CN110427815A - Realize the method for processing video frequency and device of the effective contents interception of gate inhibition - Google Patents
Realize the method for processing video frequency and device of the effective contents interception of gate inhibition Download PDFInfo
- Publication number
- CN110427815A CN110427815A CN201910551347.8A CN201910551347A CN110427815A CN 110427815 A CN110427815 A CN 110427815A CN 201910551347 A CN201910551347 A CN 201910551347A CN 110427815 A CN110427815 A CN 110427815A
- Authority
- CN
- China
- Prior art keywords
- video pictures
- pixel
- target
- human body
- video
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/254—Analysis of motion involving subtraction of images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- 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
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N5/00—Details of television systems
- H04N5/76—Television signal recording
- H04N5/91—Television signal processing therefor
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
- H04N7/188—Capturing isolated or intermittent images triggered by the occurrence of a predetermined event, e.g. an object reaching a predetermined position
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20224—Image subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the method for processing video frequency for realizing the effective contents interception of gate inhibition, the following steps are included: carrying out moving object detection to the video pictures of acquisition first, to identify the video pictures for including motion target area, then whether judgement belongs to human body target comprising moving target corresponding in the video pictures of motion target area, whether human body target region included in video pictures of the detection comprising human body target includes effective facial image later, and finally the video pictures comprising effective facial image are saved and uploaded.This method is only saved and is uploaded to the video image for including effective facial image, without all collected monitor video images are all saved and are uploaded, the requirement to data space is reduced in terms of data storage, data carrying cost is reduced, while also reducing transmitted data on network amount and transmission cost in terms of uploaded videos picture data.
Description
Technical field
The present invention relates to technical field of image processing, in particular to a kind of video processing for realizing the effective contents interception of gate inhibition
Method and a kind of video process apparatus for realizing the effective contents interception of gate inhibition.
Background technique
More and more communities, building replace traditional access control system of swiping the card using video access control system at present.Video
Access control system shoots the video pictures of front certain space range with camera, therefrom identifies character face region, and from people
Object plane portion extracted region face characteristic, by the registration face characteristic progress of the face characteristic of extraction and registration in advance in the database
It compares, judges whether it has gate inhibition's right-of-way by way of identifying piece identity, gate inhibition is opened if having and is let pass, it is no
Then refuse to let pass.For not having the personage of gate inhibition's right-of-way, then the target room number that can be provided according to the personage, by video
Picture or character face's area transmissions give the target room, carry out manual verification.
Other than the basic function of gate inhibition's access control, video gate inhibition can also save above-mentioned video pictures, or will
The data of video pictures upload to background server by network and are saved, and enter depositing for community or building as personage
Grade, the purposes such as realization admission record, security are recorded, traced afterwards.
But for video gate inhibition, most of video pictures of shooting do not contain effective character face region.
Such as: the picture shot when unmanned current is unmanned video pictures, when someone is current due to task distance farther out, personage front
The video pictures for causing face not recognized effectively towards factors such as not positive (such as backwards to cameras), these video pictures are all
Not comprising effective character face region, it is not belonging to effective video picture, therefore these video pictures are for the note of access control system
Record the necessity for not having to save for shelves.And these video pictures without containing effective character face region are saved,
Memory capacity can be wasted, improves data carrying cost, while also will increase transmitted data on network amount, improve data transfer cost.
Summary of the invention
(1) goal of the invention
It is complete with lower data carrying cost and transmission cost for the ease of the monitoring system of community and building based on this
The identification of the facial image acquired in pairs of video monitoring regional and retention record are guaranteeing that retaining record can accurately reflect
Under the premise of face information, monitoring operation cost is reduced, the invention discloses following technical schemes.
(2) technical solution
As the first aspect of the present invention, the invention discloses realize the effective contents interception of gate inhibition method for processing video frequency,
Include:
Moving object detection is carried out to the video pictures of acquisition, to identify the video picture for including motion target area
Face；
Whether judgement belongs to human body target comprising moving target corresponding in the video pictures of the motion target area；
Whether human body target region included in video pictures of the detection comprising the human body target includes effective face
Image；
Video pictures comprising effective facial image are saved and/or uploaded.
In a kind of possible embodiment, the video pictures of described pair of acquisition carry out moving object detection and include:
Foreground area, and the prospect that will be obtained are identified from the video pictures of the acquisition using background subtraction method
Region is as the motion target area；
Wherein, the background model that the background subtraction method uses is mixed Gauss model or pixel grey scale mean value model.
In a kind of possible embodiment, using mixed Gauss model as background model to the video pictures of acquisition into
The identification of row foreground area includes:
Each Gauss model progress sequence that each pixel in video pictures is sorted from high to low with priority
Match, judgement and the matched Gauss model of the pixel；
Parameter update is carried out to the matched Gauss model of the pixel；
By in updated Gauss model highest priority and weight sum it is multiple high greater than background weight threshold
This model is as background；
Multiple background Gauss model carry out sequence matchings that each pixel and priority are sorted from high to low determine
The pixel for belonging to prospect out, obtains foreground area.
In a kind of possible embodiment, in judgement Gauss model matched with the pixel, in any Gauss
Model is and in the unmatched situation of the pixel: selecting the smallest Gauss model of weight as matched with the pixel
Gauss model.
In a kind of possible embodiment, drawn using pixel grey scale mean value model as video of the background model to acquisition
Face carry out foreground area identification include:
The mean value of respective pixel in the training image of gray level image is converted into as background pixel value, obtains background mould
Type；
The obtained background model is updated using the video image of present frame, obtains new background model；
The grey scale difference for calculating the video pictures to be detected and the new background model that are converted to gray level image, according to institute
It states grey scale difference and obtains foreground pixel probability distribution；
Foreground area is obtained according to the foreground pixel probability distribution.
It is described to obtain foreground area according to the foreground pixel probability distribution and include: in a kind of possible embodiment
The foreground pixel is divided into multiple grids according to the foreground pixel probability distribution, is counted in each grid
Pixel prospect probability cumulative and；
Whether the accounting for being accumulated in corresponding grid inner area as unit of grid according to each grid judges each grid
Belong to prospect grid, and then obtains the foreground area being made of the prospect grid.
In a kind of possible embodiment, the judgement is comprising corresponding in the video pictures of the motion target area
Whether moving target belongs to human body target
By the motion target area to coordinate axial projection, picture of the motion target area under each pixel rower is obtained
Prime number；
According to the feature of each pixel rower pixel number, phase corresponding at least three target sites of human body is determined
Answer the pixel rower of quantity；
Judge that each pixel column corresponding with human body target position is marked on the distance ratio of distance from each other in reference axis
Whether in corresponding human body distance than in range, if determining the movement if preset human body distance is than in range
Moving target in target area belongs to human body target.
