CN110197158A - A kind of security cloud system and its application - Google Patents
A kind of security cloud system and its application Download PDFInfo
- Publication number
- CN110197158A CN110197158A CN201910469910.7A CN201910469910A CN110197158A CN 110197158 A CN110197158 A CN 110197158A CN 201910469910 A CN201910469910 A CN 201910469910A CN 110197158 A CN110197158 A CN 110197158A
- Authority
- CN
- China
- Prior art keywords
- face
- layer
- subnet
- unit
- security
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
- G06V40/172—Classification, e.g. identification
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B13/00—Burglar, theft or intruder alarms
- G08B13/18—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength
- G08B13/189—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems
- G08B13/194—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems
- G08B13/196—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems using television cameras
- G08B13/19602—Image analysis to detect motion of the intruder, e.g. by frame subtraction
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B13/00—Burglar, theft or intruder alarms
- G08B13/18—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength
- G08B13/189—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems
- G08B13/194—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems
- G08B13/196—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems using television cameras
- G08B13/19639—Details of the system layout
- G08B13/19645—Multiple cameras, each having view on one of a plurality of scenes, e.g. multiple cameras for multi-room surveillance or for tracking an object by view hand-over
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Alarm Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of security cloud system and its applications, belong to intelligent security field, including public security alarm unit, security prewarning unit, recognition of face positioning training unit and face snap locating and tracking unit, the public security alarm unit and security prewarning unit are connect with recognition of face positioning training unit, and the recognition of face positioning training unit is connected with face snap locating and tracking unit.By the way that real-time face snap locating and tracking unit is arranged, face in scene is captured, and human face data is disclosed with public security system and is compared, when having found in public security system via the personnel to put on record, it is carried out issuing warning information or warning message etc. according to the different stage for the personnel that put on record, the staff for the informing security system that can shift to an earlier date, preferably prevent the thing to upset the public from occurring, possible trouble can be taken precautions against, preferably prevents the medical trouble, community's multiple level marketing and school of hospital from abducting child.
Description
Technical field
The present invention relates to intelligent security field more particularly to a kind of security cloud system and its applications.
Background technique
Existing many security systems are to reach the supervision to some environment by the monitoring of a scene status.But
It is this supervision is that just can be carried out the corresponding personnel of alert notice after violence or crime dramas occur, it cannot pair in advance
Personnel's situation discriminance analysis of some environment, so that early warning in advance, more effectively prevents incident of violence or crime dramas
Occur.
Such as the medical trouble personnel of present hospital, many hospitals are because some making trouble for professional medical trouble people hamper normally
Medical Order, while needing to ensure public security accordingly after medical trouble outburst and just can be carried out control, it is difficult to manage.Therefore, in many fields
Scape field could not in advance supervise corresponding personnel, while not link up well with public security department, could not data interchange,
Social unstability time advance cannot be informed and corresponding personnel are supervised.
Summary of the invention
It, cannot be to society to solve that existing security is pained the purpose of the present invention is to provide a kind of security cloud system and its application
The technical issues of meeting personnel carry out recognition of face monitoring, and the informing that gives warning in advance is supervised.
A kind of security cloud system, including public security alarm unit, security prewarning unit, recognition of face positioning training unit and people
Face captures locating and tracking unit, and the public security alarm unit and security prewarning unit connect with recognition of face positioning training unit
It connects, the recognition of face positioning training unit is connected with face snap locating and tracking unit, and the public security alarm unit is for mentioning
Training unit is positioned to recognition of face for the human face data being identified in each scene, while it is single to receive recognition of face positioning training
First alarm signal automatic alarm, display alarm information, the face snap locating and tracking unit are passed for acquiring human face data
Training unit is positioned to recognition of face, is carried out according to the tracking data that recognition of face positions training unit outflow to corresponding face
Tracked, recognition of face positioning training unit be used to receive the human face data of face snap locating and tracking unit acquisition with
The identified human face data that public security alarm unit provides compares, and when contrasting by identical human face data, issues early warning letter
Number inform security prewarning unit and/or public security alarm unit, security prewarning unit is for receiving recognition of face positioning training unit
Early warning or warning message alarmed and notify Security Personnel.
Further, the face snap locating and tracking unit includes face acquisition module and face tracking locating module,
The face acquisition module is connect with face tracking locating module, the face tracking locating module control face acquisition module rotation
Turn to track face.
Further, the recognition of face positioning training unit includes data memory module, face recognition module and face
Training module is positioned, the data memory module is connect with face recognition module, and the face recognition module and Face detection are instructed
Practice module connection, the data memory module provides identified human face data information, the people for storing public security alarm unit
Face identification module is used to the human face data of face acquisition module acquisition back compare with identified human face data, works as comparison
When by same or similar data, issues early warning or alarm signal is transmitted to public security alarm unit and/or security prewarning unit, institute
Face detection training module is stated for training pattern location model, location model is provided and is tracked to face tracking locating module
Positioning.
