CN103400382A - Abnormal panel detection algorithm based on ATM (Automatic Teller Machine) scene - Google Patents
Abnormal panel detection algorithm based on ATM (Automatic Teller Machine) scene Download PDFInfo
- Publication number
- CN103400382A CN103400382A CN2013103180054A CN201310318005A CN103400382A CN 103400382 A CN103400382 A CN 103400382A CN 2013103180054 A CN2013103180054 A CN 2013103180054A CN 201310318005 A CN201310318005 A CN 201310318005A CN 103400382 A CN103400382 A CN 103400382A
- Authority
- CN
- China
- Prior art keywords
- algorithm
- target
- background
- rechecking
- atm
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides an abnormal panel detection algorithm based on an ATM (Automatic Teller Machine) scene and application thereof in financial intelligence security and protection. The abnormal panel detection algorithm comprises the following steps of firstly calculating a dynamic change of a prospect by adopting a three frame differencing method when the algorithm is started so as to determine a self-adaptive background; then carrying out initial detection on an ATM abnormal panel based on the self-adaptive prospect so as to improve the algorithm efficiency; and finally, carrying out redetection based a method of multi-dimensional feature vectors so as to improve the algorithm accuracy. According to the abnormal panel detection algorithm and the application thereof, intelligent analysis and detection aiming at behaviours, such as adlet sticking, illegal dismantling, additional arrangement of a keyboard and a card slot and the like of an ATM can be effectively solved, so that the security of bank inspection and control is improved, a great deal of manpower is saved for bank users and the efficiency is improved at the same time.
Description
Technical field
The invention belongs to banking machine vision field, particularly a kind of based on the abnormal panel detection algorithm under the ATM scene, and the application of the method in the financial intelligent security protection.
Technical background
Along with the bank ATM Possum is day by day universal, people use bank card increasing in the ratio of the enterprising industry business of ATM Possum operation, the business such as remittance of transferring accounts from depositing and withdrawing in the past, and people are also increasing to the dependency degree of ATM Possum.Meanwhile, the offender utilizes the bank ATM Possum to implement criminal activity also in continuous increase.The crime of most of ATM to be card reader, false camera, false keyboard to be installed and to paste the property that the various means such as small advertisement are gained the user by cheating, although ATM itself is supporting perfect supervisory system arranged, because control point is numerous, considerably beyond people's monitoring capacity.
Under this background, based on the financial intelligent methods of video analyses of computer vision, arise at the historic moment.The method utilizes computer vision technique to carry out real-time analysis to the ATM guarded region, if alarm is sent in unusual circumstance, reminds the monitor staff to note, anticrime generation timely and effectively.The ATM intellectual analysis has two kinds of application models at present, the one, front end DVS or intelligent DVR pattern, being about to ATM intellectual analysis algorithm is embedded in DSP, the defect of this application is the video that a DVS can only access 2 left and right, tunnel usually, if to the whole city's all ATM implementing monitorings of the whole province even, need to arrange a large amount of front-end equipments, cost is higher.Another kind is the back-end server pattern, is about to ATM intellectual analysis algorithm and is arranged in back-end server, and the concurrent way of the benefit separate unit server of this application is higher, and Ke Da 20 tunnels, even have indivedual producers can accomplish 32 tunnels usually.But as a rule, the atm device number of large bank a medium-sized city reaches up to a hundred, namely enables to reach the concurrency on 32 tunnels, also needs multiple servers could monitor whole city's all devices, and need to take a large amount of network bandwidths, affect the operation of other monitor supervision platform.In order to solve this situation, the present invention proposes a kind of ATM poll testing mechanism, all ATM of the whole city are divided into groups by outlet, take between group concurrent, the strategy of poll in group, can guarantee that like this in the situation that the system concurrency degree is constant, the treatable number of devices of a station server increases greatly.