In a kind of possible embodiment, the detection is comprising included in the video pictures in the human body target region
Human body target region whether include that effective facial image includes:
Pass through the picture area of the search window covering in the classifier judgement video pictures comprising motion target area
Whether domain belongs to human face region；
Video pictures are realized by described search window mobile in the video pictures comprising motion target area
Traversal；
Determine the positions and dimensions for the human face for including in the human face region；
Positional relationship based on the human face determines whether the human face region belongs to effective facial image.
In a kind of possible embodiment, the described pair of video pictures comprising effective facial image are saved
And/or it uploads and includes:
The picture area for including at least face part is extracted from the video pictures comprising effective facial image
It is saved and/or is uploaded.
In a kind of possible embodiment, people included in the video pictures comprising the human body target is being detected
In the case that body target area includes effective facial image, the described pair of video pictures comprising effective facial image into
Before row is saved and/or is uploaded:
The effective facial image for including by present frame video pictures is compared with former frame video pictures, and in former frame
When in video pictures including the facial image for all human body targets that present frame video pictures include, cancel to present frame video
The preservation and upload of picture.
As a second aspect of the invention, the invention also discloses realize that the video of the effective contents interception of gate inhibition is handled to fill
It sets, comprising:
Moving object detection module, the video pictures for acquiring to video capture device carry out moving object detection, with
Identify include motion target area video pictures；
Human body target judgment module, for judging that the moving object detection module identifies comprising the moving target
Whether corresponding moving target belongs to human body target in the video pictures in region；
Face information detection module includes the human body target for detect the human body target judgment module judgement
Whether human body target region included in video pictures includes effective facial image；
Video pictures retain module, for the face information detection module detect comprising effective face figure
The video pictures of picture are saved and/or are uploaded.
In a kind of possible embodiment, the moving object detection module uses background subtraction method from the acquisition
Foreground area is identified in video pictures, and using the obtained foreground area as the motion target area；
Wherein, the background model that the background subtraction method uses is mixed Gauss model or pixel grey scale mean value model.
In a kind of possible embodiment, the moving object detection module includes first object detection sub-module, is used
In the identification that as background model the video pictures of acquisition are carried out with foreground area using mixed Gauss model；
The first object detection sub-module includes:
Model Matching unit is high for each of each pixel and the priority in video pictures sorting from high to low
This model carry out sequence matching, judgement and the matched Gauss model of the pixel；
Parameter updating unit, for the Model Matching units match go out with the matched Gauss model of the pixel
Carry out parameter update；
Foreground selection unit, for highest priority in the Gauss model after being updated the parameter updating unit
And the sum of weight greater than background weight threshold multiple Gauss models as background；
First prospect acquiring unit, multiple backgrounds for each pixel and priority to sort from high to low
Selecting unit selects background Gauss model carry out sequence matching, determines the pixel for belonging to prospect, obtains foreground area.
In a kind of possible embodiment, in the Model Matching unit judges and the matched Gaussian mode of the pixel
When type, in any Gauss model and in the unmatched situation of the pixel: the Model Matching Unit selection weight is minimum
Gauss model as with the matched Gauss model of the pixel.
In a kind of possible embodiment, the moving object detection module includes the second target detection submodule, is used
In the identification that as background model the video pictures of acquisition are carried out with foreground area using pixel grey scale mean value model；
The second target detection submodule includes:
Background acquiring unit, for being converted into the mean value of respective pixel in the training image of gray level image as background picture
Element value, obtains background model；
Context update unit, the background model that the background acquiring unit is obtained for the video image using present frame
It is updated, obtains new background model；
Probability calculation unit, for calculating the video pictures to be detected and the context update unit that are converted to gray level image
The grey scale difference of updated background model obtains foreground pixel probability distribution according to the grey scale difference；
Second prospect acquiring unit, before the foreground pixel probability distribution for calculating according to the probability calculation unit obtains
Scene area.
In a kind of possible embodiment, the second prospect acquiring unit includes:
Cumulative statistics subelement, for the foreground pixel to be divided into multiple nets according to the foreground pixel probability distribution
Lattice, count the cumulative of pixel prospect probability in each grid and；
Accounting judgment sub-unit, for the corresponding grid inner area of being accumulated in as unit of grid according to each grid
Accounting judges whether each grid belongs to prospect grid, and then obtains the foreground area being made of the prospect grid.
In a kind of possible embodiment, the human body target judgment module includes:
Row pixel number statistic unit, for the motion target area to coordinate axial projection, to be obtained the moving target
Pixel number of the region under each pixel rower；
Target site matching unit, each pixel rower pixel number for being counted according to the row pixel number statistic unit
Feature, determine the pixel rower of respective numbers corresponding at least three target sites of human body；
Distance is than judging unit, for judging the target site of each and human body that the target site matching unit is determined
Corresponding pixel column be marked in reference axis distance from each other the ratio between two-by-two whether in corresponding human body distance than range
It is interior, if determining that the moving target in the motion target area belongs to human body mesh if preset human body distance is than in range
Mark.
In a kind of possible embodiment, the face information detection module includes:
Human face region search unit, for by the classifier judgement video pictures comprising motion target area
Whether the picture area of search window covering belongs to human face region；
Picture search spread unit, for by being searched described in the movement in the video pictures comprising motion target area
The traversal of rope window realization video pictures；
Organ characteristic's acquiring unit, for determining the people for including in human face region that the human face region search unit determines
The positions and dimensions of face；
It closes the position of effective face judging unit, the human face for being determined based on organ characteristic's acquiring unit
System determines whether the human face region belongs to effective facial image.
In a kind of possible embodiment, the video pictures retain module from described comprising effective facial image
Video pictures in extract include at least face part picture area saved and/or uploaded.
In a kind of possible embodiment, described device further include:
Picture retains judgment module, for detecting the view comprising the human body target in the face information detection module
In the case that human body target region included in frequency picture includes effective facial image, module is retained in the video pictures
Before video pictures comprising effective facial image are saved and/or are uploaded: including by present frame video pictures
Effective facial image is compared with former frame video pictures, and includes present frame video pictures in former frame video pictures
When the facial image for all human body targets for including, cancels the video pictures and retain preservation of the module to present frame video pictures
And upload.
(3) beneficial effect
The method for processing video frequency and device disclosed by the invention for realizing the effective contents interception of gate inhibition, can be to access control system reality
When the monitor video image that acquires simplified, to need the monitor video image acquired to access control system to carry out that shelves is stayed to save
Using as admission record, security record and in the future tracing of material when, only to include effective facial image video image carry out
It saves and uploads, without all collected monitor video images are all saved and uploaded, in terms of data storage
The requirement to data space is reduced, data carrying cost is reduced, while also being dropped in terms of uploaded videos picture data
In addition low transmitted data on network amount and transmission cost also in a disguised form increase inspection video pictures are carried out with personage's retrieval in the future when
Rope efficiency.
Detailed description of the invention
It is exemplary below with reference to the embodiment of attached drawing description, it is intended to for the explanation and illustration present invention, and cannot manage
Solution is the limitation to protection scope of the present invention.