Further, the detailed process of the Face detection training module training location model are as follows:
The twin network model of pyramid is constructed, the training twin network model of pyramid tests the twin network model of pyramid,
Training and the twin network model of pyramid tested are transferred to the realization of face tracking locating module to track face.
Further, the twin network model of the pyramid by twin network, feature pyramid network and classification and orientation simultaneously
Row network composition, twin network is made of the subnet that two VGG are constituted, and the subnet that two VGG are constituted shares identical parameter, and two
Feature of the subnet that a VGG is constituted for respectively to target image and search image extracts, and twin network completes target figure
After the feature extraction of picture and search image, the target signature layer and search characteristics layer of different scale are obtained respectively, from different levels
With in the characteristic layer of different scale extract 6 layers of feature, for constructing pyramid network；
After the completion of feature pyramid network struction, in conjunction with classification and orientation parallel network, for the real-time positioning to target
And tracking, classification and orientation parallel network return subnet by candidate frame subnet, classifier subnet and positioning and form, candidate frame subnet,
Classifier subnet and positioning return subnet and generate candidate frame, confidence level and coordinate shift amount, and classifier subnet and positioning respectively
Subnet is returned to execute parallel.
Further, the subnet that described two VGG are constituted is targeted subnet and search subnet, to target image and is searched respectively
Rope image carries out feature extraction, and shares identical weight and biasing, and targeted subnet and search subnet are by eleventh floor convolution
Layer is constituted, and eleventh floor convolutional layer is respectively: first layer is made of 2 convolution units, and the second layer is made of 2 convolution units, the
Three layers are made of 3 convolution units, and the 4th layer is made of 3 convolution units, and layer 5 is made of 3 convolution units, layer 6
It is made of 1 convolution unit, layer 7 is made of 1 convolution unit, and the 8th layer is made of 2 convolution units, and the 9th layer by 2
Convolution unit is constituted, and the tenth layer is made of 2 convolution units, and eleventh floor is made of 2 convolution units；
The feature pyramid network is made of the characteristic layer that targeted subnet and search subnet obtain, and total number of plies is 6 layers,
Be respectively: first layer is made of the eleventh floor characteristic layer in targeted subnet；The second layer is by the tenth layer of feature in search subnet
Layer is constituted；Third layer is made of the layer 7 characteristic layer in targeted subnet；4th layer by the layer 6 characteristic layer in search subnet
It constitutes；Layer 5 is made of the 4th layer of characteristic layer in targeted subnet；Layer 6 is by the third layer characteristic layer structure in search subnet
At,
The classification and orientation parallel network returns subnet by candidate frame subnet, classifier subnet and positioning and forms, candidate frame
Subnet is made of candidate frame, and for predicting the ability of target, classifier subnet is made of normalization exponential function classifier, is used
In distinguishing target and non-targeted ability, positioning returns subnet and is made of 3x3 convolution kernel, for positioning target, wherein candidate frame
By every tomographic image Image Segmentation Methods Based on Features at n x n grid, n is positive integer, and each grid generates the candidate frame of 6 fixed sizes, is waited
It selects frame to return subnet by classifier subnet and positioning and generates confidence level and coordinate shift amount respectively.
Further, the detailed process of the training twin network model of pyramid are as follows:
Original video sequence is obtained from video database, and image preprocessing is carried out to video sequence, obtains target training
Collection and search training set, and the target of training set is all in the center of image；
After completing training set image procossing, pairs of target training set and search training set picture are input to twin network
Corresponding subnet obtains target signature layer and search characteristics layer, extracts the characteristic layer of different levels and different scale, constructs gold
Word tower network, meanwhile, it is based on candidate frame size formula and location formula, is constructed in every layer of characteristic layer of pyramid network
The different size of candidate frame of different location out；
Each layer of characteristic layer of pyramid network is input to classification and orientation parallel network, obtains the output knot of parallel network
The result of output and label true value are carried out similarity mode, obtain positive sample and negative sample by fruit；
The error between matching result and label true value is calculated using target loss function, and the error is successively reversed
Input layer is propagated to, meanwhile, based on the weight and biasing in small lot stochastic gradient descent optimization algorithm adjustment network, obtain most
Excellent error amount completes primary network model training；
It repeats the above steps, until the error amount of target loss function converges on minimum value.
