident whole process, wherein length imitate the accurate fast updating of background model, based on fast target detection and tracking, the ghost target of BLOB method accurately detect, two background subtraction stop target detection is concrete key technology means.In said flow process, step is following:
(1) length is imitated the initialization of background model.Divide thought based on mixed Gauss model and background area, the subelement that carries out length effect background model is divided, the subelement mixed Gauss model is initial.
(2) BLOB target detection, stop target, ghost target detection, the BLOB shown in Fig. 1 .5 divides and comprises motion visual object BLOB, shade BLOB, ghost BLOB, illumination variation BLOB, background noise BLOB etc.Design the systemic decision method in all kinds of BLOB zone, realize the fast detecting of target.At first eliminate illumination variation, the trichromatic variation correlation of pixel is considered in the background subtraction, come further to handle through the illumination variation characteristic of calculating pocket simultaneously.Then; Calculate adjacent frame difference as the foreground moving seed; The thick prospect BLOB of the difference of obtaining is carried out eight be communicated with mark, calculate the motion seed number of each tag block, and judge that the interior motion seed number of connected component meets certain threshold value and takes a decision as to whether real motion prospect BLOB.Consider part target BLOB motion slowly, frame-to-frame differences does not almost have seed, and design compensates and eliminates this mistake through the barycenter predicted position of following the tracks of BLOB being carried out seed.
The detection that stops target judges that two factors of foundation: the movement velocity of target in successive frame is less than given threshold value; The target continuous proper motion seed that frame difference method is confirmed is counted less than given threshold value.Stop between target following trial period,, accurately be updated in the fugitive background, need carry out the moving target association to attempting stopping target simultaneously,, then correct fugitive background, change motion target tracking simultaneously over to if do not reach the trial requirement according to the geometric properties of this target.The target of short, long-acting background subtraction output is for stopping target output result.Long-acting background is not carried out the renewal of moving region, and other zones are the same with fugitive background, carries out the mixed Gaussian background model and upgrades, to adapt to the slow variation of illumination.
Stop the target area and update in the fugitive background, long-acting background is carried out non-moving region selective updating, to adapt to the slow variation of illumination; Judge through ghost, find to stop target and leave original position, be used for 5 pairs of corresponding fugitive background areas of performing step and do accurate renewal;
(3) the BLOB fast target of heuristic is followed the tracks of, and for guaranteeing the adaptability of motion target tracking, designs heuristic fast, according to the related rule of the preferential geometric properties in position, follows the tracks of and judge position, the speed of target; Do you judge and support static target to follow the tracks of? If do not support, get into PBlob Tracker motion target tracking, if support, get into the target following of PBlob Tracker sound;
Design heuristic fast and guarantee the real-time of following the tracks of.Each target O
i(t) define following key element: position p (t), geometric properties g (t), speed v (t), lifetime L (t), identity sign I (t), whether static sign, tracking mode (attempt, normal, extrapolation) etc.Position prediction is following constantly for moving target t:
Which target detected BLOB zone belongs to, and according to the nearest preferential matching principle in position, sets up the Gaussian distribution likelihood function:
In the following formula (2),
Be t moving target prediction constantly center, P
j=1,2 ..., n is the center of detected observation area.Parameter σ ∈ [0.3,0.5].
With p
jThe position is near more, then k
IjValue big more.Consider position, observation area p
j, target predicted value
T tracking results constantly is following:
According to two continuous frames, calculate t target velocity constantly.
(4) taboo is stopped event detection and outputting alarm, prohibits and stops the requirement that event alarm need satisfy three aspects: the requirement of target geometric properties; The time of staying of target; The stop place of target.For identification of targets, this method is considered the reason of real-time, does not do requirement.Have to meet when prohibiting the incident generation that stops alarm conditions, trigger the alarm signal of acousto-optic electricity, the alarm alarm event of corresponding target is done warehouse-in and is handled, and prepares against inquiry and uses.
(5) length is imitated the background model selective updating, and is to motion, just static, just move and illumination variation the selective updating background model.Comprise fill to draw motion BLOB, firm static BLOB, just move BLOB and illumination variation rebuild, according to the fugitive background model of dissimilar realization mLongBG, the fugitive background model selective updating of mShortBG and output.
What Fig. 3 illustrated is that conventional I VS taboo is stopped the event detection flow process.Conventional I VS is used for taboo and stops event detection, can't handle the sound Target Transformation, can only set up velocity to moving target and Position Tracking, can not effectively solve the occlusion issue of sound target, can not long-time, the sound target effective tracking on a large scale of fine solution.Conventional I VS is used for prohibiting and stops event detection, to the danger that moves to static a period of time in the scene stop measuring ability very a little less than.And the method for this programme, the taboo that then can effectively solve under the unconfined condition is stopped event detection.