Fig. 1 is the embodiment flow diagram of method for processing video frequency disclosed by the invention.
Fig. 2 is the schematic diagram of a certain frame video pictures of access control system acquisition.
Fig. 3 is the example structure block diagram of video process apparatus disclosed by the invention.
Specific embodiment
To keep the purposes, technical schemes and advantages of the invention implemented clearer, below in conjunction in the embodiment of the present invention
Attached drawing, technical solution in the embodiment of the present invention is further described in more detail.
Implement below with reference to the method for processing video frequency for realizing the effective contents interception of gate inhibition disclosed in Fig. 1 the present invention is described in detail
Example.As shown in Figure 1, method for processing video frequency disclosed in the present embodiment mainly comprises the steps that
Step 100, moving object detection is carried out to the video pictures of acquisition, to identify including motion target area
Video pictures.The view that access control system passes through certain area coverage of the video capture devices such as the camera acquisition comprising gate inhibition inlet
Frequency image obtains each frame video pictures comprising gate inhibition's entrance area, then the moving object detection mould of video process apparatus
Block carries out moving object detection to each frame video pictures that access control system acquires.Moving target refers to actually occurring mobile behavior
Target.
Motion target area refers to that the picture area that moving target is indicated in video pictures, the picture area usually only account for
A part of whole frame video pictures, therefore motion target area is contained in video pictures.Such as Fig. 2 show access control system and adopts
A wherein frame video pictures schematic diagram for collection, for the access control system of office building inlet, the personage P1 in building is just in face
It comes up to access control system, it is mobile towards observation position that symbol " ⊙ " represents target；Personage P2 in building is backwards to access control system
It leaves, symbol " ⊕ " represents the direction that target faces away from observation position；Reception counseling platform R in building is fixed；Vapour outside building
Automobile C on lane is being moved to the left.In four targets in Fig. 2, P1, P2 and C are actual moving target, profile region
Motion target area should belong in domain, and the reception fixed structures such as counseling platform R and ground, load bearing wall are actual non-athletic target,
Background area should belong in its contour area.
Moving object detection module can regard each frame by the methods of background subtraction method, frame differential method, optical flow method
Frequency picture carry out moving object detection, with identify and obtain include motion target area video pictures, hereinafter referred to as transport
Moving-target video pictures.Video pictures shown in Fig. 2 can be identified as including motion target area by moving object detection module
Video pictures.In above-mentioned three kinds of moving target detecting methods, background subtraction method and frame differential method are suitble to fairly simple back
Scene area, but more sensitive to the variation of light, and access control system majority is arranged in indoor has artificially lighting environment
In, therefore background subtraction method and frame differential method are relatively specific for access control system applied by the present embodiment, specific embodiment party
Formula, which is shown in, to be hereinafter described.
Step 200, whether judgement belongs to human body mesh comprising moving target corresponding in the video pictures of motion target area
Mark.After obtaining above-mentioned Moving Targets Based on Video Streams picture, the human body target judgment module of video process apparatus moves mesh to these
The motion target area for including in mark video pictures is judged, if the moving target that motion target area indicates is human body mesh
Mark, the then moving target for being included are the video pictures of human body target, hereinafter referred to as human body target video pictures.
Such as human body target judgment module sentences four targets that Moving Targets Based on Video Streams picture shown in Fig. 2 includes
Disconnected, the region of discovery target P1 and P2 are actual human body target region, therefore video pictures shown in Fig. 2 are human body target
Video pictures.
Step 300, whether human body target region included in video pictures of the detection comprising human body target includes effective
Facial image.After obtaining above-mentioned human body target video pictures, the face information detection module of video process apparatus is to these
The human body target region for including in human body target video pictures is detected, to obtain therefrom being identified effective face figure
The video pictures of picture, hereinafter referred to as face information video pictures.Effective facial image refers to determine therefrom that out personage's body
The facial image of part information allows face information video pictures to enter script holder by enter community or building as personage
It records, the data of security record, subsequent tracing of material.
Such as face information detection module two human body target P1 including to human body target video pictures shown in Fig. 2 and
The human body target region of P2 carries out detection identification, since target P1 is moved towards access control system, the image-region of target P1
In include be P1 the positive face of personage, and target P2 is mobile towards the direction far from access control system, apart from access control system camera
Farther out, and positive face relative to camera angle not just, positive face is backwards to camera, therefore in the image-region of target P2
It absolutely not include the positive face of personage of P2.Therefore only the image-region of target P1 includes effective facial image, the figure of target P2
As region fails comprising effective facial image.But it since video pictures shown in Fig. 2 include effective facial image, is consequently belonging to
Face information video pictures.If not including any effective facial image in video pictures, it is not belonging to face information video pictures.
Step 400, the video pictures comprising effective facial image are saved and/or is uploaded.
After obtaining above-mentioned face information video pictures, the video pictures of video process apparatus retain module and believe face
It ceases video pictures and carries out local preservation, can also upload in the background server of monitoring system and be saved and shown, to make
It is recorded for the admission record for the corresponding personage of effective facial image for including in picture, security, subsequent tracing of material.
Method for processing video frequency provided in this embodiment can carry out essence to the monitor video image that access control system acquires in real time
Letter, with need the monitor video image that access control system is acquired stay shelves save using as admission record, security record and
In the future when tracing of material, only the video image for including effective facial image is saved and uploaded, without being adopted all
The monitor video image collected is all saved and is uploaded, and is reduced in terms of data storage and is wanted to data space
It asks, reduces data carrying cost, while also reducing transmitted data on network amount and transmission in terms of uploaded videos picture data
In addition cost also in a disguised form increases recall precision video pictures are carried out with personage's retrieval in the future when.
In one embodiment, carrying out moving object detection to the video pictures of acquisition in step 100 includes:
Step 110, foreground area, and the prospect that will be obtained are identified from the video pictures of acquisition using background subtraction method
Region is as motion target area.
Background subtraction method is the pixel value for carrying out approximate background image using the context parameter model pre-established, by present frame
Image and background model carry out differential comparison, to realize detection to motion target area.To current frame image and background
When image is compared, the biggish pixel region of the difference compared is considered as foreground area, distinguishes lesser pixel region
It is considered as background area.Region of the moving target in video pictures is referred to as prospect, and other do not actually occur movement
The region of target (non-athletic target) in video pictures is then background.Such as in Fig. 2, before target P1, P2 and C are actual
Scene area, target R is actual background area, therefore the region of target P1, P2 and C are motion target area.In step 110
The background model that the background subtraction method used is selected is mixed Gauss model or pixel grey scale mean value model.
In one embodiment, prospect is carried out as video pictures of the background model to acquisition using mixed Gauss model
The identification in region the following steps are included:
Step A1 carries out each pixel in video pictures with each Gauss model that priority sorts from high to low
Sequence matches, judgement and the matched Gauss model of pixel.
