CN105913464A - Multi-body target online measurement method based on videos - Google Patents
Multi-body target online measurement method based on videos Download PDFInfo
- Publication number
- CN105913464A CN105913464A CN201610204926.1A CN201610204926A CN105913464A CN 105913464 A CN105913464 A CN 105913464A CN 201610204926 A CN201610204926 A CN 201610204926A CN 105913464 A CN105913464 A CN 105913464A
- Authority
- CN
- China
- Prior art keywords
- point
- human
- foot
- foreground image
- head feature
- 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
- 238000000691 measurement method Methods 0.000 title abstract 2
- 238000000034 method Methods 0.000 claims abstract description 59
- 239000011159 matrix material Substances 0.000 claims abstract description 37
- 230000000877 morphologic effect Effects 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims description 14
- 238000001914 filtration Methods 0.000 claims description 12
- 238000001514 detection method Methods 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 11
- 238000000926 separation method Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 3
- 238000005259 measurement Methods 0.000 description 26
- 238000010586 diagram Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 230000011218 segmentation Effects 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000005315 distribution function Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241001212149 Cathetus Species 0.000 description 1
- 210000003484 anatomy Anatomy 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005530 etching Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000105 evaporative light scattering detection Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a multi-body target height real-time online measurement method based on videos. The method comprises steps that 1, a camera is calibrated, an internal parameter matrix and an external parameter matrix of the camera are acquired, and infinite vertical vanishing points are calculated according to the internal parameter matrix and the external parameter matrix; 2, a foreground image of moving bodies in a to-be-detected video is acquired; 3, morphological processing on the foreground image is carried out; 4, a rectangular bounding box of each moving body position in the foreground image is acquired; 5, head characteristic points and foot characteristic points are extracted from the rectangular bounding box of each moving body position; and 6, height of each moving body is solved according to the extracted head characteristic points and the extracted foot characteristic points.
Description
Technical field
The present invention relates to a kind of many human body targets real-time online measuring method based on video, belong to computer vision technique
Field.
Background technology
Dynamic target measurement based on video refers to utilize computer vision technique to measure the three-dimensional of dynamic object in video
Information, it is a basic task of video analysis, and wherein the height to the movement human in video measures is dynamic mesh
One important technology of mapping amount.
Conventional object height measuring method has method based on stereoscopic vision or method based on RGB-D camera, compares
In only utilizing the target measurement algorithm of single camera, these algorithms need specific hardware device, and increase the one-tenth of hardware
This, therefore target measurement algorithm based on monocular camera has bigger advantage.Some scholars propose and carry out based on single image
The method that scene is measured, these methods have obtained certain application in building surveying and scene of a crime are measured.So
And the method for great majority measurement based on single image is both for what the rigid-object of profile rule proposed, and to pedestrian or
Other non-rigid objects then cannot be suitable for, and seldom have in the middle of current existing document and survey for movement human target
The algorithm of amount.
Paper " real-time automatic body elevation carrection based on video " that Dong Qiulei etc. deliver (automatization's journal, 2009,35
(2): 137-144) proposing a kind of single object height measuring method based on video, the method first passes through Gaussian Mixture mould
Type obtains the foreground region image of movement human, then extracts head and foot's characteristic point of human body in foreground region image,
Set up constraint equation further according to the characteristic point extracted and ask for the height of approximating anatomy.The method follows the tracks of both feet simultaneously in video
Region, according to the tracking result in both feet region, introduces a geometrical constraint about spatial point corresponding to characteristic point with further
Optimize measurement result.The method can realize the automatic measurement of human height under the camera lens that resolution is relatively low.The method can
Realize human body target single in video is measured, but the elevation information of multiple human body targets in video can not be measured, and
And owing to the method is when being modeled human foot region, be by gauss hybrid models, the color of shoes to be built
Mould, therefore can only measure the human body target that shoes color is bigger with clothes colour contrast.
The method measured for the multiple movement human targets in video, rarely has document to propose concrete solution at present
Method.
Summary of the invention
For shortcoming present in prior art, it is an object of the invention to use computer vision technique, take the photograph single
As one or more human body target height in the video of head shooting measures.
