CN103049765A - Method for judging crowd density and number of people based on fish eye camera - Google Patents
Method for judging crowd density and number of people based on fish eye camera Download PDFInfo
- Publication number
- CN103049765A CN103049765A CN2012105591973A CN201210559197A CN103049765A CN 103049765 A CN103049765 A CN 103049765A CN 2012105591973 A CN2012105591973 A CN 2012105591973A CN 201210559197 A CN201210559197 A CN 201210559197A CN 103049765 A CN103049765 A CN 103049765A
- Authority
- CN
- China
- Prior art keywords
- image
- crowd
- weighted
- regression
- density
- 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
Images
Abstract
The invention provides a method for judging crowd density and number of people based on a fish eye camera. The method comprises the following steps of: (1) acquiring a background image; (2) acquiring a crowd monitoring video image; (3) preprocessing the crowd monitoring video image; (4) performing foreground segmentation; (5) extracting a crowd target characteristic; (6) grading the crowd density; and (7) counting the people. The method aims to expand the monitoring range; the problem of inconsistency in sizes of human bodies and the problem of mutual sheltering of the human bodies in the high-density crowd existing in the conventional crowd monitoring method are solved; and the method has good real-time property and can be applied to a real-time crowd monitoring system.
Description
Technical field
The invention belongs to image processing field, particularly relate to the determination methods of a kind of crowd density in image and number.
Background technology
In recent years, along with the rapid increase of urban population density, a lot of public places such as subway station, market etc. usually can welcome the peak traffic of short-term.Crowded crowd causes various accidents easily, has serious security hidden trouble.Therefore, crowd's intelligent monitoring has become a urgent problem of public safety field.At present, the main path that solves crowded problem is to carry out effectively to evacuate and the shunting measure when the crowd is dense.If but no matter whether the crowd is dense all arranges employed personnel to dredge stream, will cause serious personnel's wasting phenomenon.So, carry out crowd's management time, be a very important task to crowd's density Estimation and demographics.
The method of traditional population surveillance all is based on common rifle formula video camera, and this method monitoring range based on rifle formula video camera is little, blind spot is many.Carry out large-range monitoring and need to select splicing or many places that video camera is installed, must cause like this reduction of real-time and the increase of cost.Aspect accuracy of detection, existing crowd density estimation and demographic method, often can only crowd density lower or medium in just can obtain more satisfactory effect, and during for Dense crowd, it detects successful and reduces.In addition, existing effect is the population surveillance method preferably, and its algorithm is more complicated often, and real-time is relatively poor, is difficult to be applied in the real-time system.
Some technology occur at present, solved to a certain extent the problems referred to above.China Patent No. is to have proposed a kind of method and system of judging crowd density in image in the patent of CN102044073 A, by drawing select target zone in the video image sample that is gathered by image collecting device, block analysis unit, and in described target area, carry out the block analysis of drawing of image block, determined the array configuration of two sorters by coding unit, selected by training unit and to put the letter training sample and each two sorter is trained respectively, borrow group transmission model to obtain maximizing the crowd density graded category of posterior probability by decoding unit.This technology can be used for different scenes and obtain the crowd density grade.But, in some actual population surveillance application processes, crowd density is often larger, blocking between the human body is more serious, in the image that obtains the human body size also inconsistent, image have distortion phenomenon to occur, in this case, adopt stroke block analysis method to estimate that crowd density and number are impossible realize basically, simultaneously, the method computation complexity is high, has also affected to a certain extent real-time and the practicality of this system.
At present, can not only solve the distortion problem of the human body image of making peace not of uniform size in the image in the urgent need to a kind of, and can effectively solve the mutual occlusion issue of people's multiple targets, possess simultaneously computation complexity low, possess the method that the judgement of good real-time and practicality crowd density and number are arranged.
Summary of the invention
Technical matters to be solved by this invention provides and a kind ofly can not only solve the distortion problem of the human body image of making peace not of uniform size in the image, and can effectively solve the mutual occlusion issue of people's multiple targets, possess simultaneously computation complexity low, possess the method that the judgement of good real-time and practicality crowd density and number are arranged.