Mixed Gauss model is denoted as η (It,μi,t), i=1,2 ..., K, ItFor t moment (namely t frame number)
Pixel, μi,tIt is i-th of Gauss model in the mean value of t moment, K is the quantity of Gauss model, is typically set at three to five
It is a.In each Gauss model, ωi,tFor weight of i-th of Gauss model on t moment current pixel, andFor the priority of i-th of Gauss model.
Judge that pixel is matched with Gauss model according to formula (1) in step A1:
|It-μi,t-1|≤Di*δi,t-1Formula (1)；
Wherein, μi,t-1Mean value for i-th of Gaussian function in t-1 frame, δi,t-1For i-th of Gaussian function t-1 frame mark
Poor, the D of standardiFor constant.
Model is trained it is understood that mixed Gauss model needs to first pass through multiframe continuous video pictures in advance
It obtains.
When judgement is with pixel matched Gauss model in step A1, if any Gauss model with pixel not
Match, then select the smallest Gauss model of weight as with the matched Gauss model of pixel.
Step A2 carries out parameter update to the matched Gauss model of pixel.The parameter of update includes Gauss model
Weight ωi,t, varianceAnd mean μi,t, and be updated respectively according to formula (2) to (4):
ωi,t=(1- α) * ωi,t-1+ α formula (2)；
μi,t=(1- ρ) * μi,t-1+ρ*ItFormula (3)；
Wherein, α is customized learning rate, and the size of 0≤α≤1, α determine the speed that model is updated, with model
Renewal speed is directly proportional.ρ is parameter learning rate, ρ ≈ α/ωi,t。
Pervious mean value and variance are not kept then with the matched Gauss model of pixel, and weight is then according to ωi,t=(1-
α)*ωi,t-1Decaying.
If failing to match pixel Gauss model in step A1, therefore select the selection the smallest Gauss model of weight
When, then when carrying out parameter update to the smallest Gauss model of the selection weight, its mean value is updated to It, standard deviation is updated to
δ0, right value update ωK,t=(1- α) * ωK,t-1+α。
Step A3, by updated Gauss model highest priority and weight sum greater than background weight threshold
Nb Gauss model is as background.After mixed Gauss model parameter updates, order standard of the foundation priority as Gauss model,
Each Gauss model is subjected to sequence from big to small, the higher Gauss model of priority is more forward in the sequence, also more may be
Background.It can use foundation of the threshold value T of the part sum of background weight as screening model, if T is less than the power of preceding Nb model
Value and, then using preceding Nb model as background distributions.
Step A4, the Nb Gauss model progress sequence as background that each pixel and priority are sorted from high to low
Matching, determines the pixel for belonging to prospect, obtains foreground area.Assuming that current pixel point is It, according to the priority of model
Sequence matches the Nb Gauss model filtered out in the pixel and step A3 one by one, if meeting formula (5), determines
ItFor foreground point, I is otherwise determinedtFor background dot, the foreground area in video pictures is thus obtained, that is, target P1 in Fig. 2,
The region of P2 and C.
|It-μi,t|>D2*δi,t, i=1,2 ..., Nb formula (5)；
Wherein, D2For customized constant.
In one embodiment, it is carried out using pixel grey scale mean value model as video pictures of the background model to acquisition
The identification of foreground area includes:
Step B1 is converted into the mean value of respective pixel in the training image of gray level image as background pixel value, obtains
Background model.Training image refers to the video sequence for training background model, first one before selection current time
Divide training image, then convert gray level image for these training images, using the mean value of respective pixel in training image as back
Scape pixel value obtains background model B, that is, obtains background image.
Background model B is indicated by formula (6):
Wherein, ItFor the gray level image of t moment (t frame number), T be for train the video sequence of background model it is total when
It is long, i.e., it is above-mentioned " a part of training image before current time ", therefore T also corresponds to the frame number of video pictures.T value is bigger,
The video pictures for being equivalent to selection are more, then the background model obtained is more accurate, but operation time is longer.
Step B2 is updated obtained background model using the video image of present frame, obtains new background model.
Since background model is not forever constant, it is therefore desirable to be updated to background model.Such as worked as using formula (7) to calculate
The pixel value of preceding background model, and then the background model updated；
Wherein, ptFor the pixel value of t moment image, ut-1For the pixel value of corresponding current background model, α is customized
Habit rate, the size of 0≤α≤1, α determine current frame image to the adjustment influence degree of background model, and α is bigger, current frame image
Bigger on the adjustment of background model influence, background model adapts to faster with environmental change.When α=1, be equivalent to using present frame as
New background replaces original background model.
Step B3 calculates the gray scale difference of the video pictures to be detected and updated background model that are converted to gray level image
Point, foreground pixel probability distribution is obtained according to grey scale difference.
If there is moving target in the video sequence, the difference of comparison current video picture and background frame can be passed through
Moving target is not detected.Gray level image is converted by video pictures to be detected first, is then calculated by formula (8) to be detected
The grey scale difference dI of frame and background modelt(x):
Wherein, ItIt (x) is video pictures to be detected, B (x) is background image, and thr is constant.Grey scale difference indicates current
The degree that the gray value of frame changes relative to background gray levels shows that the variation is due to moving mesh if changing and being greater than thr
Caused by mark.
Judge whether pixel belongs to foreground pixel especially by formula (9), and then obtain foreground area:
Wherein, Pt(x) approximate representation for belonging to the probability of prospect for current pixel, if Pt(x) very little, then can tentatively by
The pixel is judged as background pixel, if Pt(x) very big, then illustrate that the pixel is foreground pixel.
Step B4 obtains foreground area according to foreground pixel probability distribution.By all pixels for being judged as foreground pixel
Point combines, and foreground area can be obtained.Specifically, in order to eliminate ambient noise, so that foreground area will not be divided into
More than the multiple regions of actual foreground region quantity, and be to try to be similar to actual foreground region, step B4 the following steps are included:
Foreground pixel is divided into multiple grids according to foreground pixel probability distribution, counts pixel in each grid by step B41
Prospect probability cumulative and.Cumulative and A is specifically calculated such as formula (10):
A=∑ P (x) x ∈ L formula (10)；
Wherein, L is a certain regional area, that is, grid, and x is the pixel in the L of region.
Step B42, the accounting for being accumulated in corresponding grid inner area as unit of grid according to each grid judge each grid
Whether belong to prospect grid, and then obtains the foreground area being made of prospect grid.Due to 0≤P (x)≤1, the maximum of A
Value is exactly the area S of region L.Then judge that region belongs to foreground area or the decision condition of background area can as unit of region
To use formula (11):
Wherein, β is invariant, and 0≤β≤1, β is bigger, and region is judged as belonging to the prospect that foreground area to be included
The quantitative requirement of pixel is higher, is also just more difficult to be judged as foreground area.