Under polling mechanism, to organizing interior every equipment, detect in turn, only obtain a two field picture at every turn and analyze, classical background modeling method at this moment, will be no longer applicable as mixed Gaussian etc.Based on this, the present invention has designed the abnormal panel detection algorithm of a kind of ATM under polling mechanism.
Summary of the invention
The objective of the invention is in order to solve the restriction of existing ATM intelligent analysis system due to concurrent way, cause the limited bottleneck of the number of devices of processing, a kind of ATM poll testing mechanism has been proposed, and designed the abnormal panel detection algorithm of ATM under a kind of polling mechanism, thereby greatly improved the atm device number of separate unit server monitoring.Summary of the invention is as follows:
ATM intelligent video analysis system poll testing mechanism, technical characterictic is as follows:
For same outlet, video camera is carried out to linearity test, namely detect one by one; For different outlets, carry out concurrent detection.If the A of outlet and B are arranged, A has video camera a, b, c; B has video camera d, e, and f, so, and system parallel processing a, the d video camera, cut out link after handling, then process b, e, c afterwards, f, a afterwards, d.Concurrent like this way is 2 but accessible number of devices is 6.The poll testing mechanism can guarantee greatly to increase the number of devices that the separate unit server can be processed under the prerequisite that does not increase concurrent way.
The abnormal panel detection algorithm of ATM under polling mechanism, technical characterictic is as follows:
(1), based on self-adaptation foreground extraction algorithm, by based on three frame difference algorithms, carrying out foreground detection early stage, determine adaptive background, namely when three frame difference method acquisition prospects change less than threshold value T, confirm that present frame is the acquiescence Background, otherwise continue the adaptive learning background.
(2), based on the self-adaptation foreground extraction, carry out initial survey, its algorithm is, after background study completes, by the background subtraction method, present image and adaptive background are carried out to difference, obtain current foreground picture, the foreground picture that extracts is carried out to the pinpoint target extraction by largest connected territory method, when pinpoint target is in certain zone of reasonableness, adds the reinspection formation to recheck this target and using this as suspicious abnormal panel target; If target sizes is directly filtered outside setting range the time, experimental results show that this just detecting method can effectively improve efficiency of algorithm, reduce the algorithm computational complexity, thereby improve system computing concurrency.
(3), based on the self-adaptation foreground extraction, carry out abnormal object and recheck algorithm, its algorithm carries out the abnormal object reinspection by the method based on multidimensional characteristic vectors, to improve the algorithm accuracy, show in algorithm by to rechecking target and background figure corresponding region, carrying out based on dutycycle, texture, edge, the method of color is advanced coupling, only have four features all to meet abnormal panel needs and just think that rechecking target passes through to recheck, if recheck unsuccessfully, carrying out abnormal panel regional background upgrades, thereby illumination is disturbed and filtered, thereby reach the purpose of reinspection, improve the algorithm accuracy.
(4), recheck in for the feature of texture, algorithm is by calculating simultaneously the gray level co-occurrence matrixes of 4 directions to rechecking target and background figure corresponding region, then calculate average and the variance of energy, entropy, moment of inertia and correlativity in gray level co-occurrence matrixes, obtain the proper vector of two octuples, whether calculate the Euclidean distance of two vectors, confirming to recheck target is abnormal object.
(5), recheck in for the feature at edge, algorithm obtains respectively and rechecks target and background figure corresponding region contour feature by the canny operator, then to two profile diagrams are carried out to the step-by-step exclusive disjunction, by the side's of card distribution, ask respectively the otherness of rechecking target figure after target and background graph region and exclusive disjunction afterwards, if two differences are all greater than threshold value T, think that current goal is abnormal object, otherwise carry out filter operation.
(6), recheck in for the feature of color, algorithm obtains at first respectively rechecks hsv color space, target and background figure corresponding region, HS Color Statistical histogram, then by two the HS Color Statistical histograms of method contrast that distribute based on card side, if diversity factor is less than threshold value T, think that rechecking target is abnormal object, otherwise filter rechecking target.