The invention provides a kind of judgement crowd density based on fisheye camera and the method for number, the method comprises:
(1) background extraction image: when number is zero in the region-of-interest of monitoring, by the method background image of fisheye camera, use mixed Gaussian background modeling;
(2) obtain the population surveillance video image: obtain the population surveillance video image by fisheye camera;
(3) pre-service of population surveillance video image: the real-time population surveillance video image that obtains is carried out the adjustment of resolution and frame per second, the division of region-of-interest ROI;
(4) foreground segmentation: population surveillance video image and background image are carried out calculus of differences, more differentiated image is carried out binary segmentation and namely obtain the foreground target image;
(5) crowd's target's feature-extraction: set up the perspective weighted model of fish eye images, extract the following feature of people's multiple targets by weighted calculation: weighted area, weighting profile girth, weighted edge are counted, Weighted H arris angle point number, the tangential gradient of weighting and weighted area gradient ratio;
(6) crowd density classification: adopt the AdaBoost sorter as the density device of classifying, the crowd density of fish eye images is divided into basic, normal, high Three Estate, and uses successively arabic numeral 1,2,3 expressions;
(7) demographics: adopt multiple linear regression analysis method to set up the regression training model, calculate the number in the current population surveillance video image.
As preferably, the installation site of described fisheye camera is the top of monitoring scene central authorities, and the primary optical axis of video camera is perpendicular to the ground, so that fisheye camera has the monitoring range of 360 ° of vertical 180 ° and levels, can effectively strengthen monitoring range like this.
As preferably, in the pre-service of described population surveillance video image:
The adjustment of its resolution be will input video image wide and high respectively be adjusted to original 1/2, namely resolution adjustment extremely original 1/4, so effectively strengthened the real-time of algorithm;
The adjustment of its frame per second is can not undergo mutation with interior at 1s according to crowd density and number, and the frame per second of video image of input was adjusted to for 1 frame/second, i.e. a crowd's of 1s clock detection density and number have so also strengthened the real-time of algorithm effectively;
The division of its region-of-interest ROI is based on a kind of method based on the match of Radius Constraint circle, and concrete steps are:
A) with the fish eye images binaryzation;
B) effective coverage of filling fish eye images;
C) with Canny operator extraction edge;
D) method of employing Radius Constraint circle match is asked the central point of effective coverage
PAnd radius
R
E) take center point P as the center of circle, with 0.85*
RFor radius tentatively limits a border circular areas as the region-of-interest of population surveillance.
Wherein, described Canny operator is the multistage edge detection algorithm that John F. Canny developed in 1986.
As preferably, described crowd's target signature is specially:
(I) weighted area WA
Wherein,
w r (
x,
y) for coordinate in the fisheye camera perspective weighted model be (
x,
y) the weighted value of pixel,
FBe the foreground image areas in the ROI;
(II) weighting profile girth WCP
Wherein,
CThe contour images that the prospect bianry image is extracted for the profile track algorithm that uses based on chain code;
(III) weighted edge WEPN that counts
Wherein,
EFor using the Canny operator that crowd's image is carried out image behind the edge extracting;
(IV) Weighted H arris angle point is counted WHCN
Wherein,
HCoordinate set for crowd's foreground area Harris angle point;
The tangential gradient WTG of (V) weighting
Wherein,
TBe the tangential gradient image after the feature enhancing;
(VI) weighted area gradient ratio WAGR
Wherein,
WABe the weighted area feature,
WTGBe the tangential Gradient Features of weighting;
As preferably, in the described demographics, adopt multiple linear regression analysis method to set up the regression training model and carry out demographics according to following formula:
Wherein,
~
Be followed successively by from
iExtract in the individual training sample image
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue,
~
Be the estimator of regression parameter, in the demographics process, calculate 6 features of test pattern
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue, the parameter in the substitution following formula successively
~
In, can calculate the number in the current test pattern
y
As preferably, in the described demographics, the estimator of regression parameter
~
Draw in the following manner:
Mathematic(al) representation according to regression model carries out the regression equation training, for each sample image, calculate respectively 6 features as independent variable, and artificial interpretation goes out in each sample image number as dependent variable, according to the least square rule, obtain the estimator of each regression parameter
~
Wherein the mathematic(al) representation of regression model is:
In the formula,
~
Be followed successively by from
iExtract in the individual training sample image
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue,
Be
iThe effective strength of individual training sample image,
~
Be regression coefficient,
NBe the number of training sample image,
Error term for regression model.
The present invention offers the user with density rating and the number information who obtains, in conjunction with the predefined alarm threshold value of user, realize the alarm functions such as warning, reach the purpose of crowd's intelligent monitoring, the predefined alarm threshold value of user can change according to actual conditions certainly.
The present invention is not only in order to enlarge the scope of monitoring, and solve the problem that human body blocks mutually in the inconsistent problem of human dimension in traditional population surveillance method and the Dense crowd, and real-time of the present invention is good, can be applied in the real-time population surveillance system.
The present invention obtains crowd's density and number information mainly for the population surveillance that still is not limited to indoor scene.