In one embodiment, judge to move accordingly in the video pictures comprising motion target area in step 200
Target whether belong to human body target the following steps are included:
Step 210, motion target area is obtained into picture of the motion target area under each pixel rower to coordinate axial projection
Prime number.By taking Fig. 2 as an example, the motion target area of moving target P1, P2 and C are converted into binary image, so that moving target
Pixel point value in region is 1, and the value in other regions is 0.Then the row of moving target profile is obtained to coordinate axial projection respectively
Mark-pixel number statistical graph, the X-axis in chart is pixel rower, and referring to motion target area altogether includes multirow pixel, Y
Axis is the pixel quantity that value is 1.Since the shape (namely profile) of motion target area is different, every row pixel rower is corresponding
Pixel quantity it is also different, the corresponding pixel quantity of a circular every row pixel rower is in the song fallen after rising along X-axis
Line, the corresponding pixel quantity of every row pixel rower of the horizontal positioned triangle in a bottom edge is along the linear ascendant trend of X-axis.
Step 220, the feature according to each pixel rower pixel number is determined corresponding at least three target sites of human body
Respective numbers pixel rower.
The feature of the pixel rower pixel number of motion target area of different shapes is different, corresponding in rower-pixel number system
What is presented on meter chart is exactly stripe shape and lifting trend difference.Since human body target has certain features of shape, because
The stripe shape that this general human body is presented in correspondence graph is usual are as follows: is to fall after rising and drop from the head in X-axis to neck
To the reduced levels of pixel quantity, and X-axis head portion is relatively close to the distance between neck, and increases since neck, intermediate
There may be some liftings, then arriving foot is then nearby decline.
It follows that wherein stablizing the most constant is exactly head, neck and foot, thus select these three positions as
Target site.And neck to the part between foot can due to the stature of human body and different, a but elder generation from head to neck
Dropped after rising and neck after rising, the decline before foot be then general human body pixel quantity linear feature.
Therefore can using head, neck and foot as the target site of human body, and since head and foot are both ends,
Therefore its pixel number is zero, and neck is due to most thin, and pixel number is one minimum in all troughs in lines, and distance
Head is closer.
It follows that the head of target site and foot are located in X-axis, the neck of target site is higher than one in X-axis
A trough determines head in three charts, neck, foot in X respectively after the chart for obtaining moving target P1, P2 and C
Position on axis.
Step 230, judgement respectively pixel column corresponding with human body target position be marked in reference axis distance from each other away from
From than whether with a distance from corresponding human body than determining to move if if preset human body distance is than in range in range
Moving target in target area belongs to human body target.
Due to general human body head to neck and neck to the distance between foot than being the presence of rule, in X
On axis, the distance between head to neck HN and neck to the distance between foot distance NF ratio in a certain range, such as
0.1≤HN/NF≤0.15, human body distance is exactly [0.1,0.15] than range, if the HN/NF of moving target is in human body
Otherwise distance is determined as non-human target than in range, being then determined as human body target.Target P1 and P2 can be judged as human body mesh
Mark, target C can be judged as non-human target.
Can also similarly choose head to the distance between neck HN and head between the distance between foot HF away from
From the judgment basis than whether belonging to human body target as moving target, applicable human body distance has phase than range at this time
It should change.
In one embodiment, human body mesh included in video pictures of the detection comprising human body target in step 300
Mark region whether include effective facial image the following steps are included:
Step 310, the picture by classifier judgement comprising the search window covering in the video pictures of motion target area
Whether face region belongs to human face region.Search window is the window for being less than video pictures size, for delimiting one piece of region
As the region of identification face, and classifier can classify to sample, therefore can use classifier and draw to search window
Fixed region carries out image classification and realizes the identification of face to obtain facial image class.
Step 320, pass through the mobile search window realization video pictures in the video pictures comprising motion target area
Traversal.Search window can be moved in video pictures with the distance for being less than search window length/width, be drawn with traversing entire video
Face.Such as Fig. 2, face information detection module can detect the face area for including in the region of target P1 by traversal
Domain.Target P2 will not be recognized as face due to not including face.
Step 330, the positions and dimensions for the human face for including in human face region are determined.Such as the angle of face is different,
The organ shown is also different, and positive face can show that all eyebrows, eye, nose, mouth, but possibly can not show ear, and side
Face is only able to display out eyebrow, eye, the ear of wherein side, and shows nose, mouth.
Step 340, the positional relationship based on human face determines whether human face region belongs to effective facial image.Distance is taken the photograph
As the close positive face of head, organ clarity height, angle just, belong to effective facial image, such as the region of the target P1 in Fig. 2 includes
Effective facial image.The side face remote apart from camera, organ clarity is not high, angle just, is not belonging to effective facial image.
In one embodiment, the picture area of search window covering is judged in step 310 by Adaboost classifier
Whether domain belongs to human face region.The mode of training Adaboost classifier is as follows:
Given training sample set S, wherein X and Y corresponds respectively to face positive example sample and non-face negative example sample, and T is instruction
Experienced maximum cycle.Training sample is N number of altogether, and the corresponding weight of each sample is identical, initialization in N number of training sample
Sample weights are 1/N, as the initial probability distribution of training sample.Under this sample distribution by N number of training sample into
Row training obtains first Weak Classifier.
For the sample of classification error, its corresponding weight is increased, and for correct sample of classifying, its weight is reduced,
The sample of misclassification so just is highlighted out, to obtain a new sample distribution.Under new sample distribution, again to sample
Originally it is trained, obtains second Weak Classifier.
And so on, it is recycled by T times, obtains T Weak Classifier, this T Weak Classifier is pressed certain weighted superposition
Get up, the strong classifier finally wanted.
After strong classifier trains, it can use finally obtained strong classifier and correctly identify people in video pictures
Face area information.Such as strong classifier can identify the human face region P1f of target P1 from Fig. 2, therefore Fig. 2 belongs to and includes
The video pictures of facial image are imitated, it includes have an effective face P1f.
The video pictures comprising cognizable character face region (effective video picture) are obtained after identified, but due to same
One personage can generate a large amount of effective video picture during passage, be directed to for the record of the personage, these are effectively
Effective content that video pictures include is all the same.For example, for the continuous M frame video pictures acquired in a period of time,
J frame video pictures (regardless of whether being continuous frame) is wherein shared comprising effective face figure if identifying through this embodiment
Whole image contents of the j frame video pictures are then locally saved or are uploaded to monitoring system in access control system by picture
Background server in, using as staying shelves to record.
But due to not needing to save background area and the low region of interest-degree sometimes, in one embodiment,
The video pictures comprising effective facial image are saved and/or uploaded in step 400, specifically may is that from comprising effective
The picture area extracted in the video pictures of facial image including at least face part is saved and/or is uploaded.Such as
For video pictures Fig. 2, the region of the target P1 comprising human face region P1f, that is, video pictures only can be saved or uploaded
The human region of middle P1, or even only save or upload human face region P1f.It can be further reduced the data volume of storage and upper in this way
The data volume of biography.
In addition, effective face that above-mentioned j frame video pictures include may be the face of same personage, therefore in step
When video pictures being saved and uploaded in 400, the content for saving and uploading has a large amount of duplicate contents, but it is not necessary to
All video pictures comprising effective people information are all preserved, memory capacity otherwise can be wasted, data is improved and deposits
Cost is stored up, while also will increase transmitted data on network amount, improve data transfer cost.Therefore in one embodiment, in step
Detect that human body target region included in the video pictures comprising human body target includes effective facial image in rapid 300
In the case of, before the video pictures comprising effective facial image are saved and/or are uploaded in step 400:
The effective facial image for first including by present frame video pictures is compared with former frame video pictures, and previous
When in frame video pictures including the facial image for all human body targets that present frame video pictures include, cancels and present frame is regarded
The preservation and upload of frequency picture.
It include effective facial image in Fig. 2, therefore meet and be saved by taking present frame video pictures shown in Fig. 2 as an example
The condition of biography, but by being found after it is compared with former frame video pictures, former frame video pictures and present frame video
Personage corresponding to effective facial image that picture includes is all the same, is all target P1, therefore this two frames video pictures is used as
Script holder record, security record meaning and contribution be identical, therefore be saved and uploaded former frame video pictures or
In the case where having saved with the meaning of former frame video pictures and the identical more previous video pictures of contribution, then without pair
Present frame video pictures are saved and are uploaded, to be further reduced the data volume of storage and the data volume of upload.
Specifically, access control system can cache former frame video pictures, it is made whether to need to present frame video pictures
When preservation upload judges, use the former frame video pictures of caching as reference, then no matter whether former frame video pictures are protected
It deposits or uploads, can guarantee among one section of continuous video frame, if the face that every frame video pictures include is all the same, to admission
Record, the meaning of security record are also all the same, then only first frame video pictures are saved and upload, alternatively, passing through every detection
When one new video frame, the quality of effective facial image will be compared, quality of human face image is higher than to the view saved
The same meaning video frame of frequency frame replaces previously stored video frame.
It should be noted that the video pictures due to preservation can be recorded as admission record and security, it is only continuous
One section of video frame comprising identical effective face in, the preservation of wherein most video frame can be saved, if otherwise target
P1 occurred primary and after being collected, uploading in monitoring system before one month, and today after one month goes out again
When now and collected, even if its video frame occurred is spaced one month, belong to discontinuous video frame, as having saved
Information and no longer save, this just cannot achieve admission record and security record function.
Implement below with reference to the video process apparatus for realizing the effective contents interception of gate inhibition disclosed in Fig. 3 the present invention is described in detail
Example.The present embodiment is for implementing method for processing video frequency above-mentioned.
As shown in figure 3, video process apparatus disclosed in the present embodiment, comprising:
Moving object detection module, the video pictures for acquiring to video capture device carry out moving object detection, with
Identify include motion target area video pictures；
Human body target judgment module, the view comprising motion target area identified for judging moving object detection module
Whether corresponding moving target belongs to human body target in frequency picture；
Face information detection module, for detecting the video pictures comprising human body target of human body target judgment module judgement
Included in human body target region whether include effective facial image；
Video pictures retain module, the video comprising effective facial image for detecting to face information detection module
Picture is saved and/or is uploaded.
In one embodiment, moving object detection module is identified from the video pictures of acquisition using background subtraction method
Foreground area out, and using obtained foreground area as motion target area；
Wherein, the background model that background subtraction method uses is mixed Gauss model or pixel grey scale mean value model.
In one embodiment, moving object detection module includes first object detection sub-module, for using mixing
Gauss model carries out the identification of foreground area as background model to the video pictures of acquisition；
First object detection sub-module includes:
Model Matching unit is high for each of each pixel and the priority in video pictures sorting from high to low
This model carry out sequence matching, judgement and the matched Gauss model of pixel；
Parameter updating unit carries out parameter with the matched Gauss model of pixel for what is gone out to Model Matching units match
It updates；
Foreground selection unit, in the Gauss model after being updated parameter updating unit highest priority and
The sum of weight greater than background weight threshold multiple Gauss models as background；
First prospect acquiring unit, multiple Foreground selection units for each pixel and priority to sort from high to low
Background Gauss model carry out sequence matching is selected, the pixel for belonging to prospect is determined, obtains foreground area.
In one embodiment, in Model Matching unit judges Gauss model matched with pixel, any high
This model is and in the unmatched situation of pixel: the smallest Gauss model of Model Matching Unit selection weight as with pixel
Matched Gauss model.
In one embodiment, moving object detection module includes the second target detection submodule, for using pixel
Gray average model carries out the identification of foreground area as background model to the video pictures of acquisition；
Second target detection submodule includes:
Background acquiring unit, for being converted into the mean value of respective pixel in the training image of gray level image as background picture
Element value, obtains background model；
Context update unit carries out the background model that background acquiring unit obtains for the video image using present frame
It updates, obtains new background model；
Probability calculation unit is updated for calculating the video pictures to be detected for being converted to gray level image and context update unit
The grey scale difference of background model afterwards obtains foreground pixel probability distribution according to grey scale difference；
Second prospect acquiring unit, the foreground pixel probability distribution for calculating according to probability calculation unit obtain foreground zone
Domain.
In one embodiment, the second prospect acquiring unit includes:
Cumulative statistics subelement is counted for foreground pixel to be divided into multiple grids according to foreground pixel probability distribution
In each grid pixel prospect probability cumulative and；
Accounting judgment sub-unit, for the accounting for being accumulated in corresponding grid inner area as unit of grid according to each grid
Judge whether each grid belongs to prospect grid, and then obtains the foreground area being made of prospect grid.
In one embodiment, human body target judgment module includes:
Row pixel number statistic unit, for motion target area to coordinate axial projection, to be obtained motion target area each
Pixel number under pixel rower；
Target site matching unit, the spy of each pixel rower pixel number for being counted according to row pixel number statistic unit
Sign, determines the pixel rower of respective numbers corresponding at least three target sites of human body；
For distance than judging unit, the target site of each and human body for judging that target site matching unit is determined is corresponding
Pixel column be marked in reference axis distance from each other the ratio between two-by-two whether in corresponding human body distance than in range, if
In preset human body distance than then determining that the moving target in motion target area belongs to human body target in range.
In one embodiment, face information detection module includes:
Human face region search unit, for including the search in the video pictures of motion target area by classifier judgement
Whether the picture area of window covering belongs to human face region；
Picture search spread unit, for passing through the mobile search window reality in the video pictures comprising motion target area
The traversal of existing video pictures；
Organ characteristic's acquiring unit, for determine human face region search unit determine human face region in include face device
The positions and dimensions of official；
Effective face judging unit, the positional relationship of the human face for being determined based on organ characteristic's acquiring unit are sentenced
Determine whether human face region belongs to effective facial image.