To achieve these goals, the technical solution used in the present invention is as follows:
According to an aspect of the present invention, it provides a kind of many human body targets height real-time online measuring side based on video
Method, comprising:
Step 1, camera is demarcated, obtain the Intrinsic Matrix of camera and outer parameter matrix, according to described intrinsic parameter
Matrix and outer parameter matrix calculate the infinite shadow point that vertically disappears;
Step 2, obtain the foreground image of movement human in video to be detected;
Step 3, described foreground image is carried out Morphological scale-space;
Step 4, obtain each movement human position in described foreground image rectangle surround frame;
Step 5, rectangle from each acquired movement human position surround and extract head feature point and foot frame
Characteristic point;
Step 6, solve the height of each movement human according to the head feature point extracted and foot's characteristic point.
According to a further aspect of the invention, it provides a kind of many human body targets height real-time online measuring based on video dress
Put, comprising:
Demarcating module, for demarcating camera, obtains the Intrinsic Matrix of camera and outer parameter matrix, according to described
Intrinsic Matrix and outer parameter matrix calculate the infinite shadow point that vertically disappears;
Foreground detection module, for obtaining the foreground image of movement human in video to be detected;
Pretreatment module, for carrying out Morphological scale-space to described foreground image;
Human region acquisition module, surrounds for obtaining the rectangle of each movement human position in described foreground image
Frame;
Feature point extraction module, surrounds for the rectangle from each acquired movement human position and extracts head frame
Characteristic point and foot's characteristic point;
Height computing module, for solving each movement human according to the head feature point extracted and foot's characteristic point
Height.The present invention is by obtaining the rectangular area at each human body place in video, from the square of each movement human target
Shape region extracts head and foot's characteristic point of human body, resettles constraint equation and solve the height of each human body.Compare
In previous methods, a kind of many human body targets real-time online measuring method based on video of the present invention has the advantages that
(1) can carry out the height of the movement human target in video measuring the most automatically, and body of not asking for help
The colour contrast of outward appearance (being primarily referred to as do not ask for help body dress ornament color and shoes) is relatively big, can be automatic according to human figure feature
Extract head and foot's characteristic point.
(2) the background modeling algorithm used only can quickly extract motion target area by a two field picture, even if
Illuminance abrupt variation or the situation of background change occur, also can split background and prospect fast and accurately.Further, this algorithm is extracted
The target area agglomerate of movement human relatively complete, can preferably carry out the extraction of single human region.
(3) during human body agglomerate extracts, have employed effective filtering method, discharged due to ambient interferences and image
The impact that noise brings, and for disconnected human region, have employed agglomerate method based on Distance Judgment, extracted
Movement human target area is more accurate.
(4) when determining single human body target region, target tracking algorism based on EKF is used to determine
The region of search of human region.Do so has two benefits: first can more improve the efficiency of algorithm;Furthermore can also prevent by
Feature point detection mistake is caused in the disconnected human body foreground area being partitioned into.
Accompanying drawing explanation
Fig. 1 is the flow chart that the present invention implements Human Height method for automatic measurement based on video.
Fig. 2 A-2B is the foreground zone that the present invention implements the movement human in Human Height method for automatic measurement based on video
Regional partition result, Fig. 2 A is original-gray image, and Fig. 2 B is foreground segmentation result.
Fig. 3 A-3B is many people targeted mass detection that the present invention implements in Human Height method for automatic measurement based on video
With the design sketch followed the tracks of, Fig. 3 A is gray level image, and Fig. 3 B is many people targeted mass design sketch.
Fig. 4 is the extraction result that the present invention implements the head feature point in Human Height method for automatic measurement based on video
Schematic diagram.
Fig. 5 A-5B is the human body target foot areas that the present invention implements in Human Height method for automatic measurement based on video
Schematic diagram with foot's characteristic point.
Fig. 6 is the single human body target elevation carrection that the present invention implements in Human Height method for automatic measurement based on video
Result schematic diagram.