Description of drawings
Fig. 1 is the judgement crowd density based on fisheye camera of the present invention and the method flow diagram of number.
Fig. 2 is the image-forming principle synoptic diagram of fisheye camera of the present invention.
Fig. 3 is that the tangential gradient image of fisheye camera of the present invention calculates synoptic diagram.
Fig. 4 is that the Mathematical Morphology Method of a kind of fish eye images of using of the present invention is carried out the morphology radial erosion and is radially expanded the processing procedure synoptic diagram tangential gradient image.
Embodiment
Below in conjunction with the accompanying drawing illustrated embodiment the present invention is described in detail, the example of described embodiment is shown in the drawings, and wherein identical or similar label represents identical or similar element or the element with identical or similar functions from start to finish.Be exemplary below by the embodiment that is described with reference to the drawings, only be used for explaining the present invention, and can not be interpreted as limitation of the present invention.
Present embodiment obtains crowd's density and number information mainly for the population surveillance of indoor scene.As shown in Figure 1, its concrete grammar flow process comprises following steps:
(1) background extraction image: when number is zero in the region-of-interest of monitoring, by the method background image of fisheye camera, use mixed Gaussian background modeling;
(2) obtain the population surveillance video image:
Fisheye camera is installed in monitoring scene, is obtained real-time population surveillance video, its installation site is the top of monitoring scene central authorities, and the primary optical axis of video camera is perpendicular to the ground, so that fisheye camera has the monitoring range of 360 ° of vertical 180 ° and levels.
(3) pre-service of population surveillance video image:
Pre-service comprises two processes among the present invention: the adjustment of resolution and frame per second, the division of region-of-interest (Region of Interest, ROI) ROI.
In order to improve the real-time of algorithm, the video image of input wide and high respectively is adjusted to original 1/2, namely resolution adjustment extremely original 1/4; In view of crowd density and number can not be undergone mutation with interior at 1s, frame per second that therefore will video to be detected was adjusted to for 1 frame/second, i.e. a crowd's of 1s clock detection density and number.
The ROI division methods of crowd's intelligent control method of the present invention is a kind of method based on the match of Radius Constraint circle.Its concrete steps are:
A) with the fish eye images binaryzation;
B) effective coverage of filling fish eye images;
C) with Canny operator extraction edge, wherein, described Canny operator is the multistage edge detection algorithm that John F. Canny developed in 1986.
D) method of employing Radius Constraint circle match is asked the central point of effective coverage
PAnd radius
R
E) take center point P as the center of circle, with
k*
RFor radius tentatively limits a border circular areas as the region-of-interest of population surveillance,
kGenerally get 0.85.
(4) foreground segmentation:
The present invention adopts a kind of mixed Gaussian background subtraction point-score based on the area feedback mechanism to carry out crowd's Target Segmentation, the basic thought of the method is to utilize the weighted area information of calculating to determine whether needing to upgrade current background, specific practice is: the people's multiple targets in the manual observation video, when number is zero in the region-of-interest of monitoring, use the method background image of mixed Gaussian background modeling; In case it is complete that background image creates, system need not all carry out context update by every frame, but the weighted area information by system feedback, and target in the present image region-of-interest what are judged.When the weighted area in this moment region-of-interest was lower than area threshold, background image began to upgrade, otherwise background image does not upgrade.If the difference of the time of the time of last background image updating and current background image updating is less than the regular hour, then background image upgrades and postpones, until the difference of the time that twice background image upgrades is when reaching certain threshold value and current weighted area less than certain threshold value, background image begins to upgrade.Generally speaking, time threshold is made as 5 seconds.After the background extraction image, raw video image and background image are carried out calculus of differences, more differentiated image is carried out binary segmentation and namely obtain the foreground target image.
(5) crowd's target's feature-extraction:
Ask for an interview Fig. 2, the present invention at first according to the image-forming principle of fish eye images, analyzes the perspective structure of fish eye images in the raiser multiple targets, set up the perspective weighted model of a fish eye images.Fig. 2 is the two-dimensional representation of fish eye images imaging, and wherein, the straight line of below represents surface level, and the semicircle in the upper right corner represents the fish-eye lens combination, and the line segment above the semicircle is expressed as the picture target surface.G is the center on the surface level, i.e. the intersection point of primary optical axis and surface level, and g is the projection of G on the imaging target surface.A is the object lens center of lens combination.F is the focal length of lens, and H is the vertical range between the dried up plane of lens.Qr is any point in the surface level ROI, its being projected as on the imaging target surface
q r ,
Q r The light that sends incides A, and the angle that forms with primary optical axis is
θThe circle expression image corresponding with the imaging target surface in the upper left corner.
