CN108363988A - A kind of people counting method of combination characteristics of image and hydrodynamics characteristic - Google Patents
A kind of people counting method of combination characteristics of image and hydrodynamics characteristic Download PDFInfo
- Publication number
- CN108363988A CN108363988A CN201810195310.1A CN201810195310A CN108363988A CN 108363988 A CN108363988 A CN 108363988A CN 201810195310 A CN201810195310 A CN 201810195310A CN 108363988 A CN108363988 A CN 108363988A
- Authority
- CN
- China
- Prior art keywords
- particle
- image
- crowd
- shearing stress
- characteristic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Analysis (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Algebra (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of people counting methods of combination characteristics of image and hydrodynamics characteristic, obtain experiment video image first, obtain foreground moving object using Gaussian modeling method, extract the characteristics of image of sport foreground；Video motion pixel is further considered as particle, particle rapidity information is sought on the basis of being handled according to optical flow method, establishes particle shearing stress model；Motion Particles shearing stress is sought by shearing stress model, particle shearing stress is summed it up to characterize the hydrodynamics characteristic of crowd's particle；The image and characteristic of fluid construction feature vector further combined, completes regression model training as input vector, further realizes the estimation to crowd's number in scene, realize to video monitoring intelligent detection and analysis.The present invention have many advantages, such as reasonable design, be widely used, predict it is accurate.
Description
Technical field
Present document relates to intelligent monitoring population analysis technical fields, more particularly to a kind of combination characteristics of image and fluid
The people counting method of mechanical characteristics.
Background technology
In recent years, taking place frequently for injures and deaths event constantly endangered the property and life security of people in public, all
Such as Bund in Shanghai's tread event, Miyun county in Beijing's swarm and jostlement event etc., this not only causes many injures and deaths and also lives to people
Many harmful effects were generated, therefore crowd massing event causes concern of more public to public safety.Due to
Early stage monitoring safety-security area, video monitoring place one's entire reliance upon manpower, and the defect of the way is not only when accident occurs, to cause
The problem of manpower can not be transferred in time, while certain waste is also resulted in resource, it is in addition to this highly dense by being accomplished manually
Degree crowd's demographics are even more a larger engineering.Therefore in Video security monitoring field, based on the dynamic progress to crowd
Holistic modeling and analysis and research intellectual analysis show more important and urgent.
Currently, people counting method is roughly divided into two major classes, one, individual pedestrian's analysis, each pedestrian in crowd is considered as
Independent individual, it is therefore desirable to which individual in detect and track video monitoring marks pedestrian by cluster, to carry out statistics people
Group's number, disadvantage however is that when numerous crowds are blocked, it is difficult to precise positioning tagging target in the scene.Two,
Crowd is considered as entirety, extracts the characteristics of image of whole man's multiple targets in video by analysis of regression model, special using crowd's target
Sign training regression model, to pass through forecast of regression model pedestrian's number.
From the point of view of present Research, since single pedestrian detection is influenced by the factor of blocking, independent individual pedestrian detection error can
Can it is larger, while in analysis of regression model characteristics of image extraction type it is more single, regression model training on may
It is unable to get preferable fitting, consequently, it is possible to causing to cause some unnecessary errors on crowd's prediction.
Invention content
Present invention aims at providing a kind of reasonable design, the good combination characteristics of image of accuracy and hydrodynamics characteristic
People counting method.
To achieve the above object, following technical scheme is used：Steps are as follows for the method for the invention：
Step 1 obtains video frame images, the moving foreground object of image is obtained according to mixed Gaussian method, before movement
The characteristics of image of scape Objective extraction sport people；
Image slices vegetarian refreshments is considered as Motion Particles by step 2, builds particle shearing stress model；
Step 3 seeks particle shearing stress according to particle shearing stress model, is characterized by being summed up to particle shearing stress
The hydrodynamics characteristic of target group；
The characteristics of image of target group is combined structure mesh by step 4, the feature vector for building crowd with hydrodynamics characteristic
The feature vector of mark crowd completes the estimated number of target group using regression model.Regression model can return for neural network
Return model, can also be Support vector regression model.