Further, the public security alarm unit includes public security database module and public security automatic alarm module, the public affairs
Peace database module is connect with public security automatic alarm module, and the public security database module is used to store what public security inside was identified
Personnel's human face data and related labeled concerns information, the public security automatic alarm module are used for automatic alarm, lead to
Know local police station's responding.
A kind of application of security cloud system, above-mentioned system are applied in school's security system, hospital's security system, cell peace
Insurance system or industry park security system.
Present invention employs above-mentioned technical proposal, the present invention is had following technical effect that
The present invention captures the face in scene by the real-time face snap locating and tracking unit of setting, and with
Public security system discloses human face data and compares, when having found in public security system via the personnel to put on record, according to the people that puts on record
The different stage of member carries out issuing warning information or warning message etc., the staff for the informing security system that can shift to an earlier date,
Preferably prevent the thing to upset the public from occurring, possible trouble can be taken precautions against, preferably prevent the medical trouble of hospital, community's multiple level marketing and
Child abducts in school.
Detailed description of the invention
Fig. 1 is present system block diagram.
Fig. 2 is that recognition of face of the present invention positions training unit and face snap locating and tracking unit module block diagram.
Fig. 3 is present system working timing figure.
Fig. 4 is public security alarm unit module frame chart of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, referring to the drawings and preferred reality is enumerated
Example is applied, the present invention is described in more detail.However, it is necessary to illustrate, many details listed in specification are only to be
Reader is set to have a thorough explanation to one or more aspects of the present invention, it can also be with even without these specific details
Realize the aspects of the invention.
Referring to Fig. 1, the present invention provides a kind of security cloud system, including public security alarm unit, security prewarning unit, face
Identification positioning training unit and face snap locating and tracking unit.Public security alarm unit and security prewarning unit are and recognition of face
Position training unit connection.Recognition of face positioning training unit is connected with face snap locating and tracking unit.Public security alarm unit
Training unit is positioned to recognition of face for providing the human face data being identified in each scene, while receiving recognition of face positioning
Training unit alarm signal automatic alarm, display alarm information.Face snap locating and tracking unit is used to acquire human face data, and
It is transmitted to recognition of face positioning training unit, is carried out according to the tracking data that recognition of face positions training unit outflow to corresponding people
Face is tracked.Recognition of face positioning training unit is used to receive the human face data and public affairs of face snap locating and tracking unit acquisition
The identified human face data that peace alarm unit provides compares, and when contrasting by identical human face data, issues warning signal
Inform security prewarning unit and/or public security alarm unit, security prewarning unit is used to receive recognition of face positioning training unit
Early warning or warning message are alarmed and notify Security Personnel.Security prewarning unit is each scene, such as the security personnel of hospital
Room, entrance guard's room security of cell etc..In different scenes, public security alarm unit is identical.Recognition of face positioning training is single
Member is primarily as recognition of face processing center or the data processing centre of camera.Face snap locating and tracking unit
The camera being mostly installed in each scene.
In the embodiment of the present invention, the face snap locating and tracking unit includes face acquisition module and face tracking positioning
Module, the face acquisition module are connect with face tracking locating module, the face tracking locating module control face acquisition
Module rotation tracks face.
In the embodiment of the present invention, the recognition of face positioning training unit includes data memory module, face recognition module
With Face detection training module, the data memory module is connect with face recognition module, the face recognition module and face
Training module connection is positioned, the data memory module provides identified human face data information for storing public security alarm unit,
The face recognition module is used to the human face data of face acquisition module acquisition back compare with identified human face data,
When comparison is by same or similar data, issues early warning or alarm signal is transmitted to public security alarm unit and/or security early warning
Unit, the Face detection training module are used for training pattern location model, provide location model and give face tracking locating module
Carry out tracking and positioning.
In the embodiment of the present invention, the detailed process of the Face detection training module training location model are as follows:
The twin network model of pyramid is constructed, the training twin network model of pyramid tests the twin network model of pyramid,
Training and the twin network model of pyramid tested are transferred to the realization of face tracking locating module to track face.
The training twin network model of pyramid:
(1) original video sequence is obtained from the video database in ILSVRC, and image procossing is carried out to the video sequence,
To obtain target training set and search training set.Wherein the size of target training set image is 127x127x3, searches for training set figure
The size of picture is 255x255x3, and the target of above-mentioned training set is all in the center of image；
(2) after completing training set image procossing, pairs of training set picture is input to the corresponding subnet of twin network, with
Target signature layer and search characteristics layer are obtained, the characteristic layer of above-mentioned different levels and different scale is extracted, to construct the present invention
Pyramid network, meanwhile, be based on candidate frame size formula and location formula, in every layer of feature of above-mentioned pyramid network
The different size of candidate frame of different location is constructed in layer；
(3) each layer of characteristic layer of above-mentioned pyramid network is input to classification and orientation parallel network of the invention, to obtain
The output of this parallel network is obtained as a result, the result of the output and label true value are carried out similarity mode, to obtain positive sample
Sheet and negative sample.