P G Be picture centre,
P r With
P R Be respectively
q r With
q R Pixel in the corresponding image.
The perspective weighted model of fisheye camera is based on three assumed conditions: the height of all human bodies is all close; All human bodies all are on the same level face; The center of image is the center of video camera imaging target surface.In addition, order
εBe the length of the single pixel of image on the imaging target surface, i.e. pixel dimension, unit is millimeter/pixel; Order
rFor
P r To picture centre
P G Distance,
RFor
P R Arrive
P G Distance, i.e. the radius of fish eye images ROI,
rWith
RUnit also is pixel; Order
d r Be the point on the imaging target surface
q r To the target surface center
gBetween distance,
d R For
q R Arrive
gBetween distance,
d r With
d R Unit be the millimeter; Make point on the surface level
Q r To the object lens center
ABetween distance be
D r , unit is millimeter.
Can obtain any point on the surface level by the equidistant projection formula of the geometric relationship of Fig. 2 and fisheye camera
Q r To the distance between (object lens center A) between the camera be:
Make the RC pixel of image
P G The weights of point are
w G =1, then in the image from
P G The distance of point is
rThe weights of pixel
w r Should be:
After obtaining the perspective weighted model of fish eye images, extract following 6 features of people's multiple targets by weighted calculation:
(I) weighted area WA (Weighted Area)
Wherein,
w r (
x,
y) for coordinate in the fisheye camera perspective weighted model be (
x,
y) the weighted value of pixel,
FBe the foreground image areas in the ROI.
(II) weighting profile girth WCP (Weighted Contour Perimeter)
Wherein,
CThe contour images that the prospect bianry image is extracted for the profile track algorithm that uses based on chain code.
(III) weighted edge WEPN (Weighted Edge Pixel Number) that counts
Wherein,
EFor using the Canny operator that crowd's image is carried out image behind the edge extracting.
(IV) Weighted H arris angle point is counted WHCN (Weighted Harris Corner Number)
Wherein,
HCoordinate set for crowd's foreground area Harris angle point.
The tangential gradient WTG of (V) weighting (Weighted Tangential Gradient)
As shown in Figure 3, P is any point in the fish eye images, its coordinate be (
x,
y), P1 ~ P8 is 8 neighborhoods that P is ordered, O is the center of circle of fish eye images, coordinate be (
x c ,
y c );
Because the direction of straight line PO is centripetal direction, and with the straight line P4P8 at a P4 and some P8 place be same straight line, so the position angle of straight line P4P8 can be expressed as:
The coordinate that P is ordered is known, and according to the polar equation of straight line, the coordinate that can calculate the upper any point of straight line P4P8 is:
Order
kEqual respectively 1 and-1, calculate and round with the method that rounds up according to following formula, then can obtain the coordinate of a P4 and P8
With
With P4,3 of P and P8 are reference point respectively, use the method for calculation level P1 and P2 coordinate in the 2 neighborhood methods namely can calculate the coordinate of P1, P2, P3, P5, P6, P7;
After the coordinate of known point P1 ~ P8, the tangential gradient of 8 neighborhoods that P is ordered can be calculated by following formula:
Wherein,
IBe original-gray image,
G t Be the tangential gradient image of trying to achieve;
After obtaining tangential gradient image, in order to make tangential Gradient Features more obvious, used a kind of Mathematical Morphology Method of fish eye images that tangential gradient image is carried out the morphology processing among the present invention, comprised radial erosion and be radially expanded, its mathematic(al) representation is respectively:
Radial erosion
Be radially expanded:
Wherein,
, as shown in Figure 4, (
x c ,
y c ) be hypothesis image center O point coordinate, (
x , y) be the coordinate that any point P is ordered in the image.
I(
x,
y) be pending bianry image,
T(
i,
j) as the two-value template of radial structure element,
nBe the length of radial structure element,
E(
x,
y) and
D(
x,
y) respectively expression corrosion and expand after image;
After obtaining the tangential gradient image of crowd's image, use one
n* 1 radial structure element carries out radial erosion and is radially expanded operation tangential gradient image, when carrying out the radial erosion operation,
nGet 5; When being radially expanded operation,
nGet 7, carrying out after gradient strengthen to process, just can utilize the tangential Gradient Features of weighting of the tangential gradient image calculating people multiple targets after the enhancing, its mathematic(al) representation is:
Wherein,
TBe the tangential gradient image after the feature enhancing.