Further, in the step 2, structure particle shearing stress model specifically includes：
First, crowd's particle rapidity is sought using optical flow method, it is u and vertical direction speed to define particle levels direction speed
The sum velocity of particle levels and vertical direction is represented for v, uv；
Then, particle shearing stress model is built according to the low of paste in hydrodynamics：
In formula, μ is that coefficient of viscosity perseverance is positive value,For particle current gradient, ds is particle contact area, due to inciting somebody to action
Particle is considered as particle, therefore contact area is negligible, and ds values are 1.
Further, the particle flow velocity gradiometer of (x, y) point is calculated as follows：
In formula, r Euclidean distances between particle；
In carrying out current gradient calculating process, crowd's particle is divided into the particle for being in image border and is in non-side
Two kinds of situations of particle of edge are calculated, and non-edge particle flow velocity gradiometer is calculated as follows in image：
When someone comes in and goes out in experiment frame sequence just, crowd is just at image border at this time, to arrange, place
If being divided at two kinds of situation crowd's particles in first row in the calculating of image border particle, the processing of crowd's particle current gradient is such as
Under：
If the processing of particle current gradient is as follows when crowd's particle is in last row：
Euclidean distance calculates specific as follows between particle：
(x1,y1), (x2,y2) respectively represent generate current difference two crowd's particles position coordinates.
Since the characteristic of fluid of crowd's target need to be characterized, it is therefore desirable into further summing up meter to particle shearing stress
It calculates.In the step 3, it is as follows that particle shearing stress sums it up formula：
fvisIt is particle shearing stress, i is the position of particle, and N indicates population.
Compared with prior art, the method for the present invention reasonable design, convenient and practical, easy to operate, calculating are accurately, specific next
It says：
(1) as a result of hydrodynamics characteristic, the effect of characterization crowd movement's feature can be realized；
(2) feature vector of target group is built as a result of hydrodynamics characteristic and feature combinations, utilized
Regression model can realize the effect to crowd's number Accurate Prediction；
(3) it as a result of the scheme for combining hydrodynamics characteristic, can realize to low middle density crowd accurate counting
Effect.
Description of the drawings
Fig. 1 is the entire block diagram of the method for the present invention；
Fig. 2 is the specific implementation step block diagram of the method for the present invention；
Fig. 3 is present example result figure.
Specific implementation mode
The present invention will be further described below in conjunction with the accompanying drawings：
As shown in Figure 1 and Figure 2, it the described method comprises the following steps：
Step 1 obtains video frame images, the moving foreground object of image is obtained using mixed Gaussian method, before movement
The characteristics of image of scape Objective extraction sport people.Wherein, it is according to the foreground target specific method of mixed Gaussian method acquisition image：
Mixed Gaussian method is selected to extract crowd movement's target prospect, mixed Gaussian background modeling method is to each picture in image
Vegetarian refreshments establishes K Gauss model, for judging which of image pixel for background dot.The size of K can be according to computer
Processing capacity determine that theoretically K values are bigger, and the characteristic information that can be obtained is also more, the effect handled also can
It is better.Specifically judge some pixel whether be background dot way be with by the value of the pixel be brought into before establish
Gauss model in matched, if successful match, judge the point for background dot.To extract crowd's foreground in video；
Wherein, it is specially according to the characteristics of image of moving foreground object extraction sport people：According to special with image in foreground
Sign mapping relations can obtain area features and perimeter feature：
Area features：S is the following formula of areal calculation of crowd's foreground pixel B：
I (i, j) indicates that foreground pixel point, (i, j) indicate that foreground pixel position, B indicate foreground crowd's pixel total collection；
Perimeter feature：Defining the perimeter that C is crowd's foreground pixel P can acquire according to following mapping equation：
I (i, j) indicates that edge pixel point, (i, j) indicate that edge pixel location, P indicate foreground edge pixel total collection.