(4) error between above-mentioned matching result and label true value is calculated using target loss function, and by the error
Input layer successively is propagated back to, meanwhile, based on the weight in small lot stochastic gradient descent optimization algorithm adjustment network and partially
It sets, to obtain optimal error amount, to complete primary network model training；
(5) it repeats the above steps, until the error amount of target loss function converges on minimum value.
In the embodiment of the present invention, the twin network model of pyramid is by twin network, feature pyramid network and classification
Parallel network composition is positioned, twin network is made of the subnet that two VGG are constituted, and the subnet that two VGG are constituted is shared identical
Parameter, feature of the subnet that two VGG are constituted for respectively to target image and search image extract, and twin network is completed
After the feature extraction of target image and search image, the target signature layer and search characteristics layer of different scale are obtained respectively, never
6 layers of feature are extracted in the characteristic layer of same level and different scale, for constructing pyramid network.
Twin network is formed by two using the subnet that VGG is constituted, they are known respectively as targeted subnet and search subnet,
It carries out feature extraction to target image and search image respectively, and shares identical weight and biasing.Wherein VGG is as twin
The basic network of network is made of eleventh floor convolutional layer.Above-mentioned eleventh floor convolutional layer is respectively: first layer convolutional layer conv1
It is made of the convolution conv1_1 and conv1_2 that 2 sizes are 224x224x64；Second layer convolutional layer conv2 is by 2 sizes
The convolution conv2_1 and conv2_2 of 112x112x128 is constituted；Third layer convolutional layer conv3 is 56x56x256's by 3 sizes
Convolution conv3_1, conv3_2 and conv3_3 are constituted；The convolution that 4th layer of convolutional layer conv4 is 28x28x512 by 3 sizes
Conv4_1, conv4_2 and conv4_3 are constituted；The convolution that layer 5 convolutional layer conv5 is 14x14x512 by 3 sizes
Conv5_1, conv5_2 and conv5_3 are constituted；Layer 6 convolutional layer is made of the convolution conv6 that size is 3x3x1024；7th
Layer convolutional layer is made of the convolution conv7 that size is 1x1x1024；, the 8th layer of convolutional layer conv8 is 1x1x256 by 2 sizes
Convolution conv8_1 and 3x3x512 convolution conv8_2 constitute；9th layer of convolutional layer conv9 is 1x1x128's by 2 sizes
The convolution conv9_2 of convolution conv9_1 and 3x3x256 are constituted；Tenth layer of convolutional layer conv10 is 1x1x128's by 2 sizes
The convolution conv10_2 of convolution conv10_1 and 3x3x256 are constituted；Eleventh floor convolutional layer conv11 is by 2 sizes
The convolution conv11_2 of the convolution conv11_1 and 3x3x256 of 1x1x128 are constituted.
After the completion of feature pyramid network struction, in conjunction with classification and orientation parallel network, for the real-time positioning to target
And tracking, classification and orientation parallel network return subnet by candidate frame subnet, classifier subnet and positioning and form, candidate frame subnet,
Classifier subnet and positioning return subnet and generate candidate frame, confidence level and coordinate shift amount, and classifier subnet and positioning respectively
Subnet is returned to execute parallel.
In the embodiment of the present invention, the subnet that described two VGG are constituted is targeted subnet and search subnet, respectively to target figure
Picture and search image carry out feature extraction, and share identical weight and biasing, and targeted subnet and search subnet are by 11
Layer convolutional layer is constituted, and eleventh floor convolutional layer is respectively: first layer is made of 2 convolution units, and the second layer is by 2 convolution unit structures
At, third layer is made of 3 convolution units, and the 4th layer is made of 3 convolution units, and layer 5 is made of 3 convolution units, the
Six layers are made of 1 convolution unit, and layer 7 is made of 1 convolution unit, and the 8th layer is made of 2 convolution units, and the 9th layer
It is made of 2 convolution units, the tenth layer is made of 2 convolution units, and eleventh floor is made of 2 convolution units；
The feature pyramid network is made of the characteristic layer that targeted subnet and search subnet obtain, and total number of plies is 6 layers,
Be respectively: first layer is made of the eleventh floor characteristic layer in targeted subnet；The second layer is by the tenth layer of feature in search subnet
Layer is constituted；Third layer is made of the layer 7 characteristic layer in targeted subnet；4th layer by the layer 6 characteristic layer in search subnet
It constitutes；Layer 5 is made of the 4th layer of characteristic layer in targeted subnet；Layer 6 is by the third layer characteristic layer structure in search subnet
At,
The classification and orientation parallel network returns subnet by candidate frame subnet, classifier subnet and positioning and forms, candidate frame
Subnet is made of candidate frame, and for predicting the ability of target, classifier subnet is made of normalization exponential function classifier, is used
In distinguishing target and non-targeted ability, positioning returns subnet and is made of 3x3 convolution kernel, for positioning target, wherein candidate frame
By every tomographic image Image Segmentation Methods Based on Features at n x n grid, n is positive integer, and each grid generates the candidate frame of 6 fixed sizes, is waited
It selects frame to return subnet by classifier subnet and positioning and generates confidence level and coordinate shift amount respectively.