(VI) weighted area gradient ratio WAGR (Weighted Area-Gradient Ratio)
Wherein,
WABe the weighted area feature,
WTGBe the tangential Gradient Features of weighting.
(6) crowd density classification
Adopt the AdaBoost sorter as the density device of classifying among the present invention, the crowd density of fish eye images is divided into basic, normal, high Three Estate, and uses successively arabic numeral 1,2,3 expressions.
(7) demographics
The present invention adopts the method for multiple linear regression, and people's multiple targets is carried out demographics, and the mathematic(al) representation of its regression model is:
Wherein,
~
Be followed successively by from
iExtract in the individual training sample image
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue,
Be
iThe effective strength of individual training sample image,
~
Be regression coefficient,
NBe the number of training sample image,
Error term for regression model.
In the regression equation training process, for each sample image, calculate respectively 6 features as independent variable, and artificial interpretation goes out in each sample image number as dependent variable.According to the least square rule, obtain the estimator of each regression parameter
~
, can obtain regression equation:
In the demographics process, calculate 6 features of test pattern
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue, the parameter in the substitution following formula successively
~
In, can calculate the number in the current test pattern
y
Wherein, the step of the training of sorter and regression function is as follows:
When (1) selecting nobody, gather a background image as the initial background image;
(2) gather crowd's video image of training usefulness, each frame sample image is carried out pre-service, comprise the adjustment of resolution, the method for actionradius constraint circle match is carried out ROI and is divided;
(3) use the mixed Gaussian background subtraction point-score based on the area feedback mechanism to carry out cutting apart of people's multiple targets;
(4) weighted area, weighting profile girth, the weighted edge of extracting each frame training sample image counted, Weighted H arris angle point number, the tangential gradient of weighting and these 6 features of weighted area gradient ratio, forms the eigenvectors of 6 dimensions;
(5) eigenvector of the used training sample image of storage;
(6) density rating and the number of artificial each width of cloth sample image of interpretation, and carry out the data storage;
(7) with the eigenvector and the density rating that artificial interpretation obtains of all sample images, train the sorter that obtains training in the input AdaBoost sorter;
(8) with the eigenvector and the number that artificial interpretation obtains of all sample images, respectively as independent variable and the dependent variable of linear function recurrence, use the least square rule to calculate the estimator of regression parameter, obtain homing method.
Wherein, crowd density estimation and demographics step are in real time:
(1) obtains real-time population surveillance video by network cameras;
(2) video image that gathers is carried out pre-service, comprise the adjustment of resolution and frame per second, and the division of ROI;
(3) the mixed Gaussian background subtraction separating method of employing area feedback mechanism carries out foreground segmentation to the image in the ROI;
(4) weighted area, weighting profile girth, the weighted edge of extracting each frame monitoring image counted, Weighted H arris angle point number, the tangential gradient of weighting and these 6 features of weighted area gradient ratio, forms the eigenvectors of 6 dimensions;
(5) with the eigenvector that calculates, input the AdaBoost sorter that has trained, obtain the crowd density grade of current frame image;
(6) with the eigenvector that calculates, as independent variable substitution regression equation, calculate the number of current frame image.
The present invention offers the user with density rating and the number information who obtains, in conjunction with the predefined alarm threshold value of user, realize the alarm functions such as warning, reach the purpose of crowd's intelligent monitoring, the predefined alarm threshold value of user is to change according to actual conditions certainly.
The present invention is not only in order to enlarge the scope of monitoring, and solve the problem that human body blocks mutually in the inconsistent problem of human dimension in traditional population surveillance method and the Dense crowd, and real-time of the present invention is good, can be applied in the real-time population surveillance system.
The present invention obtains crowd's density and number information mainly for the population surveillance that still is not limited to indoor scene.The above embodiment is the preferred embodiment that proves absolutely that the present invention lifts, and protection scope of the present invention is not limited to this.Being equal to that those skilled in the art do on basis of the present invention substitutes or conversion, all within protection scope of the present invention.Protection scope of the present invention is as the criterion with claims.