(7), based on personnel, enter the ATM panel method for detecting abnormality of detection, show, at first algorithm judges and stands personnel whether someone enters in district by the background subtraction method, if nobody stands, carry out abnormal panel detection, otherwise do not detect; If detect for a long time someone simultaneously, stand, algorithm is rechecked by three frame difference methods, if the people detected, continues to wait for, if detect nobody, carries out the illumination ELIMINATION OF ITS INTERFERENCE, carries out current background figure renewal.
The accompanying drawing explanation
Fig. 1 is basic flow sheet of the present invention;
Fig. 2 is self-adaptation foreground extraction initial survey algorithm flow chart of the present invention;
Fig. 3 is that self-adaptation foreground extraction of the present invention is rechecked algorithm flow chart;
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the present invention's part embodiment, rather than whole embodiment.Based on the embodiment in the present invention, those of ordinary skills, not making under the creative work prerequisite the every other embodiment that obtains, belong to the scope of protection of the invention.
ATM intelligent analysis system polling mode arranges:
Check the environment of all ATM, comprise whether panel reflective, whether image clear etc., count the atm device that all can carry out intellectual analysis, by outlet, divide into groups, if total outlet number is greater than the concurrent way of maximum, the outlet that the ATM number of units is less is incorporated to other outlet, guarantees that as far as possible the atm device number of every group equates.
Under polling mechanism, the abnormal panel detection algorithm of ATM scene, implementing procedure is as follows:
1, based on self-adaptation foreground extraction algorithm, by based on three frame difference algorithms, carrying out foreground detection early stage, determine adaptive background, at first, algorithm is deposited into nearest three two field pictures in a vector, the two field picture if newly arrive, by current frame image push_back in vector, and pop_front the first two field picture, in store three nearest two field pictures all the time in vector.Adjacent two frames are done poor and binaryzation in twos, step-by-step phase as a result or, as the dynamic change of prospect, unmanned if dynamic change less than the threshold value of certain setting, illustrate present frame, the use current frame image is as initial background.
2, based on personnel, enter the ATM panel method for detecting abnormality that zone is detected, at first by the background subtraction method, extract prospect, according to the stand prospect in zone of personnel, judge whether that someone enters, if nobody stands, enter the abnormal panel detection of step 3, otherwise do not detect; If detect for a long time someone simultaneously, stand, algorithm is rechecked by three frame difference methods, if the people detected, continues to wait for, if detect nobody, carries out the illumination ELIMINATION OF ITS INTERFERENCE, carries out current background figure renewal; After if the people being detected and leaving the T frame, adopt three frame difference methods to carry out abnormal panel detection, because the method does not need background, can effectively reduce the wrong report that produces not in time due to context update.
3, based on the self-adaptation foreground extraction, carry out initial survey, algorithm carries out difference by the background subtraction method to present image and adaptive background, obtain current foreground picture, the foreground picture of counter plate surveyed area carries out the pinpoint target extraction by largest connected territory method, when the pinpoint target size is in certain zone of reasonableness, adds the reinspection formation to recheck this target and using this as suspicious abnormal panel target; If target sizes or too large, directly filter in the time of outside setting range, experimental results show that this first detecting method can effectively improve efficiency of algorithm, reduce the algorithm computational complexity, thereby improve system computing concurrency.
4, based on the self-adaptation foreground extraction, carry out abnormal object and recheck algorithm, algorithm carries out the abnormal object reinspection by the method based on multidimensional characteristic vectors, to improve the algorithm accuracy.From recheck formation, take out and recheck target, to rechecking target and background figure corresponding region, carry out advancing coupling based on the method for dutycycle, texture, edge, color, only have four features all to meet abnormal panel needs and just think that rechecking target passes through to recheck, if recheck unsuccessfully, carrying out abnormal panel regional background upgrades, thereby illumination is disturbed and to be filtered, thereby reach the purpose of reinspection, improve the algorithm accuracy.