In the embodiment of the present invention, the detailed process of the training twin network model of pyramid are as follows:
Original video sequence is obtained from video database, and image preprocessing is carried out to video sequence, obtains target training
Collection and search training set, and the target of training set is all in the center of image；
After completing training set image procossing, pairs of target training set and search training set picture are input to twin network
Corresponding subnet obtains target signature layer and search characteristics layer, extracts the characteristic layer of different levels and different scale, constructs gold
Word tower network, meanwhile, it is based on candidate frame size formula and location formula, is constructed in every layer of characteristic layer of pyramid network
The different size of candidate frame of different location out；
Each layer of characteristic layer of pyramid network is input to classification and orientation parallel network, obtains the output knot of parallel network
The result of output and label true value are carried out similarity mode, obtain positive sample and negative sample by fruit；
The error between matching result and label true value is calculated using target loss function, and the error is successively reversed
Input layer is propagated to, meanwhile, based on the weight and biasing in small lot stochastic gradient descent optimization algorithm adjustment network, obtain most
Excellent error amount completes primary network model training；
It repeats the above steps, until the error amount of target loss function converges on minimum value.
In the embodiment of the present invention, the public security alarm unit includes public security database module and public security automatic alarm module,
The public security database module is connect with public security automatic alarm module, and the public security database module is for storing quilt inside public security
Personnel's human face data of mark and related labeled concerns information, the public security automatic alarm module for reporting automatically
It is alert, notify local police station's responding.
A kind of application of security cloud system, above system are applied in school's security system, hospital's security system, community security
System or industry park security system.Hospital field mainly use object be Security Personnel, Security Personnel according to comparing result, and
When be on the scene to medical trouble, illegally accompany and attend to and take corresponding measure；School field can pick people's identity validation for child, prevent small
Child is counterfeited relative's false claiming；Community can be used for monitoring and the early warning such as pyramid schemes personnel, confirmed thief, unstable fix the number of workers；It is specific to answer
With: it can be used for tracking the whereabouts etc. for studying and judging specific people.
Public security organ provides the portrait data of all kinds of runaway convicts, and hospital can then provide mental disease violence portrait data (society
Area, study, gas station, automobile 4S are according to the crowd of itself concern dimension), a multi-level various dimensions portrait is constructed jointly
Big data.When runaway convict removes hospital admission, that hospital pays close attention to this people as public security organ.When multiple level marketing, personnel go to hospital,
Perhaps hospital and multiple level marketing personnel are not concerned with, and are then paid close attention in terms of community.System is that the wide general and fine big data of society mentions
For entrance and a set of supervision and examination mechanism, guarantees that each level of society is concerned about his crowd of concern, customized for different crowd different
Business game.
The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention, for this field skill
For art personnel, it is clear that invention is not limited to the details of the above exemplary embodiments, and without departing substantially from spirit of the invention or
In the case where essential characteristic, the present invention can be realized in other specific forms.Therefore, in all respects, should all incite somebody to action
Embodiment regards exemplary as, and is non-limiting, the scope of the present invention by appended claims rather than on state
Bright restriction, it is intended that including all changes that fall within the meaning and scope of the equivalent elements of the claims in the present invention
It is interior.Any reference signs in the claims should not be construed as limiting the involved claims.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the principle of the present invention, it can also make several improvements and retouch, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (9)
1. a kind of security cloud system, which is characterized in that including public security alarm unit, security prewarning unit, recognition of face positioning instruction
Practice unit and face snap locating and tracking unit, the public security alarm unit and security prewarning unit, which are positioned with recognition of face, instructs
Practice unit connection, the recognition of face positioning training unit is connected with face snap locating and tracking unit, and the public security alarm is single
Member positions training unit to recognition of face for providing the human face data being identified in each scene, while it is fixed to receive recognition of face
Position training unit alarm signal automatic alarm, display alarm information, the face snap locating and tracking unit is for acquiring face
Data, and it is transmitted to recognition of face positioning training unit, it is carried out pair according to the tracking data that recognition of face positions training unit outflow
Corresponding face is tracked, and the recognition of face positioning training unit is used to receive the acquisition of face snap locating and tracking unit
Human face data is compared with the identified human face data that public security alarm unit provides, when contrasting by identical human face data,
It issues warning signal and informs security prewarning unit and/or public security alarm unit, security prewarning unit is fixed for receiving recognition of face
The early warning of position training unit or warning message are alarmed and notify Security Personnel.
