CN109101888A - A kind of tourist's flow of the people monitoring and early warning method - Google Patents
A kind of tourist's flow of the people monitoring and early warning method Download PDFInfo
- Publication number
- CN109101888A CN109101888A CN201810763293.7A CN201810763293A CN109101888A CN 109101888 A CN109101888 A CN 109101888A CN 201810763293 A CN201810763293 A CN 201810763293A CN 109101888 A CN109101888 A CN 109101888A
- Authority
- CN
- China
- Prior art keywords
- people
- model
- value
- tourist
- pixel value
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of tourist's flow of the people monitoring and early warning methods, are related to intelligent tour field, are capable of the resident amount of real-time statistics tourist, supply the information in scenic spot in safe early warning information about resident amount.The present invention includes: the video image using the camera acquisition intensive sight spot of scenic spot flow of the people；Homogenization Treatments are carried out to the video acquired under different illumination using illumination compensation method, tourist's Objective extraction is carried out based on mixed Gauss model；The judgement of the high density stream of people is with the area ratio of tourist's image segmentation pixel in ROI region, each sight spot threshold value is set, exceeds threshold value, is judged as the high density stream of people, it is tracked by the residence time to the high density stream of people, then starts high density tourist's output program beyond predetermined value and carry out flow of the people monitoring；Domestic visitors calculate use the number of people detection technique based on deep learning network, the front, side, back side head feature of people can be accurately identified, it can be achieved that the high density stream of people accurate detection.Numerical value exceeds preset threshold, then issues early warning.
Description
Technical field
The present invention relates to intelligent tour field more particularly to a kind of tourist's flow of the people monitoring and early warning methods.
Background technique
With the development of compatriots' improvement of living standard and tourist industry, the quantity rapid development of tourist in scenic spot.Especially
It is all that tourist is full in National Holidays Famous sceneries, the degree of comfort decreased for going sight-seeing tourist also brings the safety of visit
Hidden danger.The statistics of scenic spot middle reaches passenger flow has following several method at present: first is that carrying out passenger flow statistics, the technology based on ticketing system
It is suitble in closing scenic spot, it is desirable that hold specific medium, the regional scope of statistics is limited, and open formula scenic spot is not suitable for.Second is that
Based on video monitoring system, personal identification is carried out by face recognition technology, to further realize the system of flowing tourist's quantity
Meter.The technology is influenced seriously by weather, light, and the standard of tourist's quantity statistics is influenced under the natural conditions such as rainy day, dense fog, night
True property.Third is that, using mobile phone signaling data, collecting mobile phone real-time position information by development of Mobile Internet technology, accurately grasping scape
The information such as area's passenger number, position distribution, source place distribution, real-time to crowd, dynamically can be monitored and count, need
The support of mobile communications network, is related to the privacy of user, there is very big limitation in practical applications.These technologies respectively have excellent
It is bad, but it is accomplished that the statistics of tourist's quantity in scenic spot, there are no correlative studys in terms of the resident amount monitoring of tourist.Resident amount refers to
Tourist's quantity of certain time is stopped at some sight spot, tourist's Attraction Degree that resident amount can reflect the sight spot and tourist are in this scape
The duration that point stops is different from the monitoring of the tourist flow of traditional sense.The main reason for scenic spot flow of the people congestion is in the short time
A large amount of tourists pour in and resident for a long time, if only real-time statistics domestic visitors to carry out tourist go sight-seeing safe early warning, can make
It is not comprehensive enough and accurate at early warning mechanism.
Therefore, lack a kind of statistical method for the resident amount of tourist in the prior art, be capable of staying for real-time statistics tourist
Allowance supplies the information in scenic spot in safe early warning information about resident amount.
Summary of the invention
The present invention provides a kind of tourist's flow of the people monitoring and early warning method, deep learning technology can be applied to high density people
Flow detection proposes headform's detection method based on transfer learning, the resident amount of tourist in implementing monitoring scenic spot, realizes
The accurate detection and early warning of scenic spot high density flow of the people.
In order to achieve the above objectives, the present invention adopts the following technical scheme:
A kind of tourist's flow of the people monitoring and early warning method is run using a kind of tourist's flow of the people monitoring early-warning system.A kind of trip
Guest's traffic monitoring early warning system includes: camera, network hard disk video recorder, monitoring host computer and warning box, and camera connects net
Network hard disk video recorder, network hard disk video recorder connect monitoring host computer, and monitoring host computer connects warning box, and monitoring host computer is in discovery failure
When triggering warning box alarm.
