CN107133607B - Demographics' method and system based on video monitoring - Google Patents
Demographics' method and system based on video monitoring Download PDFInfo
- Publication number
- CN107133607B CN107133607B CN201710395675.4A CN201710395675A CN107133607B CN 107133607 B CN107133607 B CN 107133607B CN 201710395675 A CN201710395675 A CN 201710395675A CN 107133607 B CN107133607 B CN 107133607B
- Authority
- CN
- China
- Prior art keywords
- model
- crowd
- local feature
- human body
- texture
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/758—Involving statistics of pixels or of feature values, e.g. histogram matching
-
- 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
- G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of Demographics' method and system based on video monitoring, method includes the following steps: S1: obtaining the real-time monitoring images of pre- monitoring area；S2: foreground segmentation is carried out to real-time monitoring images, is partitioned into crowd；S3: complete characteristics of human body's model is divided into five local feature models, and configures corresponding weight for each local feature model；S4: the feature of the crowd after segmentation is extracted, and matching detection is carried out with five local feature models, comprehensive matching degree is calculated according to the positional shift between weight and the five local feature models and complete characteristics of human body's model of extraction, if being more than specific threshold, successful match；S5: counting crowd's quantity according to the number of successful match, draw crowd's quantity change curve, when curve or curvilinear trend triggering anomalous event, is warned.Demographics' method and system based on video monitoring of the invention, are applicable to different occasions, accuracy is higher.
Description
Technical field
The present invention relates to field of intelligent monitoring more particularly to a kind of Demographics' method and system based on video monitoring.
Background technique
Transport hub, the people streams in public places congestion phenomenon such as large-scale activity scene and megastore are more and more frequent.Number
Statistics has become the importance of society's total safety and most optimum distribution of resources.The competition of business etc., with accurate
Data based on analyzed and managed, establish intelligentized passenger number statistical system as development trend.
There are mainly two types of current existing demographic methods: 1) being examined using body templates to the people in video image
It surveys；2) people in video image is detected using the number of people or head and shoulder template.Two methods have some limitations, only
Can be used for specific scene, portable poor, method 1) detection effect is poor in the higher scene of crowd density, method 2) in people
Accuracy rate is low in group's lower scene of density.
Summary of the invention
Demographics' method and system based on video monitoring that technical problem to be solved by the invention is to provide a kind of, can
Suitable for different occasions, accuracy is higher.
To solve the above problems, the present invention proposes a kind of Demographics' method based on video monitoring, comprising the following steps:
S1: the real-time monitoring images of pre- monitoring area are obtained；
S2: foreground segmentation is carried out to real-time monitoring images, is partitioned into crowd；
S3: complete characteristics of human body's model is divided into five local feature models, and configures phase for each local feature model
The weight answered；
S4: extracting the feature of the crowd after segmentation, and carries out matching detection with five local feature models,
Comprehensive is calculated according to the positional shift between weight and the five local feature models and complete characteristics of human body's model of extraction
With degree, if being more than specific threshold, successful match；
S5: crowd's quantity is counted according to the number of successful match, crowd's quantity change curve is drawn, when curve or curve become
When gesture triggers anomalous event, warned.
According to one embodiment of present invention, the step S3 the following steps are included:
S31: by complete characteristics of human body's model partition be five local feature models, respectively head, the upper part of the body a left side
Two up and down of right two and the lower part of the body, obtain model M1, model M2, model M3, model M4, model M5, complete characteristics of human body
Model M=M1+M2+M3+M4+M5；
S32: weight beta={ β of five local feature models of configuration1, β2, β3, β4, β5, so that β1+β2+β3+β4+β5=1；
S33: the HOG feature of characteristics of human body's model sample and non-human characteristic model sample is extracted, is obtained by SVM classifier
The template of five local feature models.