2. a kind of security cloud system according to claim 1, it is characterised in that: the face snap locating and tracking unit packet
Face acquisition module and face tracking locating module are included, the face acquisition module is connect with face tracking locating module, described
The control face acquisition module rotation of face tracking locating module tracks face.
3. a kind of security cloud system according to claim 2, it is characterised in that: the recognition of face positions training unit packet
Include data memory module, face recognition module and Face detection training module, the data memory module and face recognition module
Connection, the face recognition module are connect with Face detection training module, and the data memory module is for storing public security alarm
Unit, which provides, is identified human face data information, and the face recognition module is used for the face number the acquisition of face acquisition module back
According to compared with identified human face data, when comparison is by same or similar data, issues early warning or alarm signal passes
To public security alarm unit and/or security prewarning unit, the Face detection training module is used for training pattern location model, provides
Location model carries out tracking and positioning to face tracking locating module.
4. a kind of security cloud system according to claim 3, it is characterised in that: the Face detection training module training is fixed
The detailed process of bit model are as follows:
The twin network model of pyramid is constructed, the training twin network model of pyramid is tested the twin network model of pyramid, will be instructed
White silk and the twin network model of pyramid tested are transferred to the realization of face tracking locating module and track to face.
5. a kind of security cloud system according to claim 4, it is characterised in that: the twin network model of pyramid is by twin
Raw network, feature pyramid network and classification and orientation parallel network composition, twin network are made of the subnet that two VGG are constituted,
The subnet that two VGG are constituted shares identical parameter, and the subnet that two VGG are constituted is for respectively to target image and search image
Feature extract, after twin network completes target image and searches for the feature extraction of image, obtain different scale respectively
Target signature layer and search characteristics layer, from 6 layers of feature are extracted in the characteristic layer of different levels and different scale, for constructing gold
Word tower network；
After the completion of feature pyramid network struction, in conjunction with classification and orientation parallel network, for target it is real-time positioning and with
Track, classification and orientation parallel network return subnet by candidate frame subnet, classifier subnet and positioning and form, candidate frame subnet, classification
Device subnet and positioning return subnet and generate candidate frame, confidence level and coordinate shift amount respectively, and classifier subnet and positioning return
Subnet executes parallel.
6. a kind of security cloud system according to claim 1, it is characterised in that: the subnet that described two VGG are constituted is mesh
Subnet and search subnet are marked, feature extraction is carried out to target image and search image respectively, and share identical weight and biasing,
Targeted subnet and search subnet are made of eleventh floor convolutional layer, and eleventh floor convolutional layer is respectively: first layer is by 2 convolution lists
Member is constituted, and the second layer is made of 2 convolution units, and third layer is made of 3 convolution units, and the 4th layer by 3 convolution unit structures
At, layer 5 is made of 3 convolution units, and layer 6 is made of 1 convolution unit, and layer 7 is made of 1 convolution unit, the
Eight layers are made of 2 convolution units, and the 9th layer is made of 2 convolution units, and the tenth layer is made of 2 convolution units, and the 11st
Layer is made of 2 convolution units；
The feature pyramid network is made of the characteristic layer that targeted subnet and search subnet obtain, and total number of plies is 6 layers, respectively
Be: first layer is made of the eleventh floor characteristic layer in targeted subnet；The second layer is by the tenth layer of characteristic layer structure in search subnet
At；Third layer is made of the layer 7 characteristic layer in targeted subnet；4th layer is made of the layer 6 characteristic layer in search subnet；
Layer 5 is made of the 4th layer of characteristic layer in targeted subnet；Layer 6 is made of the third layer characteristic layer in search subnet,
The classification and orientation parallel network returns subnet by candidate frame subnet, classifier subnet and positioning and forms, candidate frame subnet
It is made of candidate frame, for predicting the ability of target, classifier subnet is made of normalization exponential function classifier, is used for area
Partial objectives for and non-targeted ability, positioning returns subnet and is made of 3x3 convolution kernel, for positioning target, wherein candidate frame will be every
For tomographic image Image Segmentation Methods Based on Features at n x n grid, n is positive integer, and each grid generates the candidate frame of 6 fixed sizes, candidate frame
Subnet, which is returned, by classifier subnet and positioning generates confidence level and coordinate shift amount respectively.