A kind of tourist's flow of the people monitoring and early warning method, comprising:
S1, each intraday camera shooting and video of sight spot camera of acquisition, use the Gauss based on illumination compensation to camera shooting and video
Model Detection Algorithm carries out foreground extraction, obtains prospect output.
S2, progress area ratio meter is exported to prospect in ROI (Region of interest, area-of-interest) region
It calculates, Dense crowd tracking is carried out according to sight spot density of stream of people threshold value, when reaching the predetermined time, export Dense crowd figure
Piece.Wherein, start to choose ROI in camera shooting and video first frame image as regional scope to be monitored, carry out prospect according to S1
It extracts, connected component labeling is carried out to the prospect output of extraction, obtains largest connected domain, i.e. prospect agglomerate area in prospect frame；It will
The stationary point region area that prospect agglomerate area is selected divided by frame in present frame, judges whether ratio is greater than early warning value, to beyond early warning
The agglomerate area of value is tracked, and ratio is all larger than early warning value within a preset time, and system provides Dense crowd judgement.
S3, using transfer learning technology, on deep learning network utilize with Dense crowd mark picture carry out
The training of number of people detection model obtains trained model.Model training is completed offline, can load trained model
Online detection output is carried out afterwards.Headform's detection method based on deep learning, number of people detection model use residual error network
Model structure, that is, joined the profound convolutional neural networks of residual block, which includes input layer, convolutional layer, pond layer, entirely
Five articulamentum, output layer parts.Picture is imported by input layer, extracts feature in convolutional layer, dimensionality reduction selection is special in the layer of pond
Sign links validity feature by full articulamentum and realizes number of people detection in output layer.There are many detection algorithm based on deep learning, can
The step of being selected according to actual needs, carrying out headform's training detection by taking R-FCN algorithm as an example below is introduced:
(1) it is realized using open source annotation tool Labeling and the number of people of scenic spot Dense crowd picture is marked, input mark
The number of people picture being poured in is generated by the full convolutional neural networks of FCN (Fully-connection Network, full-mesh network)
The characteristic pattern of picture；
(2) characteristic pattern calculated is inputted into RPN (Region Propsal Network, extracted region network), into
And generate ROIS (Region of Interest S is plural number, multiple semi-cylindrical hills)；Then by the ROIS input pair of generation
The pond the ROI layer of position sensing predicts target area to subnet study；
(3) candidate region of the feature that ROI subnet extracts FCN and RPN output, will be between prediction target and labeled targets
Error carry out backpropagation, calculate trained penalty values, make penalty values reach possible minimum value by successive ignition, with
This classification and positioning to complete number of people region.
(4) by the training of certain number, judge whether network weight is optimal with total losses curve graph, obtaining can
Judge the detection model of the number of people and position.Number of people detection is carried out to the test set picture of selection with the detection model that training obtains,
The judge of model is carried out using accuracy rate and false detection rate as standard.
S4, the Dense crowd picture of output is inputted into trained model, carries out domestic visitors detection, when number is more than threshold
When value, trained model exports alarm signal.Previous frame judgement provides Dense crowd and is resident early warning, opens tourist's demographics
Algorithm.It is counted with the detection that trained headform's detection algorithm carries out high density flow of the people, number is more than preset value can be into
Pedestrian's flow early warning.
Further, in S1, the Gauss model detection algorithm based on illumination compensation includes: by camera shooting and video present frame figure
As carry out single channel luminance proportionization processing, using brightness interpolating construct single channel overall situation matrix of differences, to triple channel image into
Row brightness enhancing processing, the video after obtaining illumination compensation；
Foreground extraction is carried out using mixed Gauss model to the video after illumination compensation, prospect is carried out using morphological operation
Target completely exports, and obtains prospect output.
