CN109191768A - A kind of kinsfolk's security risk monitoring method based on deep learning - Google Patents
A kind of kinsfolk's security risk monitoring method based on deep learning Download PDFInfo
- Publication number
- CN109191768A CN109191768A CN201811052053.2A CN201811052053A CN109191768A CN 109191768 A CN109191768 A CN 109191768A CN 201811052053 A CN201811052053 A CN 201811052053A CN 109191768 A CN109191768 A CN 109191768A
- Authority
- CN
- China
- Prior art keywords
- face
- kinsfolk
- deep learning
- topography
- security risk
- 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
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B19/00—Alarms responsive to two or more different undesired or abnormal conditions, e.g. burglary and fire, abnormal temperature and abnormal rate of flow
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
- Alarm Systems (AREA)
Abstract
The present invention discloses a kind of kinsfolk's security risk monitoring method based on deep learning, the method is based on customer data base constructed by each member's facial characteristics of family and behavior, following face characteristic matching process and behavioral value method are executed, realizes the discovery of kinsfolk's security risk；The invention has the following advantages that (1) realizes effectively discovery and early warning for kinsfolk's security risk.(2) it is providing except face identification functions, also offer unusual checking function.(3) behavioral value method along the dense trajectory extraction behavior of behavior point of interest external appearance characteristic and motion feature.
Description
Technical field
Kinsfolk's security risk monitoring method based on deep learning that the present invention relates to a kind of belongs to image vision detection
Analysis technical field.
Background technique
In recent years, as social senilization's problem is aggravated, more and more old men stay in alone family, because older, solely
From being in, when something unexpected happened, (such as burst disease, the events such as gas leakage) are more intractable, or even will cause personal safety
Problem, there are huge hidden danger.It is difficult to understand the security situation of old man there are household, meets with emergency case alarm, first aid is stranded
The problems such as difficult.In addition to this, nowadays there is also stranger's privates to rush private residence for society, and nurse maltreats the thing of child, old man, for
These things need a kind of monitoring method supervised people's behavior at home, guarantee household's life security.
Summary of the invention
Kinsfolk's security risk monitoring based on deep learning that technical problem to be solved by the invention is to provide a kind of
Method can effectively ensure that the safety of member in family, avoid the generation of security risk.
In order to solve the above-mentioned technical problem the present invention uses following technical scheme: a kind of kinsfolk based on deep learning
Security risk monitoring method, the method are held based on customer data base constructed by each member's facial characteristics of family and behavior
The following face characteristic matching process of row and behavioral value method realize the discovery of kinsfolk's security risk；
The face characteristic matching process, comprising:
Step A1. is used and is trained face detection module obtained in advance, each frame captured for monitor camera device
Video image carries out Face datection, obtains each face topography, and enter step A2；
Step A2. is directed to each face topography respectively, is shifted for the face in face topography, makes it
Face position is identical as face position of the preset standard to its facial image, thus updates each face topography respectively, so
After enter step A3；
Step A3. is directed to each face topography respectively, using preparatory trained deep learning network model, structure
Face characteristic extraction module is built, face feature vector corresponding to face topography is extracted by face characteristic extraction module,
And match it with kinsfolk's facial characteristics in customer data base, confirm whether the face topography is real
Kinsfolk realizes that bad person invades the discovery of private residence security risk hereby based on the characteristic matching for being directed to each face topography；
The behavioral value method, comprising:
Step B1: it based on resulting each face topography is updated in step A2, is captured in conjunction with monitor camera device
Continuous videos image frame each kinsfolk's moving target is obtained by the edge Canny acquiring method and Background difference；
Step B2: abnormal behaviour sample action is preset using public database CAVAIAR and BOSS, such as falls in a swoon, fall down, take out
It the abnormal behaviours such as jerks, hit the person, being trained for predetermined deep learning network, obtain trained deep learning network, i.e.,
Abnormal behaviour motion detection model；
Step B3: being directed to each kinsfolk's moving target respectively, is directed to family using abnormal behaviour motion detection model
Member's moving target carries out abnormal behaviour motion detection, judges whether that the kinsfolk acts with the presence or absence of abnormal behaviour, realizes
The discovery of kinsfolk's security risk.
Preferably,
In the step A1: using default kinsfolk's Face datection sample, using HOG algorithm, at least two
Face datection unit based on hog feature, is trained respectively, obtains each trained Face datection unit；Then into
Row cascade, obtains multi-stage cascade regression tree formula Face datection unit；Finally using default Face datection sample, for multi-stage cascade
Regression tree formula Face datection unit carries out face regression training, obtains face detection module.