5, to rechecking the target exploitation dutycycle, filter.The foreground point of rechecking target is added up, and, divided by the area of rechecking target, obtained rechecking the dutycycle of target, if dutycycle, greater than certain threshold value, enters the texture filtering of step 6, otherwise recheck unsuccessfully, judge that this zone is normal panel.Get the next row target of rechecking in formation, jump to step 4.
6, to rechecking the target exploitation texture filtering.At first current figure and the Background of rechecking target area are used respectively to sobel operator compute gradient figure, then calculate gray level co-occurrence matrixes, wherein adopt 16 gray shade scales, 0 degree, 45 degree, 90 degree, 4 directions of 135 degree.Calculate energy, entropy, moment of inertia and the correlativity of gray level co-occurrence matrixes, average respectively, variance, obtain the proper vector of an octuple.Calculate the Euclidean distance between two texture feature vectors, if greater than certain threshold value, enter the edge filter of step 7, otherwise recheck unsuccessfully, judge that this zone is normal panel.Get the next row target of rechecking in formation, jump to step 4.
7, to rechecking the target exploitation edge filter.At first to rechecking target area, with the canny boundary operator, calculate profile, then to two profile diagrams are carried out to the step-by-step exclusive disjunction, by the side's of card distribution, ask respectively the otherness of rechecking target figure after target and background graph region and exclusive disjunction afterwards, if two differences are all greater than certain threshold value, enter the color filtering of step 8, otherwise recheck unsuccessfully, judge that this zone is normal panel.Get the next row target of rechecking in formation, jump to step 4.
8, to rechecking the target exploitation color filtering.Obtain and recheck hsv color space, target and background figure corresponding region at first respectively, HS Color Statistical histogram, then by two the HS Color Statistical histograms of method contrast that distribute based on card side, if diversity factor is less than threshold value T, think that rechecking target is abnormal object, send alarm, otherwise recheck unsuccessfully, judge that this zone is normal panel.Whole reinspection finishes.Get the next row target of rechecking in formation, jump to step 4.
Claims (8)
1. one kind based on the abnormal panel detection algorithm under the ATM scene, it is characterized in that utilizing carrying out initial survey based on the self-adaptation foreground extraction, to improve efficiency of algorithm; Utilization is rechecked based on the method for multidimensional characteristic vectors, to improve the algorithm accuracy rate.
2. method according to claim 1, it is characterized in that by based on three frame difference algorithms, carrying out foreground detection early stage, determining adaptive background based on self-adaptation foreground extraction algorithm, namely when three frame difference method acquisition prospects change less than threshold value T, confirm that present frame is the acquiescence Background, otherwise continue the adaptive learning background.
3. method according to claim 1, it is characterized in that carrying out initial survey based on the self-adaptation foreground extraction, its algorithm is, after by the described algorithm of claim 2, carrying out background study, by the background subtraction method, present image and adaptive background are carried out to difference, obtain current foreground picture, the foreground picture that extracts is carried out to the pinpoint target extraction by largest connected territory method, when pinpoint target is in certain zone of reasonableness, adds the reinspection formation to recheck this target and using this as suspicious abnormal panel target; If target sizes is directly filtered outside setting range the time, experimental results show that this just detecting method can effectively improve efficiency of algorithm, reduce the algorithm computational complexity, thereby improve system computing concurrency.
4. method according to claim 1, it is characterized in that based on the self-adaptation foreground extraction, carrying out abnormal object rechecks algorithm, its algorithm carries out the abnormal object reinspection by the method based on multidimensional characteristic vectors, to improve the algorithm accuracy, show in algorithm by to rechecking target and background figure corresponding region, carrying out based on dutycycle, texture, edge, the method of color is advanced coupling, only have four features all to meet abnormal panel needs and just think that rechecking target passes through to recheck, if recheck unsuccessfully, carrying out abnormal panel regional background upgrades, thereby illumination is disturbed and filtered, thereby reach the purpose of reinspection, improve the algorithm accuracy.