Further, mixed Gauss model method includes:
SS1, mixed Gauss model is initialized, each gray-scale pixels of video sequence image is equal in calculating period T
Value μ0And variances sigma0 2, use μ0And variances sigma0 2The parameter of k Gauss model is initialized, k is positive integer, μ0And σ0 2Calculation formula
It is as follows
Wherein, ItFor new pixel value, the value of t is 1,2 ... T；
SS2, by each new pixel value ItIt is compared with k-th of Gauss model, until finding matched pixel Distribution value
Model, matching refer to, new pixel value ItFor mean bias with k-th of Gauss model in 2.5 σ, the formula for comparing use is as follows:
|It-μk,t-1|≤2.5σk,t-1
μ in formulak,t-1、σk,t-1The respectively distribution mean value and variance of t-1 moment Gauss model；
If SS3, matched pixel value distributed model meet background necessary requirement, matched pixel value distributed model is corresponding
Pixel be labeled as background parts, otherwise be labeled as foreground part；
If SS4, new pixel value ItMatch with one or several in k Gauss model, illustrates new pixel value ItMore accord with
The distribution for closing current pixel value, needs suitably to increase weight, at this time new pixel value ItMean value, variance, right value update formula such as
Under:
μk,t=(1- α) μk,t-1+αIt
ωk,t=(1- β) ωk,t-1+βθ
Wherein, ωk,tFor the weight of k-th of Gaussian Profile of t moment, ωk,t-1For the power of k-th of Gauss of t-1 moment point distribution
Weight, μk,t, σk,tRespectively the mean value and variance of k-th of Gaussian Profile of t moment, θ are match parameter, when new pixel value meets k
θ=1 when Gaussian Profile, θ=0 when not meeting；α is parameter turnover rate, indicates background changing speed, β is learning rate, when new pixel
θ=1 when value meets k Gaussian Profile, θ=0 when not meeting；
If SS5, in SS2, new pixel value ItThere is no any Gauss model matching, then the smallest Gaussian Profile of weight
Mode is replaced, i.e., the mean value of the mode is current pixel value, and standard deviation is initial the larger value, and weight is smaller value；
SS6, each Gauss model are according to its corresponding ωk,tValue sort from large to small, the Gauss that weight is big, standard deviation is small
Model arrangement is forward, obtains the sequence of Gauss model；
SS7, b Gaussian distribution model before sequence being labeled as background B, B meets following formula, and parameter T indicates background institute accounting,
Value range for given threshold, T is 0.5≤T≤1, and b is positive integer
The beneficial effects of the present invention are:
Gardens scenic spot scene is complicated, bridge, corridor, pavilion, the Room, and the places such as artificial hillock, platform, portal keep video background various, in height
Under density complex scene, situations such as minimum face, a large amount of faces are blocked with the number of people back side of head, make detection algorithm in the prior art
Accuracy is not high, and the present invention improves current human face detection tech, sets about from test object, and Face datection range is expanded
Greatly to the entire number of people.Using the number of people detection model based on deep learning, mass data collection is learnt by neural network,
Number of people detection model based on deep learning can promote the multi-angle of target and the detectability under blocking, algorithm adaptability pole
It is big to improve, performance and performance of the algorithm of target detection in individual detection are improved, to monitor staying for tourist in scenic spot in real time
Allowance realizes the accurate detection and early warning of scenic spot high density flow of the people.
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to needed in the embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for ability
For the those of ordinary skill of domain, without creative efforts, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is the flow diagram of the method for the present invention；
Fig. 2 is the mixed Gaussian video foreground extraction algorithm block diagram based on illumination compensation；
Fig. 3 is color image illumination compensation schematic diagram；
Fig. 4 is R-FCN network structure；
Fig. 5 is the domestic visitors statistical technique route block diagram based on number of people detection model；
Fig. 6 is scenic spot people flow rate statistical and early warning system block diagram based on deep learning and number of people detection model.
Specific embodiment
Technical solution in order to enable those skilled in the art to better understand the present invention, With reference to embodiment to this
Invention is described in further detail.
The present embodiment provides a kind of tourist's flow of the people monitoring and early warning methods, using a kind of tourist's flow of the people monitoring early-warning system
Operation.It includes: camera shooting that a kind of operation of tourist's flow of the people monitoring early-warning system, which includes: a kind of tourist's flow of the people monitoring early-warning system,
Head, network hard disk video recorder, monitoring host computer and warning box, camera connect network hard disk video recorder, and network hard disk video recorder connects
Monitoring host computer is connect, monitoring host computer connects warning box, and monitoring host computer triggers warning box alarm when finding failure.
A kind of tourist's flow of the people monitoring and early warning method, method flow diagram are as shown in Figure 1, comprising:
S1, each intraday camera shooting and video of sight spot camera of acquisition, use the Gauss based on illumination compensation to camera shooting and video
Model Detection Algorithm carries out foreground extraction, obtains prospect output, and process is as shown in Figure 2.
Gauss model detection algorithm based on illumination compensation, as shown in Figure 3, comprising: by camera shooting and video current frame image into
The processing of row single channel luminance proportionization constructs single channel overall situation matrix of differences using brightness interpolating, carries out to triple channel image bright
Spend enhancing processing, the video after obtaining illumination compensation；
Foreground extraction is carried out using mixed Gauss model to the video after illumination compensation, prospect is carried out using morphological operation
Target completely exports, and obtains prospect output.
Wherein, mixed Gauss model method includes:
SS1, mixed Gauss model is initialized, each gray-scale pixels of video sequence image is equal in calculating period T
Value μ0And variance, use μ0And variances sigma0 2To initialize k Gaussian mode
The parameter of type, k are positive integer, μ0And σ0Calculation formula it is as follows
Wherein, ItFor new pixel value, the value of t is 1,2 ... T
SS2, by each new pixel value ItIt is compared with k-th of Gauss model, until finding matched pixel Distribution value
Model, matching refer to, new pixel value ItFor mean bias with k-th of Gauss model in 2.5 σ, the formula for comparing use is as follows:
|It-μk,t-1|≤2.5σk,t-1
μ in formulak,t-1、σk,t-1The respectively distribution mean value and variance of t-1 moment Gauss model；
If SS3, matched pixel value distributed model meet background necessary requirement, matched pixel value distributed model is corresponding
Pixel be labeled as background parts, otherwise be labeled as foreground part；
If SS4, new pixel value ItMatch with one or several in k Gauss model, illustrates new pixel value ItMore accord with
The distribution for closing current pixel value, needs suitably to increase weight, at this time new pixel value ItMean value, variance, right value update formula such as
Under:
μk,t=(1- α) μk,t-1+αIt
ωk,t=(1- β) ωk,t-1+βθ
Wherein, wherein ωk,tFor the weight of k-th of Gaussian Profile of t moment, ωk,t-1For k-th of Gauss of t-1 moment point point
The weight of cloth, μk,t, σk,tRespectively the mean value and variance of k-th of Gaussian Profile of t moment, θ are match parameter, when new pixel value accords with
θ=1 when closing k Gaussian Profile, θ=0 when not meeting；α is parameter turnover rate, indicates background changing speed, and β is that learning rate is to work as
θ=1 when new pixel value meets k Gaussian Profile, θ=0 when not meeting；
If SS5, in SS2, new pixel value ItThere is no any Gauss model matching, then weight Gaussian minimum distributed mode
Type, the i.e. mean value of the mode are current pixel value, and standard deviation is initial the larger value, and weight is smaller value；
SS6, each Gauss model are according to its corresponding ωk,tValue sort from large to small, the Gauss that weight is big, standard deviation is small
Model arrangement is forward, obtains the sequence of Gauss model；
SS7, b Gaussian distribution model before sequence being labeled as background B, B meets following formula, and parameter T indicates background institute accounting,
Value range for given threshold, T is 0.5≤T≤1, and b is positive integer,
S2, progress area ratio meter is exported to prospect in ROI (Region of interest, area-of-interest) region
It calculates, Dense crowd tracking is carried out according to sight spot density of stream of people threshold value, when reaching the predetermined time, export Dense crowd figure
Piece.Wherein, start to choose ROI in camera shooting and video first frame image as regional scope to be monitored, carry out prospect according to S1
It extracts, connected component labeling is carried out to the prospect output of extraction, obtains largest connected domain, i.e. prospect agglomerate area in prospect frame；It will
The stationary point region area that prospect agglomerate area is selected divided by frame in present frame, judges whether ratio is greater than early warning value, to beyond early warning
The agglomerate area of value is tracked, and ratio is all larger than early warning value within a preset time, and system provides Dense crowd judgement.
S3, using transfer learning technology, on deep learning network utilize with Dense crowd mark picture carry out
The training of number of people detection model obtains trained model.Model training is completed offline, can load trained model
Online detection output is carried out afterwards.Headform's detection method based on deep learning, number of people detection model use residual error network
Model structure, that is, joined the profound convolutional neural networks of residual block, which includes input layer, convolutional layer, pond layer, entirely
Five articulamentum, output layer parts.Picture is imported by input layer, extracts feature in convolutional layer, dimensionality reduction selection is special in the layer of pond
Sign links validity feature by full articulamentum and realizes number of people detection in output layer.There are many detection algorithm based on deep learning, can
The step of being selected according to actual needs, carrying out headform's training detection by taking R-FCN algorithm as an example below is introduced, wherein R-
FCN network structure is as shown in Figure 4:
(1) it is realized using open source annotation tool Labeling and the number of people of scenic spot Dense crowd picture is marked, input mark
The number of people picture being poured in is generated by the full convolutional neural networks of FCN (Fully-connection Network, full-mesh network)
The characteristic pattern of picture；
(2) characteristic pattern calculated is inputted into RPN (Region Propsal Network, extracted region network), into
And generate ROIS (Region of Interest, S are plural number, multiple semi-cylindrical hills)；Then by the ROIS input pair of generation
The pond the ROI layer of position sensing predicts target area to subnet study；
(3) candidate region of the feature that ROI subnet extracts FCN and RPN output, will be between prediction target and labeled targets
Error carry out backpropagation, calculate trained penalty values, make penalty values reach possible minimum value by successive ignition, with
This classification and positioning to complete number of people region.
(4) by the training of certain number, judge whether network weight is optimal with total losses curve graph, obtaining can
Judge the detection model of the number of people and position.Number of people detection is carried out to the test set picture of selection with the detection model that training obtains,
As shown in figure 5, carrying out the judge of model using accuracy rate and false detection rate as standard.
S4, the Dense crowd picture of output is inputted into trained model, carries out domestic visitors detection, when number is more than threshold
When value, trained model exports alarm signal.Previous frame judgement provides Dense crowd and is resident early warning, opens tourist's demographics
Algorithm.It is counted with the detection that trained headform's detection algorithm carries out high density flow of the people, number is carried out more than preset value
Flow of the people early warning.Scenic spot people flow rate statistical and early warning system block diagram based on deep learning and number of people detection model, such as Fig. 6 institute
Show.
The beneficial effects of the present invention are:
Gardens scenic spot scene is complicated, bridge, corridor, pavilion, the Room, and the places such as artificial hillock, platform, portal keep video background various, in height
Under density complex scene, situations such as minimum face, a large amount of faces are blocked with the number of people back side of head, make detection algorithm in the prior art
Accuracy is not high, and the present invention improves current human face detection tech, sets about from test object, and Face datection range is expanded
Greatly to the entire number of people.Using the number of people detection model based on deep learning, mass data collection is learnt by neural network,
Number of people detection model based on deep learning can promote the multi-angle of target and the detectability under blocking, algorithm adaptability pole
It is big to improve, performance and performance of the algorithm of target detection in individual detection are improved, to monitor staying for tourist in scenic spot in real time
Allowance realizes the accurate detection and early warning of scenic spot high density flow of the people.
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 subject to the protection scope in claims.
Claims (3)
1. a kind of tourist's flow of the people monitoring and early warning method is run, the system using a kind of tourist's flow of the people monitoring early-warning system
Including camera, network hard disk video recorder, monitoring host computer and warning box, camera connects network hard disk video recorder, network hard disc
Video recorder connects monitoring host computer, and monitoring host computer connects warning box, and monitoring host computer is touched when finding the resident amount of tourist beyond early warning value
Warning box alarm is sent out,
The method characterized by comprising
S1, each intraday camera shooting and video of sight spot camera of acquisition, use the Gauss model based on illumination compensation to camera shooting and video
Detection algorithm carries out foreground extraction, obtains prospect output；
S2, progress area ratio calculating, root are exported to prospect in ROI (Region of interest, area-of-interest) region
Dense crowd tracking is carried out according to sight spot density of stream of people threshold value, when reaching the predetermined time, exports Dense crowd picture；
S3, using transfer learning technology, on deep learning network utilize with Dense crowd mark picture carry out the number of people
The training of detection model obtains trained model；
S4, the Dense crowd picture of output is inputted into trained model, carries out domestic visitors detection, when number is more than threshold value
When, trained model exports alarm signal.
2. the method according to claim 1, which is characterized in that in S1, the Gauss model based on illumination compensation, which detects, to be calculated
Method includes: that camera shooting and video current frame image is carried out the processing of single channel luminance proportionization, complete using brightness interpolating construction single channel
Office's matrix of differences carries out brightness enhancing processing to triple channel image, the video after obtaining illumination compensation；
Foreground extraction is carried out using mixed Gauss model to the video after illumination compensation, foreground target is carried out using morphological operation
Complete output obtains prospect output.
3. method according to claim 2, which is characterized in that mixed Gauss model method includes:
SS1, initialization mixed Gauss model, calculate the mean μ of each gray-scale pixels of video sequence image in period T0With
Variances sigma0 2, use μ0And variances sigma0 2The parameter of k Gauss model is initialized, k is positive integer, μ0And σ0 2Calculation formula it is as follows
Wherein, ItFor new pixel value, the value of t is 1,2 ... T；
SS2, by each new pixel value ItIt is compared with k-th of Gauss model, until finding matched pixel value distributed model,
Matching refers to, new pixel value ItFor mean bias with k-th of Gauss model in 2.5 σ, the formula for comparing use is as follows:
|It-μK, t-1|≤2.5σK, t-1
μ in formulaK, t-1、σK, t-1The respectively distribution mean value and variance of k-th of Gauss model of t-1 moment；
If SS3, matched pixel value distributed model meet background necessary requirement, the corresponding picture of matched pixel value distributed model
Element label is otherwise to be labeled as foreground part；
If SS4, new pixel value ItMatch with one or several in k Gauss model, illustrates new pixel value ItMeet current picture
The distribution of element value, needs suitably to increase weight, at this time new pixel value ItMean value, variance, right value update formula it is as follows:
μK, t=(1- α) μK, t-1+αIt
ωK, t=(1- β) ωK, t-1+βθ
Wherein, ωK, tFor the weight of k-th of Gaussian Profile of t moment, ωK, t-1For the weight of k-th of Gauss of t-1 moment point distribution,
μK, t, σK, tRespectively the mean value and variance of k-th of Gaussian Profile of t moment, θ are match parameter, when new pixel value meets k Gauss
θ=1 when distribution, θ=0 when not meeting；α is parameter turnover rate, indicates background changing speed, β is learning rate；
If SS5, in SS2, new pixel value ItThere is no any Gauss model matching, then the smallest Gaussian distribution model of weight
It is replaced, i.e., the mean value of the model is current pixel value, and standard deviation is initial the larger value, and weight is smaller value；
SS6, each Gauss model are according to its corresponding ωK, tValue sort from large to small, the Gauss model that weight is big, standard deviation is small
It arranges forward, obtains the sequence of Gauss model；
SS7, b Gaussian distribution model before sequence is labeled as background B, B meets following formula, and parameter T indicates background institute accounting, to set
Determine threshold value, the value range of T is 0.5≤T≤1, and b is positive integer,
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810763293.7A CN109101888B (en) | 2018-07-11 | 2018-07-11 | Visitor flow monitoring and early warning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810763293.7A CN109101888B (en) | 2018-07-11 | 2018-07-11 | Visitor flow monitoring and early warning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109101888A true CN109101888A (en) | 2018-12-28 |
| CN109101888B CN109101888B (en) | 2022-06-14 |
Family
ID=64846107
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810763293.7A Active CN109101888B (en) | 2018-07-11 | 2018-07-11 | Visitor flow monitoring and early warning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109101888B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109819208A (en) * | 2019-01-02 | 2019-05-28 | 江苏警官学院 | A kind of dense population security monitoring management method based on artificial intelligence dynamic monitoring |
| CN109948550A (en) * | 2019-03-20 | 2019-06-28 | 北京百分点信息科技有限公司 | A kind of wisdom railway station flow of the people monitoring system and method |
| CN110135274A (en) * | 2019-04-19 | 2019-08-16 | 佛山科学技术学院 | A kind of people flow rate statistical method based on recognition of face |
| CN110390266A (en) * | 2019-06-24 | 2019-10-29 | 黄燕 | A kind of system and its measurement method of the measurement scenic spot flow of the people based on area algorithm |
| CN110517251A (en) * | 2019-08-28 | 2019-11-29 | 嘉应学院 | A kind of scenic spot region overload detection and early warning system and method |
| CN110688924A (en) * | 2019-09-19 | 2020-01-14 | 天津天地伟业机器人技术有限公司 | RFCN-based vertical monocular passenger flow volume statistical method |
| CN111010439A (en) * | 2019-12-16 | 2020-04-14 | 重庆锐云科技有限公司 | Scenic spot comfort level monitoring and early warning method |
| CN111640150A (en) * | 2019-09-20 | 2020-09-08 | 于贵庆 | Video data source analysis system and method |
| CN111885202A (en) * | 2020-08-03 | 2020-11-03 | 南京亚太嘉园智慧空间营造有限公司 | Information processing platform for exhibition hall of internet of things based on VGG algorithm |
| CN112146666A (en) * | 2019-06-27 | 2020-12-29 | 奥迪股份公司 | Vehicle driving route marking method and device, computer equipment and storage medium |
| CN114926973A (en) * | 2022-04-06 | 2022-08-19 | 珠海市横琴渤商数字科技有限公司 | Video monitoring method, device, system, server and readable storage medium |
| CN116523319A (en) * | 2023-06-30 | 2023-08-01 | 中国市政工程西南设计研究总院有限公司 | Comprehensive management method and system for intelligent park |
| CN116542509A (en) * | 2023-06-21 | 2023-08-04 | 广东致盛技术有限公司 | Campus logistics task management method and device |
| CN117041484A (en) * | 2023-07-18 | 2023-11-10 | 中建科工集团运营管理有限公司 | People stream dense area monitoring method and system based on Internet of things |
| CN117041484B (en) * | 2023-07-18 | 2024-05-24 | 中建科工集团运营管理有限公司 | People stream dense area monitoring method and system based on Internet of things |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106845344A (en) * | 2016-12-15 | 2017-06-13 | 重庆凯泽科技股份有限公司 | Demographics' method and device |
| CN107229894A (en) * | 2016-03-24 | 2017-10-03 | 上海宝信软件股份有限公司 | Intelligent video monitoring method and system based on computer vision analysis technology |
| CN107301387A (en) * | 2017-06-16 | 2017-10-27 | 华南理工大学 | A kind of image Dense crowd method of counting based on deep learning |
| CN107679502A (en) * | 2017-10-12 | 2018-02-09 | 南京行者易智能交通科技有限公司 | A kind of Population size estimation method based on the segmentation of deep learning image, semantic |
-
2018
- 2018-07-11 CN CN201810763293.7A patent/CN109101888B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107229894A (en) * | 2016-03-24 | 2017-10-03 | 上海宝信软件股份有限公司 | Intelligent video monitoring method and system based on computer vision analysis technology |
| CN106845344A (en) * | 2016-12-15 | 2017-06-13 | 重庆凯泽科技股份有限公司 | Demographics' method and device |
| CN107301387A (en) * | 2017-06-16 | 2017-10-27 | 华南理工大学 | A kind of image Dense crowd method of counting based on deep learning |
| CN107679502A (en) * | 2017-10-12 | 2018-02-09 | 南京行者易智能交通科技有限公司 | A kind of Population size estimation method based on the segmentation of deep learning image, semantic |
Non-Patent Citations (1)
| Title |
|---|
| 杨文学: "留园景区"园"空间游客驻留分布密度研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109819208A (en) * | 2019-01-02 | 2019-05-28 | 江苏警官学院 | A kind of dense population security monitoring management method based on artificial intelligence dynamic monitoring |
| CN109819208B (en) * | 2019-01-02 | 2021-01-12 | 江苏警官学院 | Intensive population security monitoring management method based on artificial intelligence dynamic monitoring |
| CN109948550A (en) * | 2019-03-20 | 2019-06-28 | 北京百分点信息科技有限公司 | A kind of wisdom railway station flow of the people monitoring system and method |
| CN110135274A (en) * | 2019-04-19 | 2019-08-16 | 佛山科学技术学院 | A kind of people flow rate statistical method based on recognition of face |
| CN110135274B (en) * | 2019-04-19 | 2023-06-16 | 佛山科学技术学院 | Face recognition-based people flow statistics method |
| CN110390266A (en) * | 2019-06-24 | 2019-10-29 | 黄燕 | A kind of system and its measurement method of the measurement scenic spot flow of the people based on area algorithm |
| CN112146666A (en) * | 2019-06-27 | 2020-12-29 | 奥迪股份公司 | Vehicle driving route marking method and device, computer equipment and storage medium |