Preferably,
In the step A2: each face topography is directed to respectively, using two dimensional affine transformation for face part
Face in image are shifted, and are kept its face position identical as face position of the preset standard to its facial image, are thus divided
Each face topography is not updated.
Preferably,
In the step A3: acquiring the face-image of kinsfolk, and carry out size initialization operation, then construct house
Front yard member's face training sample data；It is finely adjusted for deep learning frame caffe and open source vgg model, obtains default frame
Structure, the deep learning network model corresponding to kinsfolk's facial feature extraction, as face characteristic extraction module.
Preferably,
The behavioral value method, firstly, using accumulation denoising encoder along dense trajectory extraction appearance of depth feature
With Depth Motion feature；Then, in order to improve the classification capacity of feature, this two kinds of features are carried out using weighted correlation method
Fusion；Finally, carrying out unusual checking using sparse reconstruction.
The present invention is due to taking above technical scheme, kinsfolk's security risk monitoring method based on deep learning, tool
It has the advantage that
(1) effectively discovery and early warning are realized for kinsfolk's security risk, face is identified using deep learning, resisted
Interference performance is strong, can improve discrimination, more accurate identification from the ability of a few sample collection learning data set feature essence
Stranger, and then can effectively notify the behavior that stranger enters in family at once.
(2) it is providing except face identification functions, also offer unusual checking function, is introducing and calculated using deep learning
Method, for falling in a swoon, falling down, twitch, hit the person etc., abnormal behaviours realize detection, find the fortuitous event of kinsfolk in time, and
And it is able to detect maltreatment of the nurse to kinsfolk, timely early warning is carried out for safety accident.
(3) behavioral value method along the dense trajectory extraction behavior of behavior point of interest external appearance characteristic and motion feature,
It on the one hand, can using the powerful learning ability of accumulation denoising encoder due to there is motion information abundant near dense track
To extract more effective behavioural characteristic；On the other hand, it is not directly to extract entire behavioural characteristic with depth network, only extracts row
For the feature of the sampled point in region, and the number of these sampled points is enough to train depth network, therefore does not need a large amount of sample
This training depth network solves influence of the lack of training samples to deep learning.
Detailed description of the invention
Fig. 1 show the flow chart of the application face feature matching method；
Fig. 2 show the flow chart of the application behavioral value method.
Specific embodiment
The present invention is described in further detail below in conjunction with the drawings and specific embodiments.It should be appreciated that described herein
Specific embodiment be only used to explain the present invention, be not intended to limit the present invention.
It should be noted that " connection " described herein and the word for expressing " connection ", as " being connected ",
" connected " etc. had both included that a certain component is directly connected to another component, and had also included that a certain component passes through other component and another portion
Part is connected.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also be intended to include plural form, additionally, it should be understood that, when in the present specification using belong to "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, component or module, component and/or their combination.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.
As Figure 1-Figure 2, kinsfolk's security risk monitoring method provided in this embodiment based on deep learning, step
Suddenly include:
(1) it according to shown in Fig. 1, using face detection module obtained is trained in advance, is caught for monitor camera device
Each frame video image obtained carries out kinsfolk's Face datection, each face topography is obtained, as training sample.
(2) it is directed to each face topography respectively, is shifted for the face in face topography, makes its face
Position is identical as face position of the preset standard to its facial image, thus updates each face topography respectively, and sample is pre-
Processing.
(3) it is directed to each face topography respectively, using preparatory trained deep learning network model, constructs people
Face characteristic extracting module extracts face feature vector corresponding to face topography by face characteristic extraction module, and will
It is matched with kinsfolk's facial characteristics in customer data base, confirms whether the face topography is real family
Member realizes that bad person invades the discovery of private residence security risk hereby based on the characteristic matching for being directed to each face topography.
(4) according to step shown in Fig. 2, on the basis of being based on kinsfolk's face characteristic matching process, abnormal row is carried out
For detection, based on updating resulting each face topography, in conjunction with the continuous videos image frame that monitor camera device is captured,
By the edge Canny acquiring method and Background difference, each kinsfolk's moving target is obtained；
(5) abnormal behaviour sample action is preset using public database CAVAIAR and BOSS, such as fall in a swoon, fall down, twitching,
It the abnormal behaviours such as hits the person, is trained for predetermined deep learning network, obtains trained deep learning network, i.e., extremely
Behavior act detection model；
(6) it is directed to each kinsfolk's moving target respectively, kinsfolk is directed to using abnormal behaviour motion detection model
Moving target carries out abnormal behaviour motion detection, judges whether that the kinsfolk acts with the presence or absence of abnormal behaviour, realizes family
The discovery of member security's hidden danger.