Fig. 7 is multiple human body target elevation carrection that the present invention implements in Human Height method for automatic measurement based on video
Result schematic diagram.
Fig. 8 is the inventive method time efficiency analysis and real-time proof diagram.
Fig. 9 A-9B is the result schematic diagram carrying out the inventive method in part monitoring data testing.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Accompanying drawing, the present invention is described in further detail.
In order to be that those of ordinary skill in the art are better understood from the present invention, first to some of which basic conception and calculation
Explaining property of method illustrates:
1. image coordinate system is the rectangular coordinate system with the image upper left corner as initial point, with pixel as coordinate unit.Wherein u, v
Represent pixel columns in digital picture and line number respectively.
2. in the present invention, camera model is linear camera model, and the Intrinsic Matrix A of camera and outer parameter matrix [R t] are equal
Demarcating in advance, the projection matrix of camera is A [R t], HgIt it is the homography matrix of image plane and scaling board plane.In an embodiment
In, the Intrinsic Matrix H that the present invention mentionsgBeing the upper triangular matrix of 3 × 3 with A, R is the spin matrix of 3 × 3, and t is 3 × 1
Translation vector.
3. the shadow point (Vanishing point) that disappears of straight line refers to that the infinite point on the cathetus of space is in image plane
Subpoint.In the present invention, be referred to as vertically disappearing shadow point by the shadow point that disappears being perpendicular to the straight line of ground level in space.As intrinsic parameter A and
When outer parameter [R t] is known, vertically disappear shadow point mvCan determine according to following formula:
smv=A (r1×r2) (1)
Wherein, s is non-zero scale factor, does not affect result of calculation in homogeneous coordinates computing.r1And r2First row for R
And secondary series.
4. Kalman filtering (Kalman filter) is that one utilizes linear system state equation, is inputted defeated by system
Go out to observe data, system mode is carried out the algorithm of optimal estimation.Owing to observation data include the noise in system and interference
Impact, so optimal estimation is also considered as filtering.When utilizing Kalman filtering to carry out optimal estimation, it is not necessary to storage
Measurement data before.After system obtains new measurement data, utilize new measurement data and previous moment state variable
Estimated value, utilizes the state transition equation of system itself, can calculate the estimating of state variable of current time by the mode of recursion
Evaluation, and its amount of calculation is little, can calculate in real time.Kalman filtering algorithm is the filtering algorithm of linear system, but actual
In the case of, state equation and measurement equation are all probably nonlinear system equation.In this case spreading kalman can be used
Filtering (Extended Kalman Filter, EKF) algorithm, its basic thought is: near filter value, applies Taylor expansion
Nonlinear system is launched, and saves the higher order term of more than second order, thus former nonlinear system has reformed into a linear system,
It is filtered processing with standard Kalman filtering algorithm again.The present invention use expanded Kalman filtration algorithm human body target is entered
Line trace.
4. Vibe background segment (Visual Background Extractor) algorithm is that a kind of background based on pixel is built
Mould or the algorithm of foreground detection, from the different of traditional method maximum, the method is that traditional method is by each in image
Individual pixel carries out statistical modeling, and calculates the probability-distribution function of each pixel, then by the statistics of statistical distribution functions
Parameter (such as meansigma methods and variance) determines whether this pixel belongs to background;Vibe algorithm without probability-distribution function to each
Individual pixel is modeled, but directly sets up a sample set for each background pixel point, is by this in sample set
The pixel value in pixel past and the pixel value composition of its neighborhood points some, then by each new pixel value and sample set
Pixel value compare, it may be judged whether belong to background dot.In the initialization of model, Vibe segmenting Background has only to one
Model just can be initialized by pictures, compared to the algorithm of other background segment, has greater advantage.Additionally, its model
More New Policy has also fully taken into account the spatial coherence between pixel, it is possible to the unexpected change of reply background.The present invention adopts
With Vibe algorithm, movement human target is carried out foreground segmentation.
As it is shown in figure 1, a kind of based on video many human body targets height real-time online measuring method of present invention proposition
Detailed step is as follows:
1. pair camera is demarcated, and obtains the intrinsic parameter of camera and outer parameter matrix, and calculates the image of the infinite shadow point that disappears
Point, in this step, can directly use the camera calibration program of OpenCV to demarcate camera, obtain the image plane of camera
And the homography matrix H between groundg, intrinsic parameter A and outer parameter [R t].When Intrinsic Matrix and outer parameter matrix are known, hang down
Directly disappear shadow point mvFormula (1) can be passed through be calculated;
2. utilize Vibe algorithm to obtain the foreground image of movement human in video to be detected;
In the extraction foreground segmentation step of the present invention, Vibe algorithm is used to carry out foreground segmentation, in one embodiment, base
This parameter may be configured as r=25, φ=20, and wherein r is the radius size of circle centered by current pixel, φ be each pixel with
The number of machine sampling.Fig. 2 A-2B is the foreground image of the movement human utilizing Vibe algorithm to extract, and Fig. 2 A is original gradation figure
Picture, Fig. 2 B is foreground segmentation result.
3. pair described foreground image carries out Morphological scale-space;
In the Morphological scale-space of the foreground image of the present invention, mainly image is carried out burn into expansion and intermediate value obscures three
Individual operation.First foreground image being carried out etching operation, in one embodiment, window size is set to 3 × 3;The most again to front
Scape image carries out expansive working, and arranging window size the most in one embodiment is 3 × 3;Finally to the foreground image after processing
Carrying out intermediate value to obscure, in one embodiment, window size is 3 × 3.
4. obtain the rectangle of everyone body position in foreground image and surround frame;
After the foreground image of the movement human removing fragment through step 3 acquisition, can be by foreground image be carried out
Agglomerate detection and expanded Kalman filtration algorithm determine the rectangular area of each movement human in video, concrete the doing of this step
Method: if movement human occurs in video for the first time, then use the method for agglomerate detection to detect the rectangle bag of this human body target
Enclose box position;If movement human is not to occur in video for the first time, then expanded Kalman filtration algorithm is used to follow the tracks of this motion
The rectangle of human body surrounds frame region, estimates the encirclement frame position of this human body when next frame, and re-starts agglomerate in this region
Detection.
In this step, have 2 needs to further illustrate: (a) foreground image is carried out agglomerate detection time, owing to making an uproar
The interference of sound, will detect that some noise spots, in order to extract human body agglomerate region, rejects area in foreground image and is less than 500
Agglomerate,.Owing to human body target is divided into disconnected region by segmenting Background, need the spacing to different agglomerate barycenter
Merging less than the agglomerate of some fixed threshold, in the methods of the invention, arranging this threshold value is 250;B () is to human body district
When territory rectangle encirclement frame EKF carries out target following, it is assumed that detect the rectangular area obtained through human body agglomerate
It is (x with this rectangle of diagonal angle coordinate representation1, y1, x2, y2), then the rectangular area size when being tracked should be set to
(x1-30, y1-30, x2+ 30, y2+30).Fig. 3 is the many people targeted mass in Human Height method for automatic measurement based on video
The design sketch of detect and track, Fig. 3 A is many people targeted mass gray level image, and Fig. 3 B is many people targeted mass design sketch.
5. the rectangle from everyone acquired body position surrounds frame, extract head and the picture point of foot's characteristic point;
This step is divided into head feature point to extract and two steps of foot's feature point extraction.
Head feature point vheadExtraction step be:
Calculate the shadow point m that vertically disappearsvWith on foreground image distance a little, if mvUnder image coordinate system, v direction divides
Amount is on the occasion of, the then image characteristic point that point is cephalad apex that respective distances is maximum;Otherwise, then the point that respective distances is minimum is head
The image characteristic point v on summit, portionhead, as shown in Figure 4, it illustrates the head in Human Height method for automatic measurement based on video
The extraction result of portion's characteristic point.
The extraction step of foot's characteristic point is:
(1) the peak v on foreground image v direction of principal axis is extractedmaxWith head feature point vhead, then human body can substantially be drawn
Region span on image v direction of principal axis.According to human figure feature, foot areas exists not over whole human region
On image v direction of principal axis 10%.Then it is known that foot areas drop shadow spread on image v direction of principal axis is [θ (vmax-
vhead), vmax], wherein θ is the constant between [0.8,0.9], and Fig. 5 A shows Human Height based on the video side of measurement automatically
Human body target foot areas image in method.
(2) with head feature point mhPoint is initial point, mhPut the shadow point m that vertically disappearsvVector directionBuild for y direction
Vertical interim plane coordinate system, seeks the shade of gray at foot areas edge, owing to people's bipod at the volley exists under this coordinate system
Projection in image is likely to have overlap, therefore takes two kinds of situations to process: if the projection that (a) bipod is on image is not
Overlapping, say, that two pieces of disconnected regions to be detected, then two sections of boundary curves can be detected, by these two sections of curves
Shade of gray direction withThe angular separation marginal point set more than 90 ° is designated as respectivelyWithThen each set is extracted
The point that middle vertical coordinate is maximum, is designated as respectivelyWithIf b projection that () bipod is on image has overlap, say, that can only
Image detects one piece of region connected, then one section of boundary curve can be detected, by shade of gray side on this curve
To withThe angular separation marginal point set more than 90 ° is designated as Tf, from set, then extract the minimum point of abscissa and horizontal seat
The point that mark is maximum, is designated as respectivelyWithFig. 5 B shows the human body in Human Height method for automatic measurement based on video
Target foot characteristic point, in figure, solid circles is the picture point of the foot's characteristic point extracted according to this step.
6. set up Constrained equations and solve Human Height
The Liang Ge foot image characteristic point extracted according to step 5WithMay determine that one under image coordinate system
Linear equation Lf
a1x+b1y+c1=0 (2)
Wherein a1, b1, c1It is constant.
Again by the image characteristic point m of cephalad apexhVertically disappear shadow point mv, under image coordinate system, may determine that other one
Individual linear equation Lv
a2x+b2y+c2=0 (3)
Wherein a2, b2, c2It is constant.
According to the definition of the shadow point that vertically disappears, cephalad apex is perpendicular to Liang Ge foot feature with the vertical line disappearing shadow point
The line of point, the most just says straight line LfWith straight line LvIntersection point be intersection point picture point, note intersection point picture point be mp.Simultaneous (2) and (3)
Straight line L can be tried to achievefWith straight line LvIntersection point mp, m under normal circumstancespThe most at infinity.
Assuming that the homogeneous coordinates on head part summit are P in three dimensionsh=(Xh, Yh, Zh, 1)T, because ground level is generation
The X/Y plane of boundary's coordinate system, therefore in three dimensions, the three-dimensional homogeneous coordinates of head part summit subpoint on ground level are Pp
=(Xh, Yh, O, 1)T, then have
The m that equation (2) and (3) is tried to achieve will be contactedpCoordinate substitutes into (4), can try to achieve Xh, Yh, also just say and have determined that Pp
Space coordinates.
At known HgIn the case of, it may be determined that projection matrix Pro, projection matrix can provide two about number of people summit
Individual constraint equation
smh=ProPh (5)
The X that (4) are tried to achievehAnd YhSubstitute into (5), the head part summit Z axis coordinate Z under least square meaning can be tried to achieveh,
The namely height of people.Fig. 6 is the single human body target that the present invention implements in Human Height method for automatic measurement based on video
Height measurement results, Fig. 7 is that multiple human body targets that the present invention implements in Human Height method for automatic measurement based on video are high
Degree measurement result.
Fig. 8 is to test the video that video data is gathered by Institute of Automation Intelligent Building main entrance east side monitoring camera
For experimental data, the time loss figure that the real-time of algorithm is verified, it can be seen that the operation time of each frame is at 40-
Swinging between 55ms, have fluctuation still general status steady though its curve is even, average every frame processes time 48.79ms, i.e. reality
The frame per second measured is 20.1 frames/s, can meet the online operation time need that human body targets multiple in video carry out height measurement
Ask.It is the versatility for verification algorithm shown in Fig. 9, this algorithm is applied and tests on more data set, measure knot
Fruit is as shown in 9A-9B.
Particular embodiments described above, has been carried out furtherly the purpose of the present invention, technical scheme and beneficial effect
Bright, used by be understood by, the foregoing is only the specific embodiment of the present invention, be not limited to the present invention, all at this
Within the spirit of invention and principle, any modification, equivalent substitution and improvement etc. done, should be included in the protection model of the present invention
Within enclosing.
Claims (10)
1. many human body targets height real-time online measuring method based on video, comprising:
Step 1, camera is demarcated, obtain the Intrinsic Matrix of camera and outer parameter matrix, according to described Intrinsic Matrix
The infinite shadow point that vertically disappears is calculated with outer parameter matrix;
Step 2, obtain the foreground image of movement human in video to be detected;
Step 3, described foreground image is carried out Morphological scale-space;
Step 4, obtain each movement human position in described foreground image rectangle surround frame;
Step 5, rectangle from each acquired movement human position surround and extract head feature point and foot's feature frame
Point;
Step 6, solve the height of each movement human according to the head feature point extracted and foot's characteristic point.
The most the method for claim 1, wherein step 6 specifically includes:
Step 601: determine the equation L of this Liang Ge foot characteristic point place straight line according to the Liang Ge foot characteristic point extractedf;
Step 602: determine described head feature point according to described head feature point and the infinite shadow point that vertically disappears and infinite vertically disappear
The equation L of shadow point place straight linev;
Step 603: according to equation LfAnd LvDetermine perpendicular intersection mp;
Step 604: assume that head feature point homogeneous coordinates are Ph=(Xh, Yh, Zh, 1)T, its subpoint on ground level neat
Secondary coordinate is Pp=(Xh, Yh, 0,1)T, then X is obtained according to following projection equationh, Yh, and then obtain Pp=(Xh, Yh, 0,1)T:
Wherein, s is non-zero scale factor, HgFor the image plane obtained according to camera calibration and the homography matrix of scaling board plane;
Step 605: according to HgDetermine projection matrix Pro, and then determine P according to following projection equationh=(Xh, Yh, Zh, 1)T, ZhI.e.
For Human Height:
smh=ProPh
Wherein, mhFor head feature point.
The most the method for claim 1, wherein step 2 use Vibe algorithm to obtain the foreground image of movement human.
The most the method for claim 1, wherein step 3 carries out Morphological scale-space to described foreground image to include: to institute
State foreground image and carry out burn into expansion and intermediate value fuzzy operation.
The most the method for claim 1, wherein step 4 includes:
If movement human occurs in video to be detected for the first time, then agglomerate detection method is used to detect the rectangle of movement human
Surround frame;If movement human is not to occur in video to be detected for the first time, then using expanded Kalman filtration algorithm to follow the tracks of should
The rectangle of movement human surrounds frame, estimates that the rectangle of this movement human surrounds frame when next frame, and in this rectangle surrounds frame
Re-start agglomerate detection.
6. method as claimed in claim 5, wherein, step 4 includes:
When described foreground image being carried out agglomerate detection, rejecting face amount in foreground image and being less than the agglomerate of predetermined value, and setting
Agglomerate area is in preset range;
Distance between different agglomerate barycenter is merged less than the agglomerate of predetermined threshold.
The most the method for claim 1, wherein in step 5, the extraction of head feature point includes:
Calculate the infinite shadow point m that vertically disappearsvWith on prospect profile distance a little, if mvThe component in v direction under image coordinate
For on the occasion of, then maximum in respective distances point is as head feature point;Whereas if mvThe component in v direction under image coordinate
For negative value, then the point that respective distances is minimum is as head feature point.
The most the method for claim 1, wherein in step 5, the extraction step of foot's characteristic point includes:
Extract the peak v on foreground image v direction of principal axismaxWith head feature point vhead, then can substantially show that human region is at figure
As the span on v direction of principal axis;
With head feature point mhPoint is initial point, mhPut the infinite shadow point m that vertically disappearsvVector directionSet up for y direction and face
Time plane coordinate system;The shade of gray at foot areas edge is sought under this coordinate system;If be detected that two sections of boundary curves, will
On these two sections of curves shade of gray direction withThe angular separation marginal point set more than 90 ° is designated as respectivelyWithThen
Extract the point that in each set, vertical coordinate is maximum, be designated as respectivelyWithIf be detected that one section of boundary curve, by this edge
On curve shade of gray direction withThe angular separation marginal point set more than 90 ° is designated as Tf, then from set, extract horizontal seat
The point of mark minimum and the point of abscissa maximum, be designated as respectivelyWith WithIt is foot's characteristic point.
9. many human body targets height real-time online measuring device based on video, comprising:
Demarcating module, for demarcating camera, obtains the Intrinsic Matrix of camera and outer parameter matrix, according to described internal reference
Matrix number and outer parameter matrix calculate the infinite shadow point that vertically disappears;
Foreground detection module, for obtaining the foreground image of movement human in video to be detected;
Pretreatment module, for carrying out Morphological scale-space to described foreground image;
Human region acquisition module, surrounds frame for obtaining the rectangle of each movement human position in described foreground image;
Feature point extraction module, surrounds for the rectangle from each acquired movement human position and extracts head feature frame
Point and foot's characteristic point;
Height computing module, for solving the body of each movement human according to the head feature point extracted and foot's characteristic point
High.
10. device as claimed in claim 9, wherein, described height computing module Human Height calculated as below:
The equation L of this Liang Ge foot characteristic point place straight line is determined according to the Liang Ge foot characteristic point extractedf;
Described head feature point and the infinite shadow point place that vertically disappears is determined according to described head feature point and the infinite shadow point that vertically disappears
The equation L of straight linev;
According to equation LfAnd LvDetermine perpendicular intersection mp;
Assume that head feature point homogeneous coordinates are Ph=(Xh, Yh, Zh, 1)T, the homogeneous coordinates of its subpoint on ground level are
Pp=(Xh, Yh, 0,1)T, then X is obtained according to following projection equationh, Yh, and then obtain Pp=(Xh, Yh, 0,1)T:
Wherein, s is non-zero scale factor, HgFor the image plane obtained according to camera calibration and the homography matrix of scaling board plane;
According to HgDetermine projection matrix Pro, and then determine P according to following projection equationh=(Xh, Yh, Zh, 1)T, ZhIt is human body body
High:
smh=PrOPh
Wherein, mhFor head feature point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610204926.1A CN105913464A (en) | 2016-04-05 | 2016-04-05 | Multi-body target online measurement method based on videos |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610204926.1A CN105913464A (en) | 2016-04-05 | 2016-04-05 | Multi-body target online measurement method based on videos |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105913464A true CN105913464A (en) | 2016-08-31 |
Family
ID=56744641
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610204926.1A Pending CN105913464A (en) | 2016-04-05 | 2016-04-05 | Multi-body target online measurement method based on videos |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105913464A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106353033A (en) * | 2016-11-15 | 2017-01-25 | 沈阳建筑大学 | Computing method for aero-engine barycenter |
CN108509994A (en) * | 2018-03-30 | 2018-09-07 | 百度在线网络技术(北京)有限公司 | character image clustering method and device |
CN108596098A (en) * | 2018-04-24 | 2018-09-28 | 北京京东尚科信息技术有限公司 | Analytic method, system, equipment and the storage medium of human part |
CN109147033A (en) * | 2018-07-27 | 2019-01-04 | 桂林电子科技大学 | A kind of construction account method based on real-time three-dimensional reconstruction technique |
CN109740458A (en) * | 2018-12-21 | 2019-05-10 | 安徽智恒信科技有限公司 | A kind of figure and features pattern measurement method and system based on video processing |
CN112800841A (en) * | 2020-12-28 | 2021-05-14 | 深圳市捷顺科技实业股份有限公司 | Pedestrian counting method, device and system and computer readable storage medium |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102012213A (en) * | 2010-08-31 | 2011-04-13 | 吉林大学 | Method for measuring foreground height through single image |
-
2016
- 2016-04-05 CN CN201610204926.1A patent/CN105913464A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102012213A (en) * | 2010-08-31 | 2011-04-13 | 吉林大学 | Method for measuring foreground height through single image |
Non-Patent Citations (2)
Title |
---|
姚亚夫等: "《基于位置特征的运动行人检测与跟踪方法》", 《广西大学学报》 * |
董秋雷等: "《基于视频的实时自动人体高度测量》", 《自动化学报》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106353033A (en) * | 2016-11-15 | 2017-01-25 | 沈阳建筑大学 | Computing method for aero-engine barycenter |
CN106353033B (en) * | 2016-11-15 | 2018-11-02 | 沈阳建筑大学 | A kind of aero-engine centroid computing method |
CN108509994A (en) * | 2018-03-30 | 2018-09-07 | 百度在线网络技术(北京)有限公司 | character image clustering method and device |
CN108596098A (en) * | 2018-04-24 | 2018-09-28 | 北京京东尚科信息技术有限公司 | Analytic method, system, equipment and the storage medium of human part |
CN108596098B (en) * | 2018-04-24 | 2021-10-01 | 北京京东尚科信息技术有限公司 | Human body part analysis method, system, device and storage medium |
CN109147033A (en) * | 2018-07-27 | 2019-01-04 | 桂林电子科技大学 | A kind of construction account method based on real-time three-dimensional reconstruction technique |
CN109740458A (en) * | 2018-12-21 | 2019-05-10 | 安徽智恒信科技有限公司 | A kind of figure and features pattern measurement method and system based on video processing |
CN109740458B (en) * | 2018-12-21 | 2021-07-16 | 安徽智恒信科技股份有限公司 | Method and system for measuring physical characteristics based on video processing |
CN112800841A (en) * | 2020-12-28 | 2021-05-14 | 深圳市捷顺科技实业股份有限公司 | Pedestrian counting method, device and system and computer readable storage medium |
CN112800841B (en) * | 2020-12-28 | 2024-05-17 | 深圳市捷顺科技实业股份有限公司 | Pedestrian counting method, device and system and computer readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105913464A (en) | Multi-body target online measurement method based on videos | |
US9846946B2 (en) | Objection recognition in a 3D scene | |
CN103077521B (en) | A kind of area-of-interest exacting method for video monitoring | |
CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
CN107767400B (en) | Remote sensing image sequence moving target detection method based on hierarchical significance analysis | |
CN109598794B (en) | Construction method of three-dimensional GIS dynamic model | |
CN105760846B (en) | Target detection and localization method and system based on depth data | |
KR101618814B1 (en) | Method and Apparatus for Monitoring Video for Estimating Gradient of Single Object | |
CN108197604A (en) | Fast face positioning and tracing method based on embedded device | |
CN105046206B (en) | Based on the pedestrian detection method and device for moving prior information in video | |
CN103093198B (en) | A kind of crowd density monitoring method and device | |
CN108731587A (en) | A kind of the unmanned plane dynamic target tracking and localization method of view-based access control model | |
EP2917874A1 (en) | Cloud feature detection | |
CN104751146B (en) | A kind of indoor human body detection method based on 3D point cloud image | |
CN105225230A (en) | A kind of method and device identifying foreground target object | |
CN104091324A (en) | Quick checkerboard image feature matching algorithm based on connected domain segmentation | |
CN109636771A (en) | Airbound target detection method and system based on image procossing | |
CN104715559B (en) | A kind of Smoke Detection and fire alarm method based on track identification | |
CN106204594A (en) | A kind of direction detection method of dispersivity moving object based on video image | |
CN109460764A (en) | A kind of satellite video ship monitoring method of combination brightness and improvement frame differential method | |
CN108416798B (en) | A kind of vehicle distances estimation method based on light stream | |
CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
CN114202646A (en) | Infrared image smoking detection method and system based on deep learning | |
CN106295657A (en) | A kind of method extracting human height's feature during video data structure | |
CN103700106A (en) | Distributed-camera-based multi-view moving object counting and positioning method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160831 |
|
WD01 | Invention patent application deemed withdrawn after publication |