Fig. 4 is that background model of the present invention is set up and adaptive updates stream FB(flow block) sketch map.Therefrom can know and background model set up the fugitive background model of mShortBG, the fugitive background model of mLongBG and subelement thereof to divide the subelement mixed model initial and other major parameter is initial as whole background frames contiguous frames etc. in initial frame sequence kind.Do you judge earlier to stop during new input picture and detect not? If detecting, the then traditional moving target of No comprises mixed Gauss model detection, many characteristics shadow Detection, the filtration of adjacent frame seed penalty method.If Yes then the mShortBG motion detection comprise and handle that illumination variation detects, many characteristics shade is eliminated adjacent frame seed penalty method and filtered; MLongB6 stops to detect and comprises that the judgement of processing real motion seed, ghost detect, stop foreground detection; The double-background model selective updating comprises treatments B lob, motion, static, firm motion just, illumination variation reconstruction, and according to dissimilar realization mShortB6 selective updatings and mLongBG selective updating.
The embodiment that Fig. 5 detects for ghost; Wherein tricycle (b) has left original position for tricycle for stopping in (a), (c) for not eliminating the fugitive background (conventional hybrid Gauss also has this characteristic) of ghost; (d) be the prospect that conventional method is not eliminated ghost, (e) this method is eliminated the prospect of ghost.Stop target and leave original position, the false target of Fig. 5 (d) can appear in target detection, and for such ghost target, judge according to three factors: the static target of trial and the natural seed point of long-acting background subtraction are less than certain threshold value; The geometric properties of this target and near onrelevant moving target are complementary; The textural characteristics of this target and near moving target are complementary.Ghost is in case judge the fugitive background area of then from long-acting background, correcting a mistake.
Fig. 6 is for prohibiting the another embodiment that stops event detection; (a) shows for incoming frame and the result that comprises illegal stop among the figure; (b), (c), (d) be fugitive background for this method stops the result after the target detection for not adding the prospect that stops target detection; (e) for stopping target detection, (f) be sport foreground.As can beappreciated from fig. 6, if do not stop target detection, among the figure tricycle of (b) and passing pedestrian sticking be in the same place, stop that (c) after the target detection then avoided gluing and, thereby in (a), realized the Continuous Tracking of target BLOB.Find out from figure (a), stop the detection of target, can also partly solve blocking under the single-view.As can beappreciated from fig. 6, stop target detection integrality of the present invention is also relatively good.
Claims (5)
1. the method for detecting video stop event based on self-adapting double-background model carries out intellectuality to the situation in the scene and keeps watch on the basis of video signal collective, comprises target detection, target following, event analysis, it is characterized in that comprising the steps:
(1) length is imitated the initialization of background model, divides based on mixed Gauss model and background area, and the subelement that carries out length effect background model is divided, each subelement mixed Gauss model is accomplished initialization by the conventional hybrid Gauss model through successive video frames;
(2) the systemic decision method in all kinds of BLOB of design zone; Realize the fast detecting of target; BLOB target detection, stop target, ghost target detection; To stop the target area and update in the fugitive background, long-acting background is carried out non-moving region selective updating, to adapt to the slow variation of illumination; Judge through ghost, find to stop target and leave original position, be used for realizing accurate renewal is done in corresponding fugitive background area;
(3) the BLOB fast target of heuristic is followed the tracks of, and for guaranteeing the adaptability of motion target tracking, designs heuristic fast, according to the related rule of the preferential geometric properties in position, follows the tracks of and judge position, the speed of target;
(4) prohibit and to stop event detection and outputting alarm, accomplish on the detection basis that stops target, stop target geometric properties, the time of staying, when dwell regions meets alarm conditions, then carry out the acousto-optic electric alarm, corresponding alarm event record warehouse-in;
(5) length is imitated the background model selective updating, and just static, just move and illumination variation to motion, selective updating length is imitated background model.
2. a kind of method for detecting video stop event according to claim 1 based on self-adapting double-background model; It is characterized in that initialization, BLOB target detection that described length is imitated background model adopt the systemic decision method in all kinds of BLOB of design zone to realize the fast detecting of video object.
3. a kind of method for detecting video stop event according to claim 1 based on self-adapting double-background model; It is characterized in that the BLOB fast target of described heuristic is followed the tracks of, comprise that designing heuristic fast guarantees the real-time of following the tracks of; Judge which target detected BLOB zone belongs to; According to the nearest preferential matching principle in position, set up the Gaussian distribution likelihood function, calculate t target velocity constantly according to two continuous frames.
4. a kind of method for detecting video stop event based on self-adapting double-background model according to claim 1 is characterized in that, described stop target, ghost target detection and length are imitated the background model selective updating; Comprise the detection judgement that stops target; Stop between target following trial period,, accurately be updated in the fugitive background according to the geometric properties of this target; Carry out the moving target association to attempting stopping target simultaneously; If do not reach the trial requirement, then correct fugitive background, change motion target tracking simultaneously over to; The target of short, long-acting background subtraction output is for stopping target output result, and long-acting background is not carried out the renewal of moving region, and other zones are the same with fugitive background, carry out the mixed Gaussian background model and upgrade, to adapt to the slow variation of illumination.