The above is only a preferred embodiment of the present invention, it is noted that for the common skill of the art
For art personnel, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications
Also it should be regarded as protection scope of the present invention.
Claims (5)
1. a kind of kinsfolk's security risk monitoring method based on deep learning, which is characterized in that the method is based on family
Customer data base constructed by each member's facial characteristics and behavior executes following face characteristic matching process and behavioral value side
Method realizes the discovery of kinsfolk's security risk；
The face characteristic matching process, comprising:
Step A1. is used and is trained face detection module obtained in advance, each frame video captured for monitor camera device
Image carries out Face datection, obtains each face topography, and enter step A2；
Step A2. is directed to each face topography respectively, is shifted for the face in face topography, its face is made
Position is identical as face position of the preset standard to its facial image, thus updates each face topography respectively, then into
Enter step A3；
Step A3. is directed to each face topography respectively, using preparatory trained deep learning network model, constructs people
Face characteristic extracting module extracts face feature vector corresponding to face topography by face characteristic extraction module, and will
It is matched with kinsfolk's facial characteristics in customer data base, confirms whether the face topography is real family
Member realizes that bad person invades the discovery of private residence security risk hereby based on the characteristic matching for being directed to each face topography；
The behavioral value method, comprising:
Step B1: based on updating resulting each face topography, the company captured in conjunction with monitor camera device in step A2
Continuous video image frame obtains each kinsfolk's moving target by the edge Canny acquiring method and Background difference；
Step B2: presetting abnormal behaviour sample action using public database CAVAIAR and BOSS, such as fall in a swoon, fall down, twitching,
It the abnormal behaviours such as hits the person, is trained for predetermined deep learning network, obtains trained deep learning network, i.e., extremely
Behavior act detection model；
Step B3: being directed to each kinsfolk's moving target respectively, is directed to kinsfolk using abnormal behaviour motion detection model
Moving target carries out abnormal behaviour motion detection, judges whether that the kinsfolk acts with the presence or absence of abnormal behaviour, realizes family
The discovery of member security's hidden danger.
2. a kind of kinsfolk's security risk monitoring method based on deep learning according to claim 1, feature exist
In,
In the step A1: being based on using HOG algorithm at least two using default kinsfolk's Face datection sample
The Face datection unit of hog feature, is trained respectively, obtains each trained Face datection unit；Then grade is carried out
Connection obtains multi-stage cascade regression tree formula Face datection unit；Finally using default Face datection sample, returned for multi-stage cascade
Tree formula Face datection unit carries out face regression training, obtains face detection module.
3. a kind of kinsfolk's security risk monitoring method based on deep learning according to claim 1, feature exist
In,
In the step A2: being directed to each face topography respectively, be directed to face topography using two dimensional affine transformation
In face shifted, keep its face position identical as face position of the preset standard to its facial image, thus respectively more
New each face topography.
4. a kind of kinsfolk's security risk monitoring method based on deep learning according to claim 1, feature exist
In,
In the step A3: acquire the face-image of kinsfolk, and carry out size initialization operation, then construct family at
The facial training sample data of member；It is finely adjusted for deep learning frame caffe and open source vgg model, obtains pre-set configuration, right
It should be in the deep learning network model of kinsfolk's facial feature extraction, as face characteristic extraction module.