Claims (6)
1. one kind based on the judgement crowd density of fisheye camera and the method for number, it is characterized in that the method comprises:
(1) background extraction image: when number is zero in the region-of-interest of monitoring, by the method background image of fisheye camera, use mixed Gaussian background modeling;
(2) obtain the population surveillance video image: obtain the population surveillance video image by fisheye camera;
(3) pre-service of population surveillance video image: the real-time population surveillance video image that obtains is carried out the adjustment of resolution and frame per second, the division of region-of-interest ROI;
(4) foreground segmentation: population surveillance video image and background image are carried out calculus of differences, more differentiated image is carried out binary segmentation and namely obtain the foreground target image;
(5) crowd's target's feature-extraction: set up the perspective weighted model of fish eye images, extract the following feature of people's multiple targets by weighted calculation: weighted area, weighting profile girth, weighted edge are counted, Weighted H arris angle point number, the tangential gradient of weighting and weighted area gradient ratio;
(6) crowd density classification: adopt the AdaBoost sorter as the density device of classifying, the crowd density of fish eye images is divided into basic, normal, high Three Estate, and uses successively arabic numeral 1,2,3 expressions;
(7) demographics: adopt multiple linear regression analysis method to set up the regression training model, calculate the number in the current population surveillance video image.
2. as claimed in claim 1 a kind of based on the judgement crowd density of fisheye camera and the method for number, it is characterized in that:
The installation site of described fisheye camera is the top of monitoring scene central authorities.
3. as claimed in claim 1 a kind of based on the judgement crowd density of fisheye camera and the method for number, it is characterized in that: in the pre-service of described population surveillance video image:
The adjustment of its resolution be will input video image wide and high respectively be adjusted to original 1/2;
The adjustment of its frame per second is can not undergo mutation with interior at 1 second according to crowd density and number, and the frame per second of video image of input was adjusted to for 1 frame/second, namely detects a crowd's density and number 1 second;
The division of its region-of-interest ROI is based on a kind of method based on the match of Radius Constraint circle, and concrete steps are:
A) with the fish eye images binaryzation;
B) effective coverage of filling fish eye images;
C) with Canny operator extraction edge;
D) method of employing Radius Constraint circle match is asked the central point of effective coverage
PAnd radius
R
E) take center point P as the center of circle, with
0.85*
RFor radius tentatively limits a border circular areas as the region-of-interest of population surveillance.
4. as claimed in claim 1 a kind of based on the judgement crowd density of fisheye camera and the method for number, it is characterized in that: described crowd's target signature is specially:
(I) weighted area WA
Wherein,
w r (
x,
y) for coordinate in the fisheye camera perspective weighted model be (
x,
y) the weighted value of pixel,
FBe the foreground image areas in the ROI;
(II) weighting profile girth WCP
Wherein,
CThe contour images that the prospect bianry image is extracted for the profile track algorithm that uses based on chain code;
(III) weighted edge WEPN that counts
Wherein,
EFor using the Canny operator that crowd's image is carried out image behind the edge extracting;
(IV) Weighted H arris angle point is counted WHCN
Wherein,
HCoordinate set for crowd's foreground area Harris angle point;
The tangential gradient WTG of (V) weighting
Wherein,
TBe the tangential gradient image after the feature enhancing;
(VI) weighted area gradient ratio WAGR
Wherein,
WABe the weighted area feature,
WTGBe the tangential Gradient Features of weighting.
5. as claimed in claim 1 a kind of based on the judgement crowd density of fisheye camera and the method for number, it is characterized in that: in the described demographics, adopt multiple linear regression analysis method to set up the regression training model and carry out demographics according to following formula:
Wherein,
~
Be followed successively by from
iExtract in the individual training sample image
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue,
~
Be the estimator of regression parameter, in the demographics process, calculate 6 features of test pattern
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue, the parameter in the substitution following formula successively
~
In, can calculate the number in the current test pattern
y
6. as claimed in claim 5 a kind of based on the judgement crowd density of fisheye camera and the method for number, it is characterized in that:
In the described demographics, the estimator of regression parameter
~
Draw in the following manner:
Mathematic(al) representation according to regression model carries out the regression equation training, for each sample image, calculate respectively 6 features as independent variable, and artificial interpretation goes out in each sample image number as dependent variable, according to the least square rule, obtain the estimator of each regression parameter
~
Wherein the mathematic(al) representation of regression model is:
In the formula,
~
Be followed successively by from
iExtract in the individual training sample image
WA,
WCP,
WEPN,
WHCN,
WTGWith
WAGRValue,
Be
iThe effective strength of individual training sample image,
~
Be regression coefficient,
NBe the number of training sample image,
Error term for regression model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105591973A CN103049765A (en) | 2012-12-21 | 2012-12-21 | Method for judging crowd density and number of people based on fish eye camera |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105591973A CN103049765A (en) | 2012-12-21 | 2012-12-21 | Method for judging crowd density and number of people based on fish eye camera |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103049765A true CN103049765A (en) | 2013-04-17 |
Family
ID=48062397
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012105591973A Pending CN103049765A (en) | 2012-12-21 | 2012-12-21 | Method for judging crowd density and number of people based on fish eye camera |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103049765A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104616432A (en) * | 2015-02-04 | 2015-05-13 | 田文华 | Intelligent identification and control method and system for people flow density |
| CN104778468A (en) * | 2014-01-15 | 2015-07-15 | 索尼公司 | Image processing device, image processing method and monitoring equipment |
| CN106651758A (en) * | 2016-12-16 | 2017-05-10 | 深圳市保千里电子有限公司 | Noisy fisheye image-based effective region extraction method and system |
| CN107368823A (en) * | 2017-08-23 | 2017-11-21 | 广州市九安光电技术股份有限公司 | A kind of stream of people's focus monitoring method and system based on panoramic picture |
| CN107784258A (en) * | 2016-08-31 | 2018-03-09 | 南京三宝科技股份有限公司 | Subway density of stream of people method of real-time |
| CN108875709A (en) * | 2018-07-18 | 2018-11-23 | 洛阳语音云创新研究院 | One kind flocks together behavioral value method, apparatus, electronic equipment and storage medium |
| CN109919064A (en) * | 2019-02-27 | 2019-06-21 | 湖南信达通信息技术有限公司 | Demographic method and device in real time in a kind of rail transit cars |
| CN111311603A (en) * | 2018-12-12 | 2020-06-19 | 北京京东尚科信息技术有限公司 | Method and apparatus for outputting target object number information |
| US10706431B2 (en) * | 2014-07-02 | 2020-07-07 | WaitTime, LLC | Techniques for automatic real-time calculation of user wait times |
| CN112733677A (en) * | 2020-12-31 | 2021-04-30 | 桂林海威科技股份有限公司 | People flow rate statistical system and method |
| CN113096406A (en) * | 2019-12-23 | 2021-07-09 | 深圳云天励飞技术有限公司 | Vehicle information acquisition method and device and electronic equipment |
| CN116012768A (en) * | 2022-08-03 | 2023-04-25 | 通号智慧城市研究设计院有限公司 | Crowd density detection method and device, electronic equipment and computer storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6795567B1 (en) * | 1999-09-16 | 2004-09-21 | Hewlett-Packard Development Company, L.P. | Method for efficiently tracking object models in video sequences via dynamic ordering of features |
| CN101325690A (en) * | 2007-06-12 | 2008-12-17 | 上海正电科技发展有限公司 | Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow |
| CN101431664A (en) * | 2007-11-06 | 2009-05-13 | 同济大学 | Automatic detection method and system for intensity of passenger flow based on video image |
-
2012
- 2012-12-21 CN CN2012105591973A patent/CN103049765A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6795567B1 (en) * | 1999-09-16 | 2004-09-21 | Hewlett-Packard Development Company, L.P. | Method for efficiently tracking object models in video sequences via dynamic ordering of features |
| CN101325690A (en) * | 2007-06-12 | 2008-12-17 | 上海正电科技发展有限公司 | Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow |
| CN101431664A (en) * | 2007-11-06 | 2009-05-13 | 同济大学 | Automatic detection method and system for intensity of passenger flow based on video image |
Non-Patent Citations (4)
| Title |
|---|
| 刘珂等: "半径约束最小二乘圆拟合方法及其误差分析", 《光电子.激光》, vol. 17, no. 5, 15 May 2006 (2006-05-15) * |
| 杨丹等: "基于区域生长的鱼眼图像轮廓提取算法", 《计算机工程》, vol. 36, no. 8, 20 April 2010 (2010-04-20) * |
| 胡学敏等: "利用加权面积透视变换对地铁站台进行人群监控", 《武汉大学学报· 信息科学版》, vol. 37, no. 3, 5 March 2012 (2012-03-05) * |
| 苏航等: "视频监控中人群流量和密度估计算法分析", 《视频应用与工程》, vol. 33, no. 11, 17 November 2009 (2009-11-17) * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778468A (en) * | 2014-01-15 | 2015-07-15 | 索尼公司 | Image processing device, image processing method and monitoring equipment |
| US10902441B2 (en) * | 2014-07-02 | 2021-01-26 | WaitTime, LLC | Techniques for automatic real-time calculation of user wait times |