5. method according to claim 4, in it is characterized in that rechecking for the feature of texture, algorithm is by calculating simultaneously the gray level co-occurrence matrixes of 4 directions to rechecking target and background figure corresponding region, then calculate average and the variance of energy, entropy, moment of inertia and correlativity in gray level co-occurrence matrixes, obtain the proper vector of two octuples, whether calculate the Euclidean distance of two vectors, confirming to recheck target is abnormal object.
6. method according to claim 4, in it is characterized in that rechecking in rechecking for the character at edge, algorithm obtains respectively and rechecks target and background figure corresponding region contour feature by the canny operator, then to two profile diagrams are carried out to the step-by-step exclusive disjunction, by the side's of card distribution, ask respectively the otherness of rechecking target figure after target and background graph region and exclusive disjunction afterwards, if two differences all greater than threshold value T, think that current goal is abnormal object, otherwise carry out filter operation.
7. method according to claim 4, in it is characterized in that rechecking, for color, recheck method, algorithm obtains at first respectively rechecks hsv color space, target and background figure corresponding region, HS Color Statistical histogram, then by two the HS Color Statistical histograms of method contrast that distribute based on card side, if diversity factor less than threshold value T, thinks that rechecking target is abnormal object, otherwise filter rechecking target.
8. one kind based on the abnormal panel detection algorithm under the ATM scene, it is characterized in that entering based on personnel the ATM panel method for detecting abnormality of detection, show, at first algorithm judges and stands personnel whether someone enters in district by the background subtraction method, if nobody stands, carry out abnormal panel detection, otherwise do not detect; If detect for a long time someone simultaneously, stand, algorithm is rechecked by three frame difference methods, if the people detected, continues to wait for, if detect nobody, carries out the illumination ELIMINATION OF ITS INTERFERENCE, carries out current background figure renewal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013103180054A CN103400382A (en) | 2013-07-24 | 2013-07-24 | Abnormal panel detection algorithm based on ATM (Automatic Teller Machine) scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013103180054A CN103400382A (en) | 2013-07-24 | 2013-07-24 | Abnormal panel detection algorithm based on ATM (Automatic Teller Machine) scene |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103400382A true CN103400382A (en) | 2013-11-20 |
Family
ID=49563994
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013103180054A Pending CN103400382A (en) | 2013-07-24 | 2013-07-24 | Abnormal panel detection algorithm based on ATM (Automatic Teller Machine) scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103400382A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103679719A (en) * | 2013-12-06 | 2014-03-26 | 河海大学 | Image segmentation method |
| CN108596944A (en) * | 2018-04-25 | 2018-09-28 | 普联技术有限公司 | A kind of method, apparatus and terminal device of extraction moving target |
| CN116128740A (en) * | 2022-10-10 | 2023-05-16 | 深圳康荣电子有限公司 | Intelligent image processing method and system for liquid crystal display screen |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060182339A1 (en) * | 2005-02-17 | 2006-08-17 | Connell Jonathan H | Combining multiple cues in a visual object detection system |
| CN101211341A (en) * | 2006-12-29 | 2008-07-02 | 上海芯盛电子科技有限公司 | Image intelligent mode recognition and searching method |
| CN102360500A (en) * | 2011-07-08 | 2012-02-22 | 西安电子科技大学 | Treelet curvelet domain denoising- based method for change detection of remote sensing image |
| CN102509078A (en) * | 2011-10-28 | 2012-06-20 | 北京安控科技股份有限公司 | Fire detection device based on video analysis |
| CN102542673A (en) * | 2010-12-30 | 2012-07-04 | 广州市浩云安防科技股份有限公司 | Automatic teller machine (ATM) pre-warning method and system based on computer vision |
-
2013
- 2013-07-24 CN CN2013103180054A patent/CN103400382A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060182339A1 (en) * | 2005-02-17 | 2006-08-17 | Connell Jonathan H | Combining multiple cues in a visual object detection system |
| CN101211341A (en) * | 2006-12-29 | 2008-07-02 | 上海芯盛电子科技有限公司 | Image intelligent mode recognition and searching method |
| CN102542673A (en) * | 2010-12-30 | 2012-07-04 | 广州市浩云安防科技股份有限公司 | Automatic teller machine (ATM) pre-warning method and system based on computer vision |
| CN102360500A (en) * | 2011-07-08 | 2012-02-22 | 西安电子科技大学 | Treelet curvelet domain denoising- based method for change detection of remote sensing image |
| CN102509078A (en) * | 2011-10-28 | 2012-06-20 | 北京安控科技股份有限公司 | Fire detection device based on video analysis |
Non-Patent Citations (2)
| Title |
|---|
| 江增伟: "融合区域颜色和纹理特征的Mean Shift目标跟踪方法研究", 《中国优秀硕士学位论文全文数据库•信息科技辑》 * |
| 陈琼: "基于计算机视觉的ATM操作的异常行为检测", 《中国优秀硕士学位论文全文数据库•信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103679719A (en) * | 2013-12-06 | 2014-03-26 | 河海大学 | Image segmentation method |
| CN108596944A (en) * | 2018-04-25 | 2018-09-28 | 普联技术有限公司 | A kind of method, apparatus and terminal device of extraction moving target |
| CN116128740A (en) * | 2022-10-10 | 2023-05-16 | 深圳康荣电子有限公司 | Intelligent image processing method and system for liquid crystal display screen |
| CN116128740B (en) * | 2022-10-10 | 2024-05-14 | 深圳康荣电子有限公司 | Intelligent image processing method and system for liquid crystal display screen |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107527009B (en) | Remnant detection method based on YOLO target detection | |
| Del Pizzo et al. | Counting people by RGB or depth overhead cameras | |
| CN101325690A (en) | Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow | |
| CN107123131B (en) | Moving target detection method based on deep learning | |
| Sharif et al. | An entropy approach for abnormal activities detection in video streams | |
| Swathi et al. | Crowd behavior analysis: A survey | |
| CN107833221A (en) | A kind of water leakage monitoring method based on multi-channel feature fusion and machine learning | |
| CN101727570B (en) | Tracking method, track detection processing unit and monitor system | |
| CN101727672A (en) | Method for detecting, tracking and identifying object abandoning/stealing event | |
| CN101957325A (en) | Substation equipment appearance abnormality recognition method based on substation inspection robot | |
| CN103605967A (en) | Subway fare evasion prevention system and working method thereof based on image recognition | |
| CN106981150A (en) | A kind of supermarket's intelligent anti-theft system and method | |
| CN109948455A (en) | One kind leaving object detecting method and device | |
| CN103605971A (en) | Method and device for capturing face images | |
| CN104463869A (en) | Video flame image composite recognition method | |
| CN112270253A (en) | High-altitude parabolic detection method and device | |
| KR101472674B1 (en) | Method and apparatus for video surveillance based on detecting abnormal behavior using extraction of trajectories from crowd in images | |
| Malhi et al. | Vision based intelligent traffic management system | |
| CN103400120A (en) | Video analysis-based bank self-service area push behavior detection method | |
| Ullah et al. | Gaussian mixtures for anomaly detection in crowded scenes | |
| CN108830204B (en) | Method for detecting abnormality in target-oriented surveillance video | |
| CN103390151A (en) | Face detection method and device | |
| CN116419059A (en) | Automatic monitoring method, device, equipment and medium based on behavior label | |
| CN106650735B (en) | A kind of LED character automatic positioning recognition methods | |
| CN109583339A (en) | A kind of ATM video brainpower watch and control method based on image procossing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20131120 |
|
| WD01 | Invention patent application deemed withdrawn after publication |