5. a kind of kinsfolk's security risk monitoring method based on deep learning according to claim 1, feature exist
In,
The behavioral value method, firstly, using accumulation denoising encoder along dense trajectory extraction appearance of depth feature and depth
Spend motion feature；Then, in order to improve the classification capacity of feature, this two kinds of features are melted using weighted correlation method
It closes；Finally, carrying out unusual checking using sparse reconstruction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811052053.2A CN109191768A (en) | 2018-09-10 | 2018-09-10 | A kind of kinsfolk's security risk monitoring method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811052053.2A CN109191768A (en) | 2018-09-10 | 2018-09-10 | A kind of kinsfolk's security risk monitoring method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109191768A true CN109191768A (en) | 2019-01-11 |
Family
ID=64915930
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811052053.2A Pending CN109191768A (en) | 2018-09-10 | 2018-09-10 | A kind of kinsfolk's security risk monitoring method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109191768A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111274883A (en) * | 2020-01-10 | 2020-06-12 | 杭州电子科技大学 | Synthetic sketch face recognition method based on multi-scale HOG (histogram of oriented gradient) features and deep features |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050068175A1 (en) * | 2002-07-08 | 2005-03-31 | Faulkner James Otis | Security system and method with realtime imagery |
| CN103198605A (en) * | 2013-03-11 | 2013-07-10 | 成都百威讯科技有限责任公司 | Indoor emergent abnormal event alarm system |
| CN103984315A (en) * | 2014-05-15 | 2014-08-13 | 成都百威讯科技有限责任公司 | Domestic multifunctional intelligent robot |
| CN106059868A (en) * | 2016-07-24 | 2016-10-26 | 哈尔滨理工大学 | Home intelligent video monitoring protection system |
| CN107563281A (en) * | 2017-07-24 | 2018-01-09 | 南京邮电大学 | A kind of construction site personal security hidden danger monitoring method based on deep learning |
-
2018
- 2018-09-10 CN CN201811052053.2A patent/CN109191768A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050068175A1 (en) * | 2002-07-08 | 2005-03-31 | Faulkner James Otis | Security system and method with realtime imagery |
| CN103198605A (en) * | 2013-03-11 | 2013-07-10 | 成都百威讯科技有限责任公司 | Indoor emergent abnormal event alarm system |
| CN103984315A (en) * | 2014-05-15 | 2014-08-13 | 成都百威讯科技有限责任公司 | Domestic multifunctional intelligent robot |
| CN106059868A (en) * | 2016-07-24 | 2016-10-26 | 哈尔滨理工大学 | Home intelligent video monitoring protection system |
| CN107563281A (en) * | 2017-07-24 | 2018-01-09 | 南京邮电大学 | A kind of construction site personal security hidden danger monitoring method based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 王军等: "基于深度学习的异常行为检测", 《湖南大学学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111274883A (en) * | 2020-01-10 | 2020-06-12 | 杭州电子科技大学 | Synthetic sketch face recognition method based on multi-scale HOG (histogram of oriented gradient) features and deep features |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107818312A (en) | A kind of embedded system based on abnormal behaviour identification | |
| CN103839373B (en) | A kind of unexpected abnormality event Intelligent Recognition alarm device and warning system | |
| CN110569720B (en) | Audio and video intelligent identification processing method based on audio and video processing system | |
| CN202130312U (en) | Driver fatigue driving monitoring device | |
| CN106241534B (en) | More people's boarding abnormal movement intelligent control methods | |
| CN105279483A (en) | Fall-down behavior real-time detection method based on depth image | |
| CN106355154B (en) | Method for detecting frequent passing of people in surveillance video | |
| CN108364374A (en) | Face access control device based on deep learning and method | |
| CN110674761B (en) | Regional behavior early warning method and system | |
| CN109271961A (en) | Behavior pattern judgment method based on the identification of piece identity's feature | |
| CN106228709B (en) | A kind of wisdom gold eyeball identifies that one adds paper money alarm method and device | |
| CN103839049A (en) | Double-person interactive behavior recognizing and active role determining method | |
| CN116994390A (en) | Security monitoring system and method based on Internet of things | |
| CN103632479A (en) | Home-based care for the aged security and protection system based on human motion trajectory analysis | |
| CN116563986A (en) | Intelligent building access control system and method | |
| Zambanini et al. | Detecting falls at homes using a network of low-resolution cameras | |
| CN109191768A (en) | A kind of kinsfolk's security risk monitoring method based on deep learning | |
| CN103247150A (en) | Fatigue driving preventing system | |
| CN206431770U (en) | A kind of mall entrance intelligent safety and defence system | |
| CN108171187A (en) | A kind of abnormal behaviour automatic identifying method and device based on the extraction of bone point | |
| CN109670435B (en) | Method for identifying identity based on human gait characteristics | |
| CN114821785B (en) | Potential sudden criminal identification method based on hypothesis evaluation and Bayesian learning | |
| Cai et al. | A novel method based on optical flow combining with wide residual network for fall detection | |
| CN111666780A (en) | Intelligent door control security method based on emotion recognition technology | |
| Nagamani et al. | Anti-Theft Monitoring for a Smart Home |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190111 |