| US10706431B2 (en) * | 2014-07-02 | 2020-07-07 | WaitTime, LLC | Techniques for automatic real-time calculation of user wait times |
| CN104616432B (en) * | 2015-02-04 | 2017-03-15 | 田文华 | A kind of density of stream of people Intelligent Recognition and control method and system |
| CN104616432A (en) * | 2015-02-04 | 2015-05-13 | 田文华 | Intelligent identification and control method and system for people flow density |
| CN107784258A (en) * | 2016-08-31 | 2018-03-09 | 南京三宝科技股份有限公司 | Subway density of stream of people method of real-time |
| CN106651758A (en) * | 2016-12-16 | 2017-05-10 | 深圳市保千里电子有限公司 | Noisy fisheye image-based effective region extraction method and system |
| CN107368823A (en) * | 2017-08-23 | 2017-11-21 | 广州市九安光电技术股份有限公司 | A kind of stream of people's focus monitoring method and system based on panoramic picture |
| CN108875709A (en) * | 2018-07-18 | 2018-11-23 | 洛阳语音云创新研究院 | One kind flocks together behavioral value method, apparatus, electronic equipment and storage medium |
| CN108875709B (en) * | 2018-07-18 | 2021-05-25 | 洛阳语音云创新研究院 | Tie-stacking behavior detection method and device, electronic equipment and storage medium |
| CN111311603A (en) * | 2018-12-12 | 2020-06-19 | 北京京东尚科信息技术有限公司 | Method and apparatus for outputting target object number information |
| CN109919064B (en) * | 2019-02-27 | 2020-12-22 | 湖南信达通信息技术有限公司 | Real-time people counting method and device in rail transit carriage |
| CN109919064A (en) * | 2019-02-27 | 2019-06-21 | 湖南信达通信息技术有限公司 | Demographic method and device in real time in a kind of rail transit cars |
| CN113096406A (en) * | 2019-12-23 | 2021-07-09 | 深圳云天励飞技术有限公司 | Vehicle information acquisition method and device and electronic equipment |
| CN113096406B (en) * | 2019-12-23 | 2022-07-26 | 深圳云天励飞技术有限公司 | Vehicle information acquisition method and device and electronic equipment |
| CN112733677A (en) * | 2020-12-31 | 2021-04-30 | 桂林海威科技股份有限公司 | People flow rate statistical system and method |
| CN112733677B (en) * | 2020-12-31 | 2021-11-30 | 桂林海威科技股份有限公司 | People flow rate statistical system and method |
| CN116012768A (en) * | 2022-08-03 | 2023-04-25 | 通号智慧城市研究设计院有限公司 | Crowd density detection method and device, electronic equipment and computer storage medium |
| CN116012768B (en) * | 2022-08-03 | 2024-03-01 | 通号智慧城市研究设计院有限公司 | Crowd density detection method and device, electronic equipment and computer storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103049765A (en) | Method for judging crowd density and number of people based on fish eye camera | |
| CN108053427B (en) | Improved multi-target tracking method, system and device based on KCF and Kalman | |
| CN106203274B (en) | Real-time pedestrian detection system and method in video monitoring | |
| KR102051829B1 (en) | Bigdata Previewport Multi-resolution video transmission CCTV using License plate detection and Bigdata Previewport Multi-resolution video transmission traffic control method using License plate detection | |
| CN104123544B (en) | Anomaly detection method and system based on video analysis | |
| CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
| Bossu et al. | Rain or snow detection in image sequences through use of a histogram of orientation of streaks | |
| CN103164858B (en) | Adhesion crowd based on super-pixel and graph model is split and tracking | |
| CN103927520B (en) | A kind of backlight environment servant's face detecting method | |
| CN101389004B (en) | Moving target classification method based on on-line study | |
| CN109255350B (en) | New energy license plate detection method based on video monitoring | |
| CN110059581A (en) | People counting method based on depth information of scene | |
| CN111696123A (en) | Remote sensing image water area segmentation and extraction method based on super-pixel classification and identification | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| CN105574488B (en) | It is a kind of to be taken photo by plane the pedestrian detection method of infrared image based on low latitude | |
| CN103824070A (en) | Rapid pedestrian detection method based on computer vision | |
| CN101800890A (en) | Multiple vehicle video tracking method in expressway monitoring scene | |
| CN101976504B (en) | Multi-vehicle video tracking method based on color space information | |
| WO2004042673A2 (en) | Automatic, real time and complete identification of vehicles | |
| CN103258432A (en) | Traffic accident automatic identification processing method and system based on videos | |
| CN102819764A (en) | Method for counting pedestrian flow from multiple views under complex scene of traffic junction | |
| CN102214309A (en) | Special human body recognition method based on head and shoulder model | |
| CN102867349A (en) | People counting method based on elliptical ring template matching | |
| Karpagavalli et al. | Estimating the density of the people and counting the number of people in a crowd environment for human safety | |
| CN110569755A (en) | Intelligent accumulated water detection method based on video |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130417 |