According to one embodiment of present invention, the step S4 the following steps are included:
S41: the HOG feature of the crowd after extracting segmentation；
S42: the feature of extraction is matched with the template of five local feature models, comprehensive matching degree S=
β1M1+β2M2+β3M3+β4M4+β5M5- b, wherein b indicates that the position between practical local feature model and ideal stance model is inclined
It moves, b=b2+b3+b4+b5, model M1Coordinate (the e of central point1, f1), model M2Coordinate (the e of central point2, f2), model M3Center
Coordinate (the e of point3, f3), model M4Coordinate (the e of central point4, f4), model M5Coordinate (the e of central point5, f5)；
S43: the comprehensive matching degree of human body each in obtained crowd is compared with specific threshold, if being more than specific
Threshold value, then successful match is primary, the successful number of statistical match.
According to one embodiment of present invention, the data of pre- monitoring area are acquired as training set, are determined using training set
Mix the weight of five local feature models of dynamic texture model and human body.
According to one embodiment of present invention, the step S5 the following steps are included:
S51: counting crowd's quantity according to the number of successful match, draws crowd's quantity change curve, and abnormal in triggering
Event alarm；
S52: recording crowd's number curve of triggered anomalous event, is placed into possible anomalous event set；
S53: it when curve or trend occur again, is given warning in advance according to warning level；If subsequent triggers anomalous event,
The danger classes of the curve element is increased in possible anomalous event set, conversely, danger classes declines；
S54: the height according to danger classes is determining or adjusts warning level.
According to one embodiment of present invention, in the step S2, to real-time monitoring images using mixing dynamic texture mould
Type carries out foreground segmentation to video image, comprising the following steps:
S21: by real-time monitoring images binaryzation, and a mixing dynamic texture model is established；
S22: the unknown parameter in the mixing dynamic texture model is solved using EM algorithm, so that it is determined that the mixing is dynamic
State texture model；
S23: according to the mixing dynamic texture model, calculating the maximum likelihood probability of monitoring image mixing dynamic texture,
Texture class label is obtained, is clustered according to texture class label by dynamic texture is mixed；
S24: the texture after scanning cluster, using the similar dynamic texture for the crowd that characterizes as prospect, remaining texture is as back
Scape is partitioned into the dynamic texture of the crowd of different directions walking.
According to one embodiment of present invention, the real-time monitoring images of the pre- monitoring area of acquisition are divided into training set and test
Collection determines the weight of mixing five local feature models of dynamic texture model and human body using training set, then by test set
Investment executes step S1-S5.
Demographics' system based on video monitoring that the present invention also provides a kind of, comprising:
Image acquisition unit obtains the real-time monitoring images of pre- monitoring area；
Crowd's cutting unit carries out foreground segmentation to real-time monitoring images, is partitioned into crowd；
Complete characteristics of human body's model is divided into five local feature models, and is each local feature by feature division unit
Model configures corresponding weight；
Characteristic matching unit extracts the feature of the crowd after segmentation, and with five local feature models into
Row matching detection, according to the positional shift between weight and the five local feature models and complete characteristics of human body's model of extraction
Comprehensive matching degree is calculated, if being more than specific threshold, successful match；
Alarm unit is counted, crowd's quantity is counted according to the number of successful match, crowd's quantity change curve is drawn, works as song
When line or curvilinear trend triggering anomalous event, warned.
After adopting the above technical scheme, the present invention has the advantages that compared with prior art
Each local feature model has corresponding weight, and the match condition of comprehensive all local feature models obtains finally
Matching result, by change local feature model weight to can be applied to different occasions, facilitate the transplanting of algorithm, really
The accuracy that guarantor group counts；Anomalous event feedback mechanism is established, enhances the stability of system, avoids false-alarm；
Foreground segmentation is carried out using mixing dynamic texture, the stream of people of different directions is distinguished, additional algorithm is not needed
Tracking crowd determines crowd direction, saves time cost, improves efficiency of algorithm；
Consider the positional shift between body local characteristic model and complete manikin, human body shape can be effectively prevent
Become.