Image slices vegetarian refreshments is considered as Motion Particles and seeks the velocity information of particle by step 2, further builds the stream of particle
Body shearing stress model.It is specific as follows：
First, the motion velocity information of crowd's particle is obtained according to optical flow method, it is u vertical with it to define particle levels speed
Speed is v, and uv represents the sum velocity of particle levels direction and vertical direction；
Then, the model according to the low of paste structure shearing stress in hydrodynamics：
In formula, μ is that coefficient of viscosity perseverance is positive valueFor particle current gradient, ds is particle contact area, due to by grain
Son is considered as particle, therefore contact area is negligible, and ds values are 1.
The particle flow velocity gradiometer of (x, y) point is calculated as follows：
In formula, r Euclidean distances between particle；
Two kinds of situations of particle that crowd's particle is further divided into the particle for being in image border and is in non-edge
It is calculated, non-edge particle flow velocity gradiometer is calculated as follows in image：
When someone comes in and goes out in experiment frame sequence just, crowd is just at image border at this time, to arrange, place
If in first row at the particle particle of image border, the processing of crowd's particle current gradient is as follows：
If the processing of particle current gradient is as follows when particle is in last row：
Euclidean distance calculates specific as follows between particle：
(x1,y1), (x2,y2) respectively represent generate current difference two crowd's particles position coordinates.
The hydrodynamics characteristic selection of step 3, further macroscopic token target group adds crowd's image particle
The method of sum realizes that particle adduction calculates specific as follows：
fvisIt is the shearing stress of particle, i is the position of particle, and N indicates population.
Since the shearing stress of particle is easily influenced by crowd's limbs, characteristic of fluid is accurately expressed to be further, by using
The method of low-pass filtering does noise reduction processing to characteristic of fluid.
Step 4, foundation crowd's target image characteristics and hydrodynamics characteristic, build the feature vector of target group, utilize
Regression model completes the estimated number of target group.Regression model can be neural net regression model, can also be support to
Amount machine regression model.Such as using feature vector as the input of the model of neural network, neural net regression training prediction is completed.
If Fig. 3 shows that public PETS2009 tests the number prediction actual value and predicted value comparison diagram of 1357View, this
Secondary experiment selects public PETS2009 experiment 1357View as training set, and which part data for training, use by remaining data
In training.
Embodiment described above is only that the preferred embodiment of the present invention is described, not to the model of the present invention
It encloses and is defined, under the premise of not departing from design spirit of the present invention, technical side of the those of ordinary skill in the art to the present invention
The various modifications and improvement that case is made should all be fallen into the protection domain of claims of the present invention determination.
Claims (4)
1. a kind of people counting method of combination characteristics of image and hydrodynamics characteristic, it is characterised in that：The method step is such as
Under：
Step 1 obtains video frame images, the moving foreground object of image is obtained according to mixed Gaussian method, according to sport foreground mesh
The characteristics of image of mark extraction sport people；
Image slices vegetarian refreshments is considered as Motion Particles by step 2, builds particle shearing stress model；
Step 3 seeks particle shearing stress according to particle shearing stress model, and target is characterized by being summed up to particle shearing stress
The hydrodynamics characteristic of crowd；
Step 4, the feature vector that the characteristics of image of target group is combined with hydrodynamics characteristic to structure target group, using returning
Model is returned to complete the estimated number of target group.
2. the people counting method of a kind of combination characteristics of image and hydrodynamics characteristic according to claim 1, feature
It is：In step 2, structure particle shearing stress model specifically includes：
First, crowd's particle rapidity being obtained according to optical flow method, definition particle levels direction speed is u, vertical direction speed is v,
Uv represents the sum velocity of particle levels direction and vertical direction；
Then, particle shearing stress model is built according to the low of paste in hydrodynamics：
In formula, μ is that coefficient of viscosity perseverance is positive value,For particle current gradient, ds is particle contact area, due to by particle
It is considered as particle, therefore contact area is ignored, ds values are 1.