7. a kind of security cloud system according to claim 1, it is characterised in that: the tool of the training twin network model of pyramid
Body process are as follows:
From video database obtain original video sequence, and to video sequence carry out image preprocessing, obtain target training set and
Training set is searched for, and the target of training set is all in the center of image；
After completing training set image procossing, by pairs of target training set and searches for training set picture to be input to twin network corresponding
Subnet, obtain target signature layer and search characteristics layer, extract the characteristic layer of different levels and different scale, construct pyramid
Network, meanwhile, it is based on candidate frame size formula and location formula, is constructed not in every layer of characteristic layer of pyramid network
With the different size of candidate frame in position；
Each layer of characteristic layer of pyramid network is input to classification and orientation parallel network, obtain the output of parallel network as a result,
The result of output and label true value are subjected to similarity mode, obtain positive sample and negative sample；
The error between matching result and label true value is calculated using target loss function, and by the layer-by-layer backpropagation of the error
To input layer, meanwhile, based on the weight and biasing in small lot stochastic gradient descent optimization algorithm adjustment network, obtain optimal
Error amount completes primary network model training；
It repeats the above steps, until the error amount of target loss function converges on minimum value.
8. a kind of security cloud system according to claim 1, it is characterised in that: the public security alarm unit includes public security number
According to library module and public security automatic alarm module, the public security database module is connect with public security automatic alarm module, the public security
Database module is used to store personnel's human face data and related labeled concerns information identified inside public security, institute
Public security automatic alarm module is stated for automatic alarm, notifies local police station's responding.
9. a kind of application of security cloud system according to claim 1, it is characterised in that: the claim 1-8 is any
One system is applied in school's security system, hospital's security system, community security system or industry park security system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910469910.7A CN110197158B (en) | 2019-05-31 | 2019-05-31 | Security cloud system and application thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910469910.7A CN110197158B (en) | 2019-05-31 | 2019-05-31 | Security cloud system and application thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110197158A true CN110197158A (en) | 2019-09-03 |
| CN110197158B CN110197158B (en) | 2023-04-18 |
Family
ID=67753645
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910469910.7A Active CN110197158B (en) | 2019-05-31 | 2019-05-31 | Security cloud system and application thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110197158B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110781778A (en) * | 2019-10-11 | 2020-02-11 | 珠海格力电器股份有限公司 | Access control method and device, storage medium and home system |
| CN111161891A (en) * | 2019-12-31 | 2020-05-15 | 重庆亚德科技股份有限公司 | Traditional Chinese medicine information management platform |
| CN113158933A (en) * | 2021-04-28 | 2021-07-23 | 广州瀚信通信科技股份有限公司 | Method, system, device and storage medium for identifying lost personnel |
| CN113762092A (en) * | 2021-08-17 | 2021-12-07 | 诺亚机器人科技(上海)有限公司 | Hospital user medical alarm detection method, system, robot and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101901340A (en) * | 2010-08-04 | 2010-12-01 | 惠州市华阳多媒体电子有限公司 | Suspect tracking method and system |
| CN108052925A (en) * | 2017-12-28 | 2018-05-18 | 江西高创保安服务技术有限公司 | A kind of cell personnel archives intelligent management |
| CN108200405A (en) * | 2018-02-05 | 2018-06-22 | 成都伦索科技有限公司 | A kind of video monitoring system based on recognition of face |
| CN109685066A (en) * | 2018-12-24 | 2019-04-26 | 中国矿业大学(北京) | A kind of mine object detection and recognition method based on depth convolutional neural networks |
| CN109711320A (en) * | 2018-12-24 | 2019-05-03 | 兴唐通信科技有限公司 | A kind of operator on duty's unlawful practice detection method and system |
| US20190147602A1 (en) * | 2017-11-13 | 2019-05-16 | Qualcomm Technologies, Inc. | Hybrid and self-aware long-term object tracking |
-
2019
- 2019-05-31 CN CN201910469910.7A patent/CN110197158B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101901340A (en) * | 2010-08-04 | 2010-12-01 | 惠州市华阳多媒体电子有限公司 | Suspect tracking method and system |
| US20190147602A1 (en) * | 2017-11-13 | 2019-05-16 | Qualcomm Technologies, Inc. | Hybrid and self-aware long-term object tracking |