Detailed description of the invention
Fig. 1 is the flow diagram of Demographics' method based on video monitoring of one embodiment of the invention；
Fig. 2 is that using to real-time monitoring images for one embodiment of the invention mixes dynamic texture model progress foreground segmentation
Flow diagram；
Fig. 3 is the flow diagram that the EM algorithm of one embodiment of the invention solves；
Fig. 4 a is the structural schematic diagram of the ideal stance model of one embodiment of the invention；
Fig. 4 b be one embodiment of the invention there are the structural schematic diagrams of the practical local feature model of positional shift；
Fig. 5 is the flow diagram of the statistics alert stage of one embodiment of the invention.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing to the present invention
Specific embodiment be described in detail.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention.But the present invention can be with
Much it is different from other way described herein to implement, those skilled in the art can be without prejudice to intension of the present invention the case where
Under do similar popularization, therefore the present invention is not limited to the specific embodiments disclosed below.
Referring to Fig. 1, in one embodiment, Demographics' method based on video monitoring the following steps are included:
S1: the real-time monitoring images of pre- monitoring area are obtained；
S2: foreground segmentation is carried out to real-time monitoring images, is partitioned into crowd；
S3: complete characteristics of human body's model is divided into five local feature models, and configures phase for each local feature model
The weight answered；
S4: extracting the feature of the crowd after segmentation, and carries out matching detection with five local feature models,
Comprehensive is calculated according to the positional shift between weight and the five local feature models and complete characteristics of human body's model of extraction
With degree, if being more than specific threshold, successful match；
S5: crowd's quantity is counted according to the number of successful match, crowd's quantity change curve is drawn, when curve or curve become
When gesture triggers anomalous event, warned.
Below to the further details of elaboration of Demographics' method based on video monitoring, but should not be as limit.
In step sl, the real-time monitoring images of pre- monitoring area are obtained, which is to need to carry out number system
In the occasion of meter as obtained by the shooting of the filming apparatus such as monitoring camera, monitoring camera passes real time monitoring picture back in real time, for holding
Row subsequent step.
In step s 2, foreground segmentation is carried out to the real-time monitoring images obtained in step S1, is partitioned into crowd.This time-division
The prospect cut out is crowd's information, can not also determine crowd's quantity.
Referring to Fig. 2, in step s 2, prospect point preferably is carried out using mixing dynamic texture model to real-time monitoring images
It cuts, comprising the following steps:
S21: by real-time monitoring images binaryzation, and a mixing dynamic texture model is established；
Wherein, ytIndicate the video image of binaryzation, i.e. observation sequence；xtIndicate the derivation ingredient of video image at any time,
That is hidden state sequence；AzFor state-transition matrix；CzFor observation matrix；vtFor the noise generated in state migration procedure, obey
Mean value is zero, covariance QzGaussian Profile；ωtFor observation noise, obeying mean value is zero, covariance Rz；Z is directed toward certain
Specific dynamic texture；
S22: institute is solved using EM algorithm (Expectation Maximization Algorithm, EM algorithm)
The unknown parameter in mixing dynamic texture model is stated, so that it is determined that the mixing dynamic texture model；
S23: according to the mixing dynamic texture model, calculating the maximum likelihood probability of monitoring image mixing dynamic texture,
Texture class label is obtained, is clustered according to texture class label by dynamic texture is mixed；
S24: the texture after scanning cluster, using the similar dynamic texture for the crowd that characterizes as prospect, remaining texture is as back
Scape is partitioned into the dynamic texture of the crowd of different directions walking.
More specifically, observation sequence ytLength be N, traditional representation isInitial hidden state variable x1Clothes
It is μ from mean valuez, covariance SzGaussian Profile；Model parameterθz={ Az, Qz, Cz, Rz, μz, Sz}；K is indicated
The type of texture, every kind of texture represent a kind of crowd in direction；α indicates the weight of texture, andEM algorithm is related to
Following formula:
One, the conditional probability distribution of hidden state sequence and observation sequence:
Conditional probability distribution p (the x of initial hidden state sequence1| z)=G (x1, μz, Sz)
(2)
Conditional probability distribution p (the x of hidden state sequencet|xt-1, z) and=G (xt, Azxt-1, Qz)
(3)
Conditional probability distribution p (the y of observation sequencet|xt, z) and=G (yt, Czxt, Rz)
(4)
T indicates the time, and t=1 indicates initial time, x1Indicate initial hidden state variable.Z is directed toward certain specific dynamic
State texture, xtIndicate the hidden state sequence of implicit t moment, ytIndicate the observation sequence of t moment.Indicate initial hidden on the right of equation
Conditional probability, the conditional probability Gaussian distributed of observation sequence of conditional probability, hidden state sequence containing status switch.Az
For state-transition matrix；CzFor observation matrix；vtFor the noise generated in state migration procedure, obeying mean value is zero, and covariance is
QzGaussian Profile；ωtFor observation noise, obeying mean value is zero, covariance Rz。
Two, expectation relevant to hidden state sequence:
The expectation of hidden state sequence
(i indicates that i-th of observation sequence, j indicate certain specific texture)；
Two above formula (6) (7) is relevant to the expectation of hidden state sequence, since it is desired that by hidden state sequence
The expectation of column acquires, and formula (10) below asks parameter needs to use.
Three, conditional probability relevant to texture:
Conditional probability relevant to texture is i.e.: according to known observation sequence y(i)Seek z(i)The probability of=j is how many, j table
Show jth kind texture, j=1,2 ..., and K }, K is the type of texture.At this point it is possible to by the dynamic texture in video image into
Row classification can extract prospect by corresponding label using the similar dynamic texture for the crowd that characterizes as prospect, that is, before
Scape segmentation.
Four, the veins clustering label of dynamic texture is mixed:
li=argmax [logp (y(i)|z(i)=j)+log αj] (9)
argmax[logp(y(i)|z(i)=j)+log αj] indicate to take logp (y(i)|z(i)=j)+log αjIn maximum value,
The occurrence of j is to mix texture classification belonging to dynamic texture at this time.
Five, the expression of dynamic texture model parameter is mixed:
Cause are as follows:
Formula (10) is to simplify expression, and the formula on the right of equation has passed through formula (5)~(8) and acquired；
So:
Formula (11) is the value that parameter is sought using formula (10), and * indicates estimated value.
Referring to Fig. 3, it is as follows to solve the step of mixing dynamic texture model parameter for EM algorithm:
A, observation sequence is inputtedInitialization model parameter { θj, αj, j=1,2 ..., K }；
B, calculation formula (5)~(8) expectation, i=1 ..., N }, j=1 ..., K }；
C, calculation formula (10) update model parameter using formula (11)
Return step B is continued to execute, until convergence.
The maximum likelihood probability of mixing dynamic texture is calculated using formula, obtains texture class label li, according to liIt will mixing
Dynamic texture cluster.Texture after scanning cluster, similar dynamic texture as prospect, remaining texture is partitioned into as background
The dynamic texture of the crowd of different directions walking.
In step s3, complete characteristics of human body's model is divided into five local feature models, and is each local feature mould
Type configures corresponding weight.The weight of different local feature models can be the same or different, and can pass through actual scene
Data are trained to obtain desired value.
Further, step S3 the following steps are included:
S31: by complete characteristics of human body's model partition be five local feature models, respectively head, the upper part of the body a left side
Two up and down of right two and the lower part of the body, obtain model M1, model M2, model M3, model M4, model M5, complete characteristics of human body
Model M=M1+M2+M3+M4+M5；
S32: weight beta={ β of five local feature models of configuration1, β2, β3, β4, β5, so that β1+β2+β3+β4+β5=1；
S33: the HOG feature of characteristics of human body's model sample and non-human characteristic model sample is extracted, is obtained by SVM classifier
The template of five local feature models.
Complete characteristics of human body's model M can be divided into five offices based on the body local characteristic model detection method of HOG
Portion's characteristic model, specifically includes: model M1, model M2, model M3, model M4, model M5, and M=M1+M2+M3+M4+M5, such as scheme
Shown in 4a.
In order to break the limitation of the prior art.Two methods as described in the background art, have some limitations,
It is only used for specific scene, it is portable poor.It can be applied to different fields by changing the weight of local feature model
It closes, in other words, the weight of local feature model is adjusted according to the demand of different occasions, such as: in high-density environments, due to
The reasons such as block, for crowd's head feature than more prominent, other features are relatively weak, assign the local feature model for representing head
Weight is big, and the weight for assigning other local feature models is small；Under case of low density case, in order to ensure crowd count accuracy,
The feature of detection whole man be it is best, give the imparting of each local feature model identical weight；Etc..Five local feature models
It is to be easier to find out model M from Fig. 4 a and 4b according to organization of human body model partition1Indicate the head of people, model M2And mould
Type M3Indicate the left and right two (being free of head) of the upper part of the body, model M4And model M5Indicate two up and down of the lower part of the body.
Wherein, in step S32, weight beta={ β of five local feature models1, β2, β3, β4, β5Initial value is respectively 0.2,
The error of the crowd's quantity and true statistical result that are gone out according to test set statistical estimate adjusts the value of weight, until i.e. error is small
In preset value.
In step s 4, the feature of (dynamic texture) crowd after foreground segmentation is extracted, and special with five parts
It levies model and carries out matching detection, according between weight and the five local feature models and complete characteristics of human body's model of extraction
Positional shift calculates comprehensive matching degree, if being more than specific threshold, successful match.
Referring to Fig. 4 a and 4b, because deformation occurs for human body in the process of movement, it will lead to and dropped with the matching degree of template
Low, so to subtract corresponding positional shift, when positional shift is more than specific threshold value, comprehensive matching degree will be less than certain threshold
Value, then match it is unsuccessful (such as happen that by the upper limb of the head of pedestrian A and pedestrian B, lower limb error detection be one
Individual, this is clearly that not pair, and the positional shift between the upper limb of pedestrian B, lower limb and the head of pedestrian A is not necessarily more than specific
Threshold value, the generation of mistake can be avoided well by subtracting positional shift this variable).
Further, step S4 the following steps are included:
S41: the HOG feature of the crowd after extracting segmentation；
S42: the feature of extraction is matched with the template of five local feature models, comprehensive matching degree S=
β1M1+β2M2+β3M3+β4M4+β5M5- b, wherein b indicates that the position between practical local feature model and ideal stance model is inclined
It moves, b=b2+b3+b4+b5, model M1Coordinate (the e of central point1, f1), model M2Coordinate (the e of central point2, f2), model M3Center
Coordinate (the e of point3, f3), model M4Coordinate (the e of central point4, f4), model M5Coordinate (the e of central point5, f5)；
S43: the comprehensive matching degree of human body each in obtained crowd is compared with specific threshold, if being more than specific
Threshold value, then successful match is primary, the successful number of statistical match.
In step s 5, crowd's quantity is counted according to the number of successful match, draws crowd's quantity change curve, works as curve
Or it when curvilinear trend triggering anomalous event, is warned.Since the crowd in scene is dynamic motion, thus crowd's quantity is not
Break and changing, can intuitively be embodied by change curve, and can be according to trend prediction.The standard of triggering anomalous event can be set,
Reach respective standard just to trigger.
Further, step S5 the following steps are included:
S51: counting crowd's quantity according to the number of successful match, draws crowd's quantity change curve, and abnormal in triggering
Event alarm；
S52: recording crowd's number curve of triggered anomalous event, is placed into possible anomalous event set；
S53: it when curve or trend occur again, is given warning in advance according to warning level；If subsequent triggers anomalous event,
The danger classes of the curve element is increased in possible anomalous event set, conversely, danger classes declines；
S54: the height according to danger classes is determining or adjusts warning level.
Anomalous event feedback mechanism is established, stability can be enhanced, avoid false-alarm, abnormal conditions can be carried out with early warning, and
And can taking place frequently degree and determine its danger classes according to anomalous event, provide guidance in time.
Preferably, the data of pre- monitoring area are acquired as training set, determine mixing dynamic texture model using training set
And the weight of five local feature models of human body.
Before formally coming into operation, needs to acquire data, the data of acquisition are divided into training set and test set, utilize training
Collection acquires the parameter of mixing dynamic texture model, and the weight of local feature model also acquires at this moment (generally to be taken under primary condition
The weight of five local feature models is 0.2, since the data manually acquired have true statistical result, according to system
The error of crowd's quantity of estimation and true statistical result adjusts the value of weight, until convergence: i.e. error is provided less than some
Value)；Test set can be used to debug.
The data of acquisition are randomly divided into training set and test set by the crowd's incremental data for manually acquiring certain public place,
Learn the parameter of mixing dynamic texture model using training setAnd weight { the β of body local characteristic model1,
β2, β3, β4, β5, data debugging is carried out using test set, it is ensured that is operated normally.Demographic method is transplanted to another public
When place, due to the change of environment, crowd density most probably changes, and manually acquires crowd's incremental data in the place, weight
Multiple above-mentioned training process and test process, can come into operation again.
Demographics' system based on video monitoring that the present invention also provides a kind of, comprising:
Image acquisition unit obtains the real-time monitoring images of pre- monitoring area；
Crowd's cutting unit carries out foreground segmentation to real-time monitoring images, is partitioned into crowd；
Complete characteristics of human body's model is divided into five local feature models, and is each local feature by feature division unit
Model configures corresponding weight；
Characteristic matching unit extracts the feature of the crowd after segmentation, and with five local feature models into
Row matching detection, according to the positional shift between weight and the five local feature models and complete characteristics of human body's model of extraction
Comprehensive matching degree is calculated, if being more than specific threshold, successful match；
Alarm unit is counted, crowd's quantity is counted according to the number of successful match, crowd's quantity change curve is drawn, works as song
When line or curvilinear trend triggering anomalous event, warned.
Particular content about Demographics' system the present invention is based on video monitoring may refer in previous embodiment
The description of Demographics' method part based on video monitoring, details are not described herein.
Although the present invention is disclosed as above with preferred embodiment, it is not for limiting claim, any this field
Technical staff without departing from the spirit and scope of the present invention, can make possible variation and modification, therefore of the invention
Protection scope should be subject to the range that the claims in the present invention are defined.
Claims (6)
1. a kind of Demographics' method based on video monitoring, which comprises the following steps:
S1: the real-time monitoring images of pre- monitoring area are obtained；
S2: foreground segmentation is carried out to real-time monitoring images, is partitioned into crowd；
S3: complete characteristics of human body's model is divided into five local feature models, and corresponding for the configuration of each local feature model
Weight；
The step S3 the following steps are included:
S31: by complete characteristics of human body's model partition be five local feature models, respectively head, the upper part of the body left and right two
Two up and down of portion and the lower part of the body, obtain model M1, model M2, model M3, model M4, model M5, complete characteristics of human body's model M
=M1+M2+M3+M4+M5；
S32: weight beta={ β of five local feature models of configuration1, β2, β3, β4, β5, so that β1+β2+β3+β4+β5=1；
S33: the HOG feature of characteristics of human body's model sample and non-human characteristic model sample is extracted, obtains five by SVM classifier
The template of local feature model；
S4: extracting the feature of the crowd after segmentation, and carries out matching detection with five local feature models, according to
Positional shift between weight and the five local feature models and complete characteristics of human body's model of extraction calculates comprehensive matching journey
Degree, if being more than specific threshold, successful match；
The step S4 the following steps are included:
S41: the HOG feature of the crowd after extracting segmentation；
S42: the feature of extraction is matched with the template of five local feature models, comprehensive matching degree S=β1M1+
β2M2+β3M3+β4M4+β5M5- b, wherein b indicates the positional shift between practical local feature model and ideal stance model, b=
b2+b3+b4+b5, model M1Coordinate (the e of central point1, f1), model M2Coordinate (the e of central point2, f2), model M3The seat of central point
Mark (e3, f3), model M4Coordinate (the e of central point4, f4), model M5Coordinate (the e of central point5, f5)；
S43: the comprehensive matching degree of human body each in obtained crowd is compared with specific threshold, if being more than certain threshold
Value, then successful match is primary, the successful number of statistical match
S5: counting crowd's quantity according to the number of successful match, draw crowd's quantity change curve, when curve or curvilinear trend touch
When sending out anomalous event, warned.
2. Demographics' method based on video monitoring as described in claim 1, which is characterized in that in the step S32, institute
State weight beta={ β of five local feature models1, β2, β3, β4, β5Initial value is respectively 0.2, gone out according to test set statistical estimate
The error of crowd's quantity and true statistical result adjusts the value of weight, until i.e. error is less than preset value.
3. Demographics' method based on video monitoring as described in any one of claim 1-2, which is characterized in that described
Step S5 the following steps are included:
S51: counting crowd's quantity according to the number of successful match, draws crowd's quantity change curve, and in triggering anomalous event
Alarm；
S52: recording crowd's number curve of triggered anomalous event, is placed into possible anomalous event set；
S53: it when curve or trend occur again, is given warning in advance according to warning level；If subsequent triggers anomalous event, can
The danger classes of the curve element can be increased in anomalous event set, conversely, danger classes declines；
S54: the height according to danger classes is determining or adjusts warning level.
4. Demographics' method based on video monitoring as described in any one of claim 1-2, which is characterized in that described
In step S2, foreground segmentation is carried out using mixing dynamic texture model to real-time monitoring images, comprising the following steps:
S21: by real-time monitoring images binaryzation, and a mixing dynamic texture model is established；
S22: the unknown parameter in the mixing dynamic texture model is solved using EM algorithm, so that it is determined that the mixing dynamic line
Manage model；
S23: according to the mixing dynamic texture model, the maximum likelihood probability of monitoring image mixing dynamic texture is calculated, is obtained
Texture class label is clustered according to texture class label by dynamic texture is mixed；
S24: the texture after scanning cluster, using the similar dynamic texture for the crowd that characterizes as prospect, remaining texture divides as background
Cut out the dynamic texture of the crowd of different directions walking.
5. Demographics' method based on video monitoring as claimed in claim 4, which is characterized in that acquire pre- monitoring area
Data determine the weight of mixing five local feature models of dynamic texture model and human body using training set as training set.
6. a kind of Demographics' system based on video monitoring characterized by comprising
Image acquisition unit obtains the real-time monitoring images of pre- monitoring area；
Crowd's cutting unit carries out foreground segmentation to real-time monitoring images, is partitioned into crowd；
Complete characteristics of human body's model is divided into five local feature models, and is each local feature model by feature division unit
Configure corresponding weight；
Characteristic matching unit extracts the feature of the crowd after segmentation, and with five local feature models carry out
With detection, calculated according to the positional shift between weight and the five local feature models and complete characteristics of human body's model of extraction
Comprehensive matching degree, if being more than specific threshold, successful match；
Count alarm unit, crowd's quantity counted according to the number of successful match, draws crowd's quantity change curve, when curve or
When curvilinear trend triggers anomalous event, warned.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710395675.4A CN107133607B (en) | 2017-05-27 | 2017-05-27 | Demographics' method and system based on video monitoring |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710395675.4A CN107133607B (en) | 2017-05-27 | 2017-05-27 | Demographics' method and system based on video monitoring |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107133607A CN107133607A (en) | 2017-09-05 |
| CN107133607B true CN107133607B (en) | 2019-10-11 |
Family
ID=59734106
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710395675.4A Active CN107133607B (en) | 2017-05-27 | 2017-05-27 | Demographics' method and system based on video monitoring |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107133607B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108009477B (en) * | 2017-11-10 | 2020-08-21 | 东软集团股份有限公司 | Image people flow number detection method and device, storage medium and electronic equipment |
| CN110795998B (en) * | 2019-09-19 | 2023-03-24 | 深圳云天励飞技术有限公司 | People flow detection method and device, electronic equipment and readable storage medium |
| CN110693501A (en) * | 2019-10-12 | 2020-01-17 | 上海应用技术大学 | Wireless walking gait detection system based on multi-sensor fusion |
| CN112699265B (en) * | 2019-10-22 | 2024-07-19 | 商汤国际私人有限公司 | Image processing method and device, processor and storage medium |
| CN110929648B (en) * | 2019-11-22 | 2021-03-16 | 广东睿盟计算机科技有限公司 | Monitoring data processing method and device, computer equipment and storage medium |
| CN114900669A (en) * | 2020-10-30 | 2022-08-12 | 深圳市商汤科技有限公司 | Scene monitoring method and device, electronic equipment and storage medium |
| CN112528861B (en) * | 2020-12-11 | 2024-05-31 | 中国铁道科学研究院集团有限公司电子计算技术研究所 | Foreign matter detection method and device applied to ballast bed in railway tunnel |
| CN112597964B (en) * | 2020-12-30 | 2021-10-22 | 上海应用技术大学 | Method for counting layered multi-scale crowd |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831472A (en) * | 2012-08-03 | 2012-12-19 | 无锡慧眼电子科技有限公司 | People counting method based on video flowing image processing |
| CN104573811A (en) * | 2015-01-08 | 2015-04-29 | 杭州天迈网络有限公司 | Pedestrian flow counting method based on infrared image and color image fusion |
| CN105138983A (en) * | 2015-08-21 | 2015-12-09 | 燕山大学 | Pedestrian detection method based on weighted part model and selective search segmentation |
| CN105184245A (en) * | 2015-08-28 | 2015-12-23 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Multi-characteristic fusion population density estimation method |
-
2017
- 2017-05-27 CN CN201710395675.4A patent/CN107133607B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831472A (en) * | 2012-08-03 | 2012-12-19 | 无锡慧眼电子科技有限公司 | People counting method based on video flowing image processing |
| CN104573811A (en) * | 2015-01-08 | 2015-04-29 | 杭州天迈网络有限公司 | Pedestrian flow counting method based on infrared image and color image fusion |
| CN105138983A (en) * | 2015-08-21 | 2015-12-09 | 燕山大学 | Pedestrian detection method based on weighted part model and selective search segmentation |
| CN105184245A (en) * | 2015-08-28 | 2015-12-23 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Multi-characteristic fusion population density estimation method |
Non-Patent Citations (2)
| Title |
|---|
| "Modeling, Clustering, and Segmenting Video with Mixtures of Dynamic Textures";Antoni B. Chan，Nuno Vasconcelos;《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》;20080531;第30卷(第5期);第1-18页 * |
| "基于加权可变形部件模型的行人检测算法";李天朔;《中国优秀硕士学位论文全文数据库 信息科技辑》;20170315;第2、6、9、40-42页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107133607A (en) | 2017-09-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107133607B (en) | Demographics' method and system based on video monitoring | |
| CN108053427A (en) | A kind of modified multi-object tracking method, system and device based on KCF and Kalman | |
| Li et al. | In-field tea shoot detection and 3D localization using an RGB-D camera | |
| US9928423B2 (en) | Efficient retrieval of anomalous events with priority learning | |
| CN110852283A (en) | Helmet wearing detection and tracking method based on improved YOLOv3 | |
| CN108229509A (en) | For identifying object type method for distinguishing and device, electronic equipment | |
| US9805256B2 (en) | Method for setting a tridimensional shape detection classifier and method for tridimensional shape detection using said shape detection classifier | |
| CN109644255A (en) | Mark includes the method and apparatus of the video flowing of a framing | |
| KR102132722B1 (en) | Tracking method and system multi-object in video | |
| JP2023528126A (en) | Analysis and selection in aquaculture | |
| CN106778687A (en) | Method for viewing points detecting based on local evaluation and global optimization | |
| CN106952293B (en) | Target tracking method based on nonparametric online clustering | |
| CN105183758A (en) | Content recognition method for continuously recorded video or image | |
| Stahlschmidt et al. | Applications for a people detection and tracking algorithm using a time-of-flight camera | |
| CN109376584A (en) | A kind of poultry quantity statistics system and method for animal husbandry | |
| CN108537129A (en) | The mask method of training sample, device and system | |
| CN109858552A (en) | A kind of object detection method and equipment for fine grit classification | |
| WO2021190470A1 (en) | Electronic device positioning method and apparatus, and server, system and storage medium | |
| CN110111370A (en) | A kind of vision object tracking methods based on TLD and the multiple dimensioned space-time characteristic of depth | |
| CN109409294A (en) | The classification method and system of trapping event based on object motion trajectory | |
| CN106529470A (en) | Gesture recognition method based on multistage depth convolution neural network | |
| Jahangiri et al. | Information pursuit: A Bayesian framework for sequential scene parsing | |
| CN109492573A (en) | A kind of pointer read method and device | |
| CN116110586B (en) | Elephant health management system based on YOLOv5 and SlowFast | |
| CN102867214A (en) | Counting management method for people within area range |
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 |