| CN110517251A (en) * | 2019-08-28 | 2019-11-29 | 嘉应学院 | A kind of scenic spot region overload detection and early warning system and method |
| CN110688924A (en) * | 2019-09-19 | 2020-01-14 | 天津天地伟业机器人技术有限公司 | RFCN-based vertical monocular passenger flow volume statistical method |
| CN111640150A (en) * | 2019-09-20 | 2020-09-08 | 于贵庆 | Video data source analysis system and method |
| CN111010439A (en) * | 2019-12-16 | 2020-04-14 | 重庆锐云科技有限公司 | Scenic spot comfort level monitoring and early warning method |
| CN111885202A (en) * | 2020-08-03 | 2020-11-03 | 南京亚太嘉园智慧空间营造有限公司 | Information processing platform for exhibition hall of internet of things based on VGG algorithm |
| CN111885202B (en) * | 2020-08-03 | 2024-05-31 | 南京亚太嘉园智慧空间营造有限公司 | VGG algorithm-based information processing platform for exhibition hall of Internet of things |
| CN114926973A (en) * | 2022-04-06 | 2022-08-19 | 珠海市横琴渤商数字科技有限公司 | Video monitoring method, device, system, server and readable storage medium |
| CN116542509A (en) * | 2023-06-21 | 2023-08-04 | 广东致盛技术有限公司 | Campus logistics task management method and device |
| CN116523319A (en) * | 2023-06-30 | 2023-08-01 | 中国市政工程西南设计研究总院有限公司 | Comprehensive management method and system for intelligent park |
| CN116523319B (en) * | 2023-06-30 | 2023-09-08 | 中国市政工程西南设计研究总院有限公司 | Comprehensive management method and system for intelligent park |
| CN117041484A (en) * | 2023-07-18 | 2023-11-10 | 中建科工集团运营管理有限公司 | People stream dense area monitoring method and system based on Internet of things |
| CN117041484B (en) * | 2023-07-18 | 2024-05-24 | 中建科工集团运营管理有限公司 | People stream dense area monitoring method and system based on Internet of things |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109101888B (en) | 2022-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109101888A (en) | A kind of tourist's flow of the people monitoring and early warning method | |
| CN107967451B (en) | Method for counting crowd of still image | |
| Li et al. | Statistical modeling of complex backgrounds for foreground object detection | |
| CN110378259A (en) | A kind of multiple target Activity recognition method and system towards monitor video | |
| CN104091176B (en) | Portrait comparison application technology in video | |
| Makris et al. | Automatic learning of an activity-based semantic scene model | |
| CN104094279B (en) | Large-range-first cross-camera visual target re-identification method | |
| CN101231755B (en) | Moving target tracking and quantity statistics method | |
| CN109919981A (en) | A kind of multi-object tracking method of the multiple features fusion based on Kalman filtering auxiliary | |
| CN106600631A (en) | Multiple target tracking-based passenger flow statistics method | |
| CN105512640A (en) | Method for acquiring people flow on the basis of video sequence | |
| CN109697435A (en) | Stream of people's quantity monitoring method, device, storage medium and equipment | |
| CN106875424A (en) | A kind of urban environment driving vehicle Activity recognition method based on machine vision | |
| CN103971386A (en) | Method for foreground detection in dynamic background scenario | |
| CN104378582A (en) | Intelligent video analysis system and method based on PTZ video camera cruising | |
| Chebrolu et al. | Deep learning based pedestrian detection at all light conditions | |
| Hsu et al. | Passenger flow counting in buses based on deep learning using surveillance video | |
| CN103208126A (en) | Method for monitoring moving object in natural environment | |
| CN112232199A (en) | Wearing mask detection method based on deep learning | |
| CN106203513A (en) | A kind of based on pedestrian's head and shoulder multi-target detection and the statistical method of tracking | |
| CN112287827A (en) | Complex environment pedestrian mask wearing detection method and system based on intelligent lamp pole | |
| Tomar et al. | Crowd analysis in video surveillance: A review | |
| CN104809742A (en) | Article safety detection method in complex scene | |
| Gade et al. | Thermal imaging systems for real-time applications in smart cities | |
| Chu et al. | Fully unsupervised learning of camera link models for tracking humans across nonoverlapping cameras |
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 |