3. the people counting method of a kind of combination characteristics of image and hydrodynamics characteristic according to claim 2, feature
It is：
The particle flow velocity gradiometer of (x, y) point calculates formula：
In formula, r Euclidean distances between particle；
In carrying out current gradient calculating process, crowd's particle is divided into the particle for being in image border and is in non-edge
Two kinds of situations of particle are calculated, and non-edge particle flow velocity gradiometer is calculated as follows in image：
When someone comes in and goes out in experiment frame sequence just, crowd is just at image border at this time, to arrange, for place
In image border, particle calculating is divided into two kinds of situations：
If crowd's particle is in first row, the processing of crowd's particle current gradient is as follows：
If the processing of particle current gradient is as follows when crowd's particle is in last row：
Euclidean distance calculates specific as follows between particle：
(x1,y1), (x2,y2) respectively represent generate current difference two crowd's particles position coordinates.
4. the people counting method of a kind of combination characteristics of image and hydrodynamics characteristic according to claim 2, feature
It is：In step 3, the specific formula of crowd's characteristic of fluid for characterizing target group by being summed up to particle shearing stress is such as
Under：
fvisIt is particle shearing stress, i is the position of particle, and N indicates population.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810195310.1A CN108363988A (en) | 2018-03-09 | 2018-03-09 | A kind of people counting method of combination characteristics of image and hydrodynamics characteristic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810195310.1A CN108363988A (en) | 2018-03-09 | 2018-03-09 | A kind of people counting method of combination characteristics of image and hydrodynamics characteristic |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108363988A true CN108363988A (en) | 2018-08-03 |
Family
ID=63003613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810195310.1A Pending CN108363988A (en) | 2018-03-09 | 2018-03-09 | A kind of people counting method of combination characteristics of image and hydrodynamics characteristic |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108363988A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112669282A (en) * | 2020-12-29 | 2021-04-16 | 燕山大学 | Spine positioning method based on deep neural network |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100322516A1 (en) * | 2008-02-19 | 2010-12-23 | Li-Qun Xu | Crowd congestion analysis |
| CN103218816A (en) * | 2013-04-18 | 2013-07-24 | 中山大学 | Crowd density estimation method and pedestrian volume statistical method based on video analysis |
| CN104063879A (en) * | 2014-06-03 | 2014-09-24 | 上海交通大学 | Pedestrian flow estimation method based on flux and shielding coefficient |
| CN104159114A (en) * | 2013-05-13 | 2014-11-19 | 哈尔滨点石仿真科技有限公司 | Method for calculating optical flow at high moving speed among image frames |
| CN105701469A (en) * | 2016-01-13 | 2016-06-22 | 北京大学深圳研究生院 | Robust population counting method based on cost-sensitive sparse linear regression |
| CN106338914A (en) * | 2016-05-03 | 2017-01-18 | 王林 | Micro-channel fluid distributor mathematical model structure and derivation and optimization method thereof |
| CN106446922A (en) * | 2015-07-31 | 2017-02-22 | 中国科学院大学 | Crowd abnormal behavior analysis method |
| CN106529465A (en) * | 2016-11-07 | 2017-03-22 | 燕山大学 | Pedestrian cause and effect relation identification method based on momentum kinetic model |
| CN106683114A (en) * | 2016-12-16 | 2017-05-17 | 河海大学 | Fluid motion vector estimation method based on feature optical flow |
| CN106874885A (en) * | 2017-03-03 | 2017-06-20 | 燕山大学 | A kind of crowd's method for detecting abnormality based on energy level changes in distribution |
-
2018
- 2018-03-09 CN CN201810195310.1A patent/CN108363988A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100322516A1 (en) * | 2008-02-19 | 2010-12-23 | Li-Qun Xu | Crowd congestion analysis |
| CN103218816A (en) * | 2013-04-18 | 2013-07-24 | 中山大学 | Crowd density estimation method and pedestrian volume statistical method based on video analysis |
| CN104159114A (en) * | 2013-05-13 | 2014-11-19 | 哈尔滨点石仿真科技有限公司 | Method for calculating optical flow at high moving speed among image frames |