5. a kind of method for detecting video stop event according to claim 4 based on self-adapting double-background model; It is characterized in that; Describedly comprise that the detection that stops target judges, according to two factors: the movement velocity of target in successive frame is less than given threshold value; The target continuous proper motion seed that frame difference method is confirmed is counted less than given threshold value, and promptly stop the detection criteria of target: speed is less than given threshold value, and the proper motion seed number is less than threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910216543A CN101848369B (en) | 2009-12-04 | 2009-12-04 | Method for detecting video stop event based on self-adapting double-background model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910216543A CN101848369B (en) | 2009-12-04 | 2009-12-04 | Method for detecting video stop event based on self-adapting double-background model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101848369A CN101848369A (en) | 2010-09-29 |
| CN101848369B true CN101848369B (en) | 2012-10-03 |
Family
ID=42772804
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200910216543A Active CN101848369B (en) | 2009-12-04 | 2009-12-04 | Method for detecting video stop event based on self-adapting double-background model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101848369B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104156942B (en) * | 2014-07-02 | 2017-02-15 | 华南理工大学 | Detection method for remnants in complex environment |
| KR102399017B1 (en) | 2015-11-16 | 2022-05-17 | 삼성전자주식회사 | Method of generating image and apparatus thereof |
| CN106228572B (en) * | 2016-07-18 | 2019-01-29 | 西安交通大学 | A kind of the long inactivity object detection and tracking of carrier state mark |
| CN107204006B (en) * | 2017-06-01 | 2020-02-07 | 大连海事大学 | Static target detection method based on double background difference |
| CN113052781B (en) * | 2019-12-26 | 2024-04-12 | 华为技术有限公司 | Image detection method, device, equipment, system and storage medium |
| CN111289967A (en) * | 2020-03-31 | 2020-06-16 | 四川长虹电器股份有限公司 | Personnel detection tracking and counting algorithm based on millimeter wave radar |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002074368A (en) * | 2000-08-25 | 2002-03-15 | Matsushita Electric Ind Co Ltd | Moving object recognizing and tracking device |
| CN101105892A (en) * | 2007-07-30 | 2008-01-16 | 深圳市融合视讯科技有限公司 | Vehicle traffic accident automatic detection method |
| CN101330605A (en) * | 2008-07-30 | 2008-12-24 | 浙江大学 | Automatic monitoring device for protecting location article |
-
2009
- 2009-12-04 CN CN200910216543A patent/CN101848369B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002074368A (en) * | 2000-08-25 | 2002-03-15 | Matsushita Electric Ind Co Ltd | Moving object recognizing and tracking device |
| CN101105892A (en) * | 2007-07-30 | 2008-01-16 | 深圳市融合视讯科技有限公司 | Vehicle traffic accident automatic detection method |
| CN101330605A (en) * | 2008-07-30 | 2008-12-24 | 浙江大学 | Automatic monitoring device for protecting location article |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101848369A (en) | 2010-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101848369B (en) | Method for detecting video stop event based on self-adapting double-background model | |
| CN105744232B (en) | A kind of method of the transmission line of electricity video external force damage prevention of Behavior-based control analytical technology | |
| Fuentes et al. | People tracking in surveillance applications | |
| CN101739686B (en) | Moving object tracking method and system thereof | |
| CN102521983B (en) | Vehicle violation detection system based on high definition video technology and method thereof | |
| CN101794385B (en) | Multi-angle multi-target fast human face tracking method used in video sequence | |
| CN101458871B (en) | Intelligent traffic analysis system and application system thereof | |
| CN107705574A (en) | A kind of precisely full-automatic capturing system of quick road violation parking | |
| CN103366569A (en) | Method and system for snapshotting traffic violation vehicle in real time | |
| KR20060031832A (en) | A smart visual security system based on real-time behavior analysis and situation cognizance | |
| CN104506804A (en) | Device and method for monitoring abnormal behavior of motor vehicle on expressway | |
| CN103208190B (en) | Traffic flow detection method based on object detection | |
| CN109598700A (en) | Using the incipient fire detection method of convolutional neural networks | |
| CN105678730A (en) | Camera movement self-detecting method on the basis of image identification | |
| CN104200466A (en) | Early warning method and camera | |
| CN110633678B (en) | Quick and efficient vehicle flow calculation method based on video image | |
| CN104811655A (en) | System and method for film concentration | |
| CN106296677A (en) | A kind of remnant object detection method of double mask context updates based on double-background model | |
| CN114781479A (en) | Traffic incident detection method and device | |
| CN105574468A (en) | Video flame detection method, device and system | |
| CN103152558B (en) | Based on the intrusion detection method of scene Recognition | |
| CN111696135A (en) | Intersection ratio-based forbidden parking detection method | |
| CN109191492A (en) | A kind of intelligent video black smoke vehicle detection method based on edge analysis | |
| Li et al. | Intelligent transportation video tracking technology based on computer and image processing technology | |
| CN103605956A (en) | Emergency intelligence identification system based on video abstraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |