CN102521578B - Method for detecting and identifying intrusion - Google Patents

Method for detecting and identifying intrusion Download PDF

Info

Publication number
CN102521578B
CN102521578B CN 201110427589 CN201110427589A CN102521578B CN 102521578 B CN102521578 B CN 102521578B CN 201110427589 CN201110427589 CN 201110427589 CN 201110427589 A CN201110427589 A CN 201110427589A CN 102521578 B CN102521578 B CN 102521578B
Authority
CN
China
Prior art keywords
identification
video
human
recognition
moving target
Prior art date
Application number
CN 201110427589
Other languages
Chinese (zh)
Other versions
CN102521578A (en
Inventor
卢林发
叶灿才
黄家祺
Original Assignee
中山爱科数字科技股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 中山爱科数字科技股份有限公司 filed Critical 中山爱科数字科技股份有限公司
Priority to CN 201110427589 priority Critical patent/CN102521578B/en
Publication of CN102521578A publication Critical patent/CN102521578A/en
Application granted granted Critical
Publication of CN102521578B publication Critical patent/CN102521578B/en

Links

Abstract

The invention discloses a method for detecting and identifying intrusion, which is applied to the field of video monitoring. By the aid of the method, human body identification, human face identification, cloud identification and manual assisted identification can be integrated, video sessions are divided, and representative frame images are selected; moving targets and areas are detected by the aid of background subtraction, and then a representative frame image which records a moving target with the size/shape closest to a threshold value is selected from a video session by the aid of a filter; the moving target is extracted from the representative frame image by the aid of background subtraction, then a head and shoulder two-dimensional model of a moving human body is built, and invariant moment of a contour of the model is computed to form feature vectors; and technical authorization of identification of a human body target and the like is carried out by a first classifier. Accordingly, video processing workload can be reduced, real-time identification and real-time alarm efficiency can be improved, and safety protection efficiency of families, communities and the like is enhanced.

Description

A kind of intrusion detection and recognition methods

Technical field

The present invention relates to technical field of video monitoring, particularly a kind of video frequency monitoring method of integrated multiple recognition technology.

Background technology

At present, video surveillance applications is very extensive.Family, office building, community etc., by automatic video monitoring, a large amount of savings manpower and materials.But deficiency is that present most of video monitoring can only carry out recording monitor to the environment in the guarded region, and can't the people in the zone be identified automatically, can't automatically identify the stranger, then takes corresponding security protection measure.

Carry out the people and automatically identify, technology commonly used comprises human body identification, recognition of face.Wherein the recognition of face discrimination is higher, but often need the identified person to ajust posture, then take pictures/record a video over against camera and just can accurately identify, but for invasion personnel such as thief, terrorists, realize that the standard face that finishes takes a picture, then identification is impossible.So, only adopting face recognition technology in the real-time video monitoring, obtained discrimination accuracy rate is very limited.With respect to recognition of face, human body identification it utilized the contour shape in shoulders of human body and above zone basicly stable, the advantage such as be not vulnerable to block, consider that bending moment does not have translation, Rotation and Zoom unchangeability, in order to process not homonymy and the variation of shoulder shape, set up the two-dimentional model of cognition of human body head and shoulder shape, and then identify.But, only use human body identification, for real-time dynamic intrusion detection and identification, or inadequate.

Summary of the invention

Purpose of the present invention is for above-mentioned existing issue, propose a kind of intrusion detection and recognition methods that is applied to field of video monitoring, it can integrated various human recognition technology, gives full play to simultaneously the distributed collaborative technology, in order to improve recognition efficiency, improve the security protection efficient of family, community etc.

The present invention is achieved by the following scheme:

A kind of intrusion detection and recognition methods, is characterized in that by video data acquiring, image recognition, last method of carrying out security control according to recognition result for successively, also comprise step:

A). by the real-time video in the imageing sensor collection area of visual field;

B). divide video-frequency band, and from video-frequency band, isolate the image of every frame, utilize background subtraction to divide to detect moving target and zone, then select the size/shape of moving target of record in the video-frequency band near the representative frame image of threshold values by filtrator;

C). utilize the background difference to extract moving target from this representative frame image, then set up the head shoulder two dimensional model of movement human and the invariant rectangle of computation model profile and become proper vector; Utilize the first sorter to carry out the identification of human body target;

D). according to the result of human body target identification, if can be judged to be the stranger, then master control system is reported to the police by control bus startup this locality or network alarming apparatus; If can not accurately determine whether the stranger, then carry out the recognition of face step;

E). when carrying out recognition of face, extract the overall situation or the local feature of people's face in the representative frame image;

If can extract the overall situation or the local feature of people's face, then judge the people's appearance image together that whether exists in the master control system database with moving target by the second sorter; Exist, then master control system allows it to carry out relative operation for this people authorizes; Otherwise starting this locality or network alarming apparatus reports to the police;

If can't be from the overall situation or the local feature of extraction people face in the representative frame image, then master control system starts the cloud identification step;

F). the cloud identification step, master control system sends to the cloud platform with this representative frame image, and then is forwarded to user's the network terminal by the cloud platform; The user utilizes the network terminal, carry out artificial cognition by human eye, if be judged as the people of understanding, allow master control system that this moving target is carried out authorization, otherwise be judged to be the stranger of invasion, the user reports to the police by network terminal control master control system or stops this moving target in the operation of system.

As preferably, described imageing sensor carries out video image acquisition with the speed of 5~15 frame/seconds; Described video-frequency band length is 1~5 minute.

Further, when the head shoulder two dimensional model of described movement human is set up, calculate first the ratio of width to height of moving target, and judge whether 0.28~0.36; Calculate the vertical direction projection histogram, find out near the local maximum in the crown, determine head width; Calculate at last a shoulder length degree, then set up head shoulder model; When extracting the failure of head shoulder model, then think the moving target that belongs to non-human body.

Further, during described recognition of face, the second used sorter integrates whole sorter for being walked abreast by global classification device and the local classifiers mode by weighted sum.The client that described user's the network terminal is installed by this locality receives the image from the cloud platform, carries out simultaneously the mutual and data transmission of control signal.

In sum, the present invention has following distinguishing feature:

1. employing multiple technologies, integrated human body identification, recognition of face, cloud identification and human assistance identification, but Effective Raise recognition efficiency and accuracy rate.

2. by dividing video-frequency band, choosing the technological means such as representative frame image, can reduce the Video processing amount, improve the efficient of Real time identification, Realtime Alerts.

Description of drawings

Fig. 1 is that the core procedure of inventive method forms synoptic diagram;

Fig. 2 is human body identification block diagram;

Fig. 3 is the particular flow sheet of inventive method.

Embodiment

With reference to figure 1, the inventive method is integrated human body identification, recognition of face, cloud recognition technology.When realizing intrusion detection and identification, at first by carrying out real-time video acquisition, the moving target that carries out in the zone is monitored, provide material for identification simultaneously.Then, video is carried out pre-service, comprising: divide video-frequency band, isolate every frame image, select representative frame image etc., last, successively by human body identification, recognition of face, cloud identification and the identification of user's artificial assistance.

With reference to figure 3, be the main flow process of realization of the present invention.At first:

See step 101, by the real-time video in the imageing sensor collection area of visual field; Acquisition rate is take the speed of 5~15 frame/seconds as suitable, and wherein preferred version was 10 frame/seconds.With traditional in order to improve real-time, and emphasize that acquisition rate is different, the video of collection of the present invention is processed for the Multiple recognition that satisfies the later stage, so acquisition rate needn't be too high, if the too high calculated amount that must strengthen identification is unfavorable for the raising of recognition accuracy and efficient on the contrary;

Step 102, the video processing module of master control system are cut apart the real-time video input that gathers and are divided into video-frequency band.Can divide in batches in the time of division, also can divide as required in order.Video-frequency band length is 1~5 minute.The zone that specifically can cover according to imageing sensor or picture pick-up device, then the assessor decides the length of video-frequency band at needed time of this zone of normally passing by.

Step 103 is isolated the image of every frame from video-frequency band, then utilize background subtraction to divide and detect moving target and zone, extracts moving target.Step 104 when adopting the background difference to extract moving target in this video-frequency band, also just illustrates the object that do not have moving target also just not identify in this time period, does not have the invader, so can carry out next video-frequency band processing, rotates back into 102; When proposing moving target, enter step 105;

Step 105, the filtrator by master control system are selected the size/shape of moving target of record in the video-frequency band near the representative frame image of threshold values.Representative frame image can be the frame of the moving target maximum of record in this video-frequency band, also can be record the moving target element at most, a profile frame the most clearly.Implementation can be done further definition.

Step 106 is extracted moving target from representative frame image;

Human body identification module in the step 107, master control system will be set up the head shoulder two dimensional model of this movement human and the invariant rectangle of computation model profile becomes proper vector.With reference to figure 2, for carrying out the human body whole FB(flow block) in identification time.After proper vector is extracted, will be that the BP network distributor is identified and the output category result by the first sorter; Wherein the BP network distributor is constantly updated or is revised by the BP network training from sample set.When the head shoulder two dimensional model of movement human is set up, calculate first the ratio of width to height of moving target, and judge whether 0.28~0.36; Calculate the vertical direction projection histogram, find out near the local maximum in the crown, determine head width; Calculate at last a shoulder length degree, then set up head shoulder model; When extracting the failure of head shoulder model, then think the moving target that belongs to non-human body.

Step 108 judges whether that can be able to form a head shoulder two dimensional model from representative frame image maybe extract proper vector, if could would carry out human body identification by the first sorter, forward 109 to; Otherwise the account for motion target is not the people, does not need to carry out next step recognition of face.

Step 109, the first sorter are carried out human body identification, if recognition result is the stranger, then forward step 110 to, and master control system starts this locality by control bus or network alarming apparatus is reported to the police, and forbids each generic operation that it is follow-up.If can't accurately identify, then start the recognition of face program, enter step 111.

Step 111, system are extracted the overall situation or the local feature of people's face in the representative frame image;

Can step 112, judgement extract the overall situation or the local feature of people's face, if can extract then forward 113 to, otherwise start cloud identification, enter 115;

Step 113 utilizes the second sorter to carry out recognition of face.The second sorter is judged the people's appearance image together that whether exists in the master control system database with moving target; Exist, then forward step 114 to, master control system allows it to carry out relative operation for this people authorizes; Otherwise starting this locality or network alarming apparatus reports to the police.If can't judge according to the face characteristic that extracts, then start cloud identification.The second sorter integrates whole sorter for being walked abreast by global classification device and the local classifiers mode by weighted sum.

Step 115, master control system sends to the cloud platform by the network image that representative frame is corresponding.

Step 116, cloud platform are forwarded to the network terminal corresponding to one or more users according to the forwarding mechanism of reaching an agreement between the master control system with this image, by artificial realization identification.

Step 117, the network terminal that the user answers is received this image, the user determines whether the people that is familiar with if it is directly or indirectly to feed back to master control system, allows this target mandate (step 119); Otherwise forbid the operation that it is follow-up, by starting this locality or network alarming apparatus report to the police (step 120).User's the network terminal can receive by the client that install this locality the image from the cloud platform, carries out simultaneously the mutual and data transmission of control signal.

Claims (6)

1. an intrusion detection and recognition methods, is characterized in that by video data acquiring, image recognition, the method for carrying out security control according to recognition result at last for successively, also comprises step:
A). by the real-time video in the imageing sensor collection area of visual field;
B). divide video-frequency band, and from video-frequency band, isolate the image of every frame, utilize background subtraction to divide to detect moving target and zone, then select the size/shape of moving target of record in the video-frequency band near the representative frame image of threshold values by filtrator;
C). utilize the background difference to extract moving target from this representative frame image, then set up the head shoulder two dimensional model of movement human and the invariant rectangle of computation model profile and become proper vector; Utilize the first sorter to carry out the identification of human body target;
D). according to the result of human body target identification, if can be judged to be the stranger, then master control system is reported to the police by control bus startup this locality or network alarming apparatus; If can not accurately determine whether the stranger, then carry out the recognition of face step;
E). when carrying out recognition of face, extract the overall situation or the local feature of people's face in the representative frame image;
If can extract the overall situation or the local feature of people's face, then judge the people's appearance image together that whether exists in the master control system database with moving target by the second sorter; Exist, then master control system is authorized and is allowed it to carry out relative operation; Otherwise starting this locality or network alarming apparatus reports to the police;
If can't be from the overall situation or the local feature of extraction people face in the representative frame image, then master control system starts the cloud identification step;
F). the cloud identification step, master control system sends to the cloud platform with this representative frame image, and then is forwarded to user's the network terminal by the cloud platform; The user utilizes the network terminal, carry out artificial cognition by human eye, if be judged as the people of understanding, allow master control system that this moving target is carried out authorization, otherwise be judged to be the stranger of invasion, the user reports to the police by network terminal control master control system or stops this moving target in the operation of system.
2. intrusion detection as claimed in claim 1 and recognition methods is characterized in that, described imageing sensor carries out video image acquisition with the speed of 5~15 frame/seconds.
3. intrusion detection as claimed in claim 1 or 2 and recognition methods is characterized in that, described video-frequency band length is 1~5 minute.
4. intrusion detection as claimed in claim 3 and recognition methods is characterized in that, when the head shoulder two dimensional model of described movement human is set up, calculate first the ratio of width to height of moving target, and judge whether 0.28~0.36; Calculate the vertical direction projection histogram, find out near the local maximum in the crown, determine head width; Calculate at last a shoulder length degree, then set up head shoulder model; When extracting the failure of head shoulder model, then think the moving target that belongs to non-human body.
5. intrusion detection as claimed in claim 4 and recognition methods is characterized in that, during described recognition of face, and the whole sorter of the second used sorter for walking abreast and integrate by global classification device and the local classifiers mode by weighted sum.
6. such as claim 1,2,4,5 arbitrary described intrusion detection and recognition methodss, it is characterized in that the client that described user's the network terminal is installed by this locality receives the image from the cloud platform, carry out simultaneously the mutual and data transmission of control signal.
CN 201110427589 2011-12-19 2011-12-19 Method for detecting and identifying intrusion CN102521578B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201110427589 CN102521578B (en) 2011-12-19 2011-12-19 Method for detecting and identifying intrusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201110427589 CN102521578B (en) 2011-12-19 2011-12-19 Method for detecting and identifying intrusion

Publications (2)

Publication Number Publication Date
CN102521578A CN102521578A (en) 2012-06-27
CN102521578B true CN102521578B (en) 2013-10-30

Family

ID=46292489

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201110427589 CN102521578B (en) 2011-12-19 2011-12-19 Method for detecting and identifying intrusion

Country Status (1)

Country Link
CN (1) CN102521578B (en)

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104010106A (en) * 2014-04-10 2014-08-27 福建伊时代信息科技股份有限公司 Copying machine security monitoring method and system based on real-time video analysis
CN104079881B (en) * 2014-07-01 2017-09-12 中磊电子(苏州)有限公司 The relative monitoring method of supervising device
CN105744345B (en) * 2014-12-12 2019-05-31 深圳Tcl新技术有限公司 Video-frequency compression method and device
CN104463174A (en) * 2014-12-16 2015-03-25 广州南方电力集团科技发展有限公司 Multi-angle dynamic people recognition and behavior prediction system
CN105006089B (en) * 2015-07-01 2017-11-10 国家电网公司 A kind of security monitoring alarm method and system based on image
CN105100724B (en) * 2015-08-13 2018-06-19 电子科技大学 A kind of smart home telesecurity monitoring method of view-based access control model analysis
CN105243773A (en) * 2015-09-25 2016-01-13 国网山东省电力公司经济技术研究院 Portable intelligent alarm fence and human proximity detection method
CN105654647B (en) * 2016-01-28 2017-11-28 中北大学 A kind of real-time judge someone invades indoor recognition methods
JP6340538B2 (en) * 2016-03-11 2018-06-13 株式会社プロドローン Biological search system
CN105979230A (en) * 2016-07-04 2016-09-28 上海思依暄机器人科技股份有限公司 Monitoring method and device realized through images by use of robot
CN106372576A (en) * 2016-08-23 2017-02-01 南京邮电大学 Deep learning-based intelligent indoor intrusion detection method and system
CN106781449A (en) * 2017-02-21 2017-05-31 青岛智能产业技术研究院 Crossing pedestrian crosses the street integrated management control system
CN107610392A (en) * 2017-09-20 2018-01-19 北京亚欧震达科技发展有限公司 A kind of pedestrian detection of motor car inspection and repair storehouse pipe gallery and the apparatus and method of alarm
CN109460787B (en) * 2018-10-26 2020-08-18 北京交通大学 Intrusion detection model establishing method and device and data processing equipment
CN109359625A (en) * 2018-11-16 2019-02-19 南京甄视智能科技有限公司 The method and system of customer identification is judged based on head and shoulder detection and face recognition technology

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6181806B1 (en) * 1993-03-29 2001-01-30 Matsushita Electric Industrial Co., Ltd. Apparatus for identifying a person using facial features
CN101174298A (en) * 2006-11-02 2008-05-07 上海银晨智能识别科技有限公司 Scattered-point high-volume face recognition system and recognizing method thereof

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6181806B1 (en) * 1993-03-29 2001-01-30 Matsushita Electric Industrial Co., Ltd. Apparatus for identifying a person using facial features
CN101174298A (en) * 2006-11-02 2008-05-07 上海银晨智能识别科技有限公司 Scattered-point high-volume face recognition system and recognizing method thereof

Also Published As

Publication number Publication date
CN102521578A (en) 2012-06-27

Similar Documents

Publication Publication Date Title
US10586113B2 (en) Stationary target detection by exploiting changes in background model
US10546186B2 (en) Object tracking and best shot detection system
US9665777B2 (en) System and method for object and event identification using multiple cameras
US10009579B2 (en) Method and system for counting people using depth sensor
US9197861B2 (en) Multi-dimensional virtual beam detection for video analytics
US8774532B2 (en) Calibration of video object classification
CN104123544B (en) Anomaly detection method and system based on video analysis
Ihaddadene et al. Real-time crowd motion analysis
US8438175B2 (en) Systems, methods and articles for video analysis reporting
US9363483B2 (en) Method for available parking distance estimation via vehicle side detection
US9070023B2 (en) System and method of alerting a driver that visual perception of pedestrian may be difficult
US20170270674A1 (en) System and method for training object classifier by machine learning
CN104123536B (en) System and method for image analysis
ES2522589T3 (en) Behavioral Recognition System
US20120321138A1 (en) Suspicious behavior detection system and method
ES2364915T3 (en) Video trap cable.
US8249301B2 (en) Video object classification
US7660439B1 (en) Method and system for flow detection and motion analysis
KR101085578B1 (en) Video tripwire
US8300890B1 (en) Person/object image and screening
CN105378752B (en) The on-line study system for detecting and counting for people
US20120188370A1 (en) Surveillance systems and methods to monitor, recognize, track objects and unusual activities in real time within user defined boundaries in an area
US8682036B2 (en) System and method for street-parking-vehicle identification through license plate capturing
US7683929B2 (en) System and method for video content analysis-based detection, surveillance and alarm management
JP5390322B2 (en) Image processing apparatus and image processing method

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
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20180802

Address after: 528400 one of the 25 floors of the 1 building of Tazhong square, No. 1 East Convention and exhibition hall, Torch Development Zone, Zhongshan, Guangdong, China.

Patentee after: Guangdong Zen science and Technology Co., Ltd.

Address before: 528400 702, room 7, No. 16 Convention and Exhibition Road, Torch Development Zone, Zhongshan, Guangdong

Patentee before: Zhongshan Iker Digital Technology Co., Ltd.

TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20190328

Address after: 510000 Room 2609,2610, 30 Huaxia Road, Tianhe District, Guangzhou City, Guangdong Province (office only)

Patentee after: Guangdong Xuzhi Travel Technology Co., Ltd.

Address before: 528400 one of the 25 floors of the 1 building of Tazhong square, No. 1 East Convention and exhibition hall, Torch Development Zone, Zhongshan, Guangdong, China.

Patentee before: Guangdong Zen science and Technology Co., Ltd.