| CN104063879A (en) * | 2014-06-03 | 2014-09-24 | 上海交通大学 | Pedestrian flow estimation method based on flux and shielding coefficient |
| CN106446922A (en) * | 2015-07-31 | 2017-02-22 | 中国科学院大学 | Crowd abnormal behavior analysis method |
| CN105701469A (en) * | 2016-01-13 | 2016-06-22 | 北京大学深圳研究生院 | Robust population counting method based on cost-sensitive sparse linear regression |
| CN106338914A (en) * | 2016-05-03 | 2017-01-18 | 王林 | Micro-channel fluid distributor mathematical model structure and derivation and optimization method thereof |
| CN106529465A (en) * | 2016-11-07 | 2017-03-22 | 燕山大学 | Pedestrian cause and effect relation identification method based on momentum kinetic model |
| CN106683114A (en) * | 2016-12-16 | 2017-05-17 | 河海大学 | Fluid motion vector estimation method based on feature optical flow |
| CN106874885A (en) * | 2017-03-03 | 2017-06-20 | 燕山大学 | A kind of crowd's method for detecting abnormality based on energy level changes in distribution |
Non-Patent Citations (5)
| Title |
|---|
| CHAN, Y.-T.等: "Extracting foreground ensemble features to detect abnormal crowd behavior in intelligent video-surveillance system", 《JOURNAL OF ELECTRONIC IMAGING》 * |
| D. KONG等: "A Viewpoint Invariant Approach for Crowd Counting", 《THE 18TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR"06)》 * |
| HABIB ULLAH等: "Density independent hydrodynamics model for crowd coherency detection", 《ELSEVIER》 * |
| XUGUANG ZHANG等: "Small scale crowd behavior classification by Euclidean distance variation-weighted network", 《SPRINGERLINK》 * |
| 卢晓威: "面向视频监控应用的人群分析研究", 《中国优秀硕士学位论文全文数据库_信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112669282A (en) * | 2020-12-29 | 2021-04-16 | 燕山大学 | Spine positioning method based on deep neural network |
| CN112669282B (en) * | 2020-12-29 | 2023-02-14 | 燕山大学 | Spine positioning method based on deep neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10735694B2 (en) | System and method for activity monitoring using video data | |
| WO2020253308A1 (en) | Human-machine interaction behavior security monitoring and forewarning method for underground belt transportation-related personnel | |
| CN108764085B (en) | Crowd counting method based on generation of confrontation network | |
| CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
| Mithun et al. | Detection and classification of vehicles from video using multiple time-spatial images | |
| CN103218816B (en) | A kind of crowd density estimation method and people flow rate statistical method based on video analysis | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| CN106203513B (en) | A kind of statistical method based on pedestrian's head and shoulder multi-target detection and tracking | |
| CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
| CN111709300B (en) | Crowd counting method based on video image | |
| JP2007128513A (en) | Scene analysis | |
| CN102156880A (en) | Method for detecting abnormal crowd behavior based on improved social force model | |
| Zheng et al. | Cross-line pedestrian counting based on spatially-consistent two-stage local crowd density estimation and accumulation | |
| CN109583366B (en) | Sports building evacuation crowd trajectory generation method based on video images and WiFi positioning | |
| CN104159088B (en) | A kind of long-distance intelligent vehicle monitoring system and method | |
| CN103577875A (en) | CAD (computer-aided design) people counting method based on FAST (features from accelerated segment test) | |
| CN106570449B (en) | A kind of flow of the people defined based on region and popularity detection method and detection system | |
| CN103049765A (en) | Method for judging crowd density and number of people based on fish eye camera | |
| KR102199252B1 (en) | Method and Apparatus for Analyzing Traffic Situation | |
| CN108471497A (en) | A kind of ship target real-time detection method based on monopod video camera | |
| CN106056078A (en) | Crowd density estimation method based on multi-feature regression ensemble learning | |
| Ua-Areemitr et al. | Low-cost road traffic state estimation system using time-spatial image processing | |
| CN104123569B (en) | Video person number information statistics method based on supervised learning | |
| CN108363988A (en) | A kind of people counting method of combination characteristics of image and hydrodynamics characteristic | |
| CN105989615A (en) | Pedestrian tracking method based on multi-feature fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180803 |
|
| WD01 | Invention patent application deemed withdrawn after publication |