| CN108052925A (en) * | 2017-12-28 | 2018-05-18 | 江西高创保安服务技术有限公司 | A kind of cell personnel archives intelligent management |
| CN108200405A (en) * | 2018-02-05 | 2018-06-22 | 成都伦索科技有限公司 | A kind of video monitoring system based on recognition of face |
| CN109685066A (en) * | 2018-12-24 | 2019-04-26 | 中国矿业大学(北京) | A kind of mine object detection and recognition method based on depth convolutional neural networks |
| CN109711320A (en) * | 2018-12-24 | 2019-05-03 | 兴唐通信科技有限公司 | A kind of operator on duty's unlawful practice detection method and system |
Non-Patent Citations (2)
| Title |
|---|
| FEI RONG等: ""A Detection algorithm based on Convolutional Neural Network"", 《PREPRINTS 2018》 * |
| SANGYUN LEE等: ""Multiple Object Tracking via Feature Pyramid Siamese Networks"", 《IEEE ACCESS》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110781778A (en) * | 2019-10-11 | 2020-02-11 | 珠海格力电器股份有限公司 | Access control method and device, storage medium and home system |
| CN110781778B (en) * | 2019-10-11 | 2021-04-20 | 珠海格力电器股份有限公司 | Access control method and device, storage medium and home system |
| CN111161891A (en) * | 2019-12-31 | 2020-05-15 | 重庆亚德科技股份有限公司 | Traditional Chinese medicine information management platform |
| CN111161891B (en) * | 2019-12-31 | 2023-06-30 | 重庆亚德科技股份有限公司 | Traditional Chinese medicine information management platform |
| CN113158933A (en) * | 2021-04-28 | 2021-07-23 | 广州瀚信通信科技股份有限公司 | Method, system, device and storage medium for identifying lost personnel |
| CN113762092A (en) * | 2021-08-17 | 2021-12-07 | 诺亚机器人科技(上海)有限公司 | Hospital user medical alarm detection method, system, robot and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110197158B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110197158A (en) | A kind of security cloud system and its application | |
| CN108062809B (en) | A kind of house access control system for realizing personnel's big data collection analysis | |
| US8648718B2 (en) | Event detection system using electronic tracking devices and video devices | |
| US7180420B2 (en) | System and method using triangulation with RF/LF and infrared devices for tracking objects | |
| WO2018116488A1 (en) | Analysis server, monitoring system, monitoring method, and program | |
| CN109711320A (en) | A kind of operator on duty's unlawful practice detection method and system | |
| CN104933819B (en) | Alarm and alarm method based on recognition of face and landmark identification | |
| US20060158308A1 (en) | System and method for automatically generating, collecting, monitoring and acting on information in an institutional setting | |
| CN111933297A (en) | Near field communication-based personnel joint sealing condition investigation method | |
| WO2022121498A1 (en) | Identity recognition method, model training method, apparatuses, and device and storage medium | |
| US20210364356A1 (en) | System and method for using artificial intelligence to enable elevated temperature detection of persons using commodity-based thermal cameras | |
| Haziq et al. | Smart IoT-based security system for residence | |
| McClain | The horizons of technological control: automated surveillance in the New York subway | |
| Gaushik et al. | Architecture design of ai and iot based system to identify covid-19 protocol violators in public places | |
| Ronoh et al. | Human-wildlife conflict early warning system using the internet of things and short message service | |
| Tran et al. | performance analysis of automatic integrated long-range RFID and webcam system | |
| KU et al. | Subduing Crime and Threat in Real-Time by Detecting Weapons Using Yolov8 | |
| Prasad et al. | Video surveillance-based security system using OpenCV and Arduino uno. | |
| CN111787287A (en) | Intelligent law enforcement and case handling center management system | |
| Krishnaveni et al. | IoT and AI based Forest Fire Prediction and Animal Monitoring System | |
| Shamnath et al. | Human Suspicious Activity Detection Using Ensemble Machine Learning Techniques | |
| KR102554835B1 (en) | System and Method for Contact Tracing of an Infectious Disease Patient | |
| CN116956980B (en) | Valuables monitoring management system based on RFID | |
| Greene et al. | Can we trust computer with body-cam vidio? Police departments are being led to believe AI will help, but they should be wary | |
| Masilamani et al. | FACE MASK DETECTION USING DEEP LEARNING |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 530007, Rooms 302 and 305, 3rd Floor, Block A, Shennengda Technology Incubation Park, R&D Building, No. 2, East Section of Gaoxin Avenue, High tech Zone, Nanning City, Guangxi Zhuang Autonomous Region Patentee after: Guangxi Aige Software Technology Co.,Ltd. Address before: 530007, Rooms 302 and 305, 3rd Floor, Block A, Shennengda Technology Incubation Park, R&D Building, No. 2, East Section of Gaoxin Avenue, High tech Zone, Nanning City, Guangxi Zhuang Autonomous Region Patentee before: GUANGXI NANNING BORUITONG SOFTWARE TECHNOLOGY Co.,Ltd. |