In one embodiment, video pictures retain module extracted from the video pictures comprising effective facial image to
The picture area comprising face part is saved and/or is uploaded less.
In one embodiment, device further include:
Picture retains judgment module, in the video pictures that face information detection module detects comprising human body target
In the case that the human body target region for being included includes effective facial image, module is retained to including effective people in video pictures
Before the video pictures of face image are saved and/or uploaded: the effective facial image for including by present frame video pictures is with before
One frame video pictures are compared, and include all human body mesh that present frame video pictures include in former frame video pictures
When target facial image, cancels video pictures and retain preservation and upload of the module to present frame video pictures.
It should be understood that in the accompanying drawings, from beginning to end same or similar label indicate same or similar element or
Element with the same or similar functions.Described embodiments are some of the embodiments of the present invention, rather than whole implementation
Example, in the absence of conflict, the features in the embodiments and the embodiments of the present application can be combined with each other.Based in the present invention
Embodiment, every other embodiment obtained by those of ordinary skill in the art without making creative efforts,
It shall fall within the protection scope of the present invention.
Herein, " first ", " second " etc. are only used for mutual differentiation, rather than indicate their significance level and sequence
Deng.
The division of module, unit or assembly herein is only a kind of division of logic function, in actual implementation may be used
To there is other division modes, such as multiple modules and/or unit can be combined or are integrated in another system.As separation
The module of part description, unit, component are also possible to indiscrete may be physically separated.It is shown as a unit
Component can be physical unit, may not be physical unit, it can is located at a specific place, may be distributed over grid
In unit.Therefore some or all of units can be selected to realize the scheme of embodiment according to actual needs.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by those familiar with the art, all answers
It is included within the scope of the present invention.Therefore, protection scope of the present invention should be with the scope of protection of the claims
It is quasi-.
Claims (10)
1. a kind of method for processing video frequency for realizing the effective contents interception of gate inhibition characterized by comprising
Moving object detection is carried out to the video pictures of acquisition, to identify the video pictures for including motion target area；
Whether judgement belongs to human body target comprising moving target corresponding in the video pictures of the motion target area；
Whether human body target region included in video pictures of the detection comprising the human body target includes effective facial image；
Video pictures comprising effective facial image are saved and/or uploaded.
2. the method as described in claim 1, which is characterized in that the video pictures of described pair of acquisition carry out moving object detection packet
It includes:
Foreground area, and the foreground area that will be obtained are identified from the video pictures of the acquisition using background subtraction method
As the motion target area；
Wherein, the background model that the background subtraction method uses is mixed Gauss model or pixel grey scale mean value model.
3. method according to claim 2, which is characterized in that using mixed Gauss model as background model to the view of acquisition
Frequency picture carry out foreground area identification include:
Each Gauss model carry out sequence matching that each pixel in video pictures is sorted from high to low with priority, sentences
The disconnected and matched Gauss model of the pixel；
Parameter update is carried out to the matched Gauss model of the pixel；
By in updated Gauss model highest priority and weight sum greater than background weight threshold multiple Gaussian modes
Type is as background；
Multiple background Gauss model carry out sequence matchings that each pixel and priority are sorted from high to low, determine to belong to
In the pixel of prospect, foreground area is obtained.
4. the method as described in claim 1, which is characterized in that the judgement includes the video pictures of the motion target area
In corresponding moving target whether belong to human body target and include:
By the motion target area to coordinate axial projection, pixel of the motion target area under each pixel rower is obtained
Number；
According to the feature of each pixel rower pixel number, respective counts corresponding at least three target sites of human body are determined
The pixel rower of amount；
Judge each pixel column corresponding with human body target position be marked on distance from each other in reference axis distance compare whether
In corresponding human body distance than in range, if determining the moving target if preset human body distance is than in range
Moving target in region belongs to human body target.
5. the method as described in claim 1, which is characterized in that the described pair of video pictures comprising effective facial image into
Row is saved and/or is uploaded
The picture area progress for including at least face part is extracted from the video pictures comprising effective facial image
It saves and/or uploads.
6. a kind of video process apparatus for realizing the effective contents interception of gate inhibition characterized by comprising
Moving object detection module, the video pictures for acquiring to video capture device carry out moving object detection, with identification
It out include the video pictures of motion target area；
Human body target judgment module, for judging that the moving object detection module identifies comprising the motion target area
Video pictures in corresponding moving target whether belong to human body target；
Face information detection module, the video comprising the human body target determined for detecting the human body target judgment module
Whether human body target region included in picture includes effective facial image；
Video pictures retain module, for the face information detection module detect comprising effective facial image
Video pictures are saved and/or are uploaded.
7. device as claimed in claim 6, which is characterized in that the moving object detection module uses background subtraction method from institute
It states and identifies foreground area in the video pictures of acquisition, and using the obtained foreground area as the motion target area；
Wherein, the background model that the background subtraction method uses is mixed Gauss model or pixel grey scale mean value model.
8. device as claimed in claim 7, which is characterized in that the moving object detection module includes first object detection
Module, for as background model the video pictures of acquisition to be carried out with the identification of foreground area using mixed Gauss model；
The first object detection sub-module includes:
Model Matching unit, each Gaussian mode for each pixel in video pictures to sort from high to low with priority
Type carry out sequence matching, judgement and the matched Gauss model of the pixel；
Parameter updating unit is carried out for what is gone out to the Model Matching units match with the matched Gauss model of the pixel
Parameter updates；
Foreground selection unit, in the Gauss model after being updated the parameter updating unit highest priority and
The sum of weight greater than background weight threshold multiple Gauss models as background；
First prospect acquiring unit, multiple Foreground selections for each pixel and priority to sort from high to low
Unit selection goes out background Gauss model carry out sequence matching, determines the pixel for belonging to prospect, obtains foreground area.
9. device as claimed in claim 6, which is characterized in that the human body target judgment module includes:
Row pixel number statistic unit, for the motion target area to coordinate axial projection, to be obtained the motion target area
Pixel number under each pixel rower；
Target site matching unit, the spy of each pixel rower pixel number for being counted according to the row pixel number statistic unit
Sign, determines the pixel rower of respective numbers corresponding at least three target sites of human body；
For distance than judging unit, the target site of each and human body for judging that the target site matching unit is determined is corresponding
Pixel column be marked in reference axis distance from each other the ratio between two-by-two whether in corresponding human body distance than in range, if
In preset human body distance than then determining that the moving target in the motion target area belongs to human body target in range.
10. device as claimed in claim 6, which is characterized in that the video pictures, which retain module, to be had from described comprising described
The picture area that extraction includes at least face part in the video pictures of facial image is imitated to be saved and/or uploaded.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910551347.8A CN110427815B (en) | 2019-06-24 | 2019-06-24 | Video processing method and device for realizing interception of effective contents of entrance guard |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910551347.8A CN110427815B (en) | 2019-06-24 | 2019-06-24 | Video processing method and device for realizing interception of effective contents of entrance guard |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110427815A true CN110427815A (en) | 2019-11-08 |
| CN110427815B CN110427815B (en) | 2020-07-10 |
Family
ID=68409468
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910551347.8A Active CN110427815B (en) | 2019-06-24 | 2019-06-24 | Video processing method and device for realizing interception of effective contents of entrance guard |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110427815B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111583485A (en) * | 2020-04-16 | 2020-08-25 | 北京澎思科技有限公司 | Community access control system, access control method and device, access control unit and medium |
| CN111784896A (en) * | 2020-06-17 | 2020-10-16 | 深圳南亿科技股份有限公司 | Access control monitoring image storage method, system and storage medium |
| CN111881866A (en) * | 2020-08-03 | 2020-11-03 | 杭州云栖智慧视通科技有限公司 | Real-time face grabbing recommendation method and device and computer equipment |
| CN112637567A (en) * | 2020-12-24 | 2021-04-09 | 中标慧安信息技术股份有限公司 | Multi-node edge computing device-based cloud data uploading method and system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101032405A (en) * | 2007-03-21 | 2007-09-12 | 汤一平 | Safe driving auxiliary device based on omnidirectional computer vision |
| CN102368301A (en) * | 2011-09-07 | 2012-03-07 | 常州蓝城信息科技有限公司 | Moving human body detection and tracking system based on video |
| CN102542271A (en) * | 2010-12-23 | 2012-07-04 | 胡茂林 | Video-based technology for informing people visiting and protecting privacy at house gates or entrances and exits of public places such as office buildings |
| CN104318202A (en) * | 2014-09-12 | 2015-01-28 | 上海明穆电子科技有限公司 | Method and system for recognizing facial feature points through face photograph |
| JP2016176816A (en) * | 2015-03-20 | 2016-10-06 | キヤノン株式会社 | Image processor, image processing method, and program |
| CN106157329A (en) * | 2015-04-20 | 2016-11-23 | 中兴通讯股份有限公司 | A kind of adaptive target tracking method and device |
| CN106372576A (en) * | 2016-08-23 | 2017-02-01 | 南京邮电大学 | Deep learning-based intelligent indoor intrusion detection method and system |
-
2019
- 2019-06-24 CN CN201910551347.8A patent/CN110427815B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101032405A (en) * | 2007-03-21 | 2007-09-12 | 汤一平 | Safe driving auxiliary device based on omnidirectional computer vision |
| CN102542271A (en) * | 2010-12-23 | 2012-07-04 | 胡茂林 | Video-based technology for informing people visiting and protecting privacy at house gates or entrances and exits of public places such as office buildings |
| CN102368301A (en) * | 2011-09-07 | 2012-03-07 | 常州蓝城信息科技有限公司 | Moving human body detection and tracking system based on video |
| CN104318202A (en) * | 2014-09-12 | 2015-01-28 | 上海明穆电子科技有限公司 | Method and system for recognizing facial feature points through face photograph |
| JP2016176816A (en) * | 2015-03-20 | 2016-10-06 | キヤノン株式会社 | Image processor, image processing method, and program |
| CN106157329A (en) * | 2015-04-20 | 2016-11-23 | 中兴通讯股份有限公司 | A kind of adaptive target tracking method and device |
| CN106372576A (en) * | 2016-08-23 | 2017-02-01 | 南京邮电大学 | Deep learning-based intelligent indoor intrusion detection method and system |
Non-Patent Citations (1)
| Title |
|---|
| 余启明: ""基于背景减法和帧差法的运动目标检测算法研究"", 《中国优秀硕士学位论文全文数据库（电子期刊） 信息科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111583485A (en) * | 2020-04-16 | 2020-08-25 | 北京澎思科技有限公司 | Community access control system, access control method and device, access control unit and medium |
| CN111784896A (en) * | 2020-06-17 | 2020-10-16 | 深圳南亿科技股份有限公司 | Access control monitoring image storage method, system and storage medium |
| CN111784896B (en) * | 2020-06-17 | 2021-02-23 | 深圳南亿科技股份有限公司 | Access control monitoring image storage method, system and storage medium |
| CN111881866A (en) * | 2020-08-03 | 2020-11-03 | 杭州云栖智慧视通科技有限公司 | Real-time face grabbing recommendation method and device and computer equipment |
| CN111881866B (en) * | 2020-08-03 | 2024-01-19 | 杭州云栖智慧视通科技有限公司 | Real-time face grabbing recommendation method and device and computer equipment |
| CN112637567A (en) * | 2020-12-24 | 2021-04-09 | 中标慧安信息技术股份有限公司 | Multi-node edge computing device-based cloud data uploading method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110427815B (en) | 2020-07-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110427815A (en) | Realize the method for processing video frequency and device of the effective contents interception of gate inhibition | |
| CN106997629B (en) | Access control method, apparatus and system | |
| CN100568262C (en) | Human face recognition detection device based on the multi-video camera information fusion | |
| CN105069472B (en) | A kind of vehicle checking method adaptive based on convolutional neural networks | |
| CN109376637B (en) | People counting system based on video monitoring image processing | |
| CN104091176B (en) | Portrait comparison application technology in video | |
| CN100397410C (en) | Method and device for distinguishing face expression based on video frequency | |
| CN104303193B (en) | Target classification based on cluster | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| CN111932583A (en) | Space-time information integrated intelligent tracking method based on complex background | |
| CN107239762A (en) | Patronage statistical method in a kind of bus of view-based access control model | |
| CN103605971B (en) | Method and device for capturing face images | |
| CN108205661A (en) | A kind of ATM abnormal human face detection based on deep learning | |
| Björklund et al. | Automatic license plate recognition with convolutional neural networks trained on synthetic data | |
| CN109063619A (en) | A kind of traffic lights detection method and system based on adaptive background suppression filter and combinations of directions histogram of gradients | |
| CN113592911B (en) | Apparent enhanced depth target tracking method | |
| CN111402298A (en) | Grain depot video data compression method based on target detection and trajectory analysis | |
| CN107483894A (en) | Judge to realize the high ferro station video monitoring system of passenger transportation management based on scene | |
| CN109918971A (en) | Number detection method and device in monitor video | |
| CN111368660A (en) | Single-stage semi-supervised image human body target detection method | |
| CN106874825A (en) | The training method of Face datection, detection method and device | |
| CN101950448A (en) | Detection method and system for masquerade and peep behaviors before ATM (Automatic Teller Machine) | |
| Ghidoni et al. | Texture-based crowd detection and localisation | |
| CN107977610A (en) | A kind of face tracking methods based on massive video processing | |
| CN103065163A (en) | Rapid target detection and recognition system and method based on static picture |
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 |