CN108596888A - A kind of Human Height Real-time Generation based on monocular image - Google Patents

A kind of Human Height Real-time Generation based on monocular image Download PDF

Info

Publication number
CN108596888A
CN108596888A CN201810356143.4A CN201810356143A CN108596888A CN 108596888 A CN108596888 A CN 108596888A CN 201810356143 A CN201810356143 A CN 201810356143A CN 108596888 A CN108596888 A CN 108596888A
Authority
CN
China
Prior art keywords
plane
formula
image
monocular image
error
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.)
Withdrawn
Application number
CN201810356143.4A
Other languages
Chinese (zh)
Inventor
李富平
冷霜
李云霞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201810356143.4A priority Critical patent/CN108596888A/en
Publication of CN108596888A publication Critical patent/CN108596888A/en
Withdrawn legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • 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
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/147—Discrete orthonormal transforms, e.g. discrete cosine transform, discrete sine transform, and variations therefrom, e.g. modified discrete cosine transform, integer transforms approximating the discrete cosine transform
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/10—Selection of transformation methods according to the characteristics of the input images
    • 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
    • G06T2207/10004—Still image; Photographic 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

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Software Systems (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Operations Research (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Discrete Mathematics (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a kind of Human Height Real-time Generation based on monocular image, including build computation model according to monocular image, the sliding-model control of computation model, the linearization process of model, setting constraints are simultaneously iterated calculating.This method directly is handled and is calculated to obtain Human Height value to the collected monocular image of all kinds of cameras, and without increasing aided-detection device at the scene, deployment is flexible, can greatly save cost.Particularly, this method has the ability that batch calculates height, in magnanimity information comparison, has unique efficiency and precision advantage.

Description

A kind of Human Height Real-time Generation based on monocular image
Technical field
The invention belongs to computer communication, image procossing and data processing fields, and in particular to one kind being based on monocular image Human Height Real-time Generation.
Background technology
Currently, the Human Height measurement based on image or video is required for based on calibrating template, calibration is template used Or reference dimension measures it is known that height then is unfolded according to optical principle, therefore the application scenario of the technology is extremely limited to.At certain Under a little special occasions or requirement, for example technology investigates occasion or relies on common monitor video image spread height and measures in real time, The prior art is difficult to solve the problems, such as this.
Invention content
To solve the above problems, the present invention provides a kind of Human Height Real-time Generation based on monocular image, it can To be not required to demarcate in advance, Human Height value in monocular image can be generated in real time, and can be as unit of group, batch calculates Height values in image.
The technical scheme is that：
A kind of Human Height Real-time Generation based on monocular image, it is characterised in that：The method includes following steps Suddenly：
A. according to monocular image, computation model is built；
B. sliding-model control is carried out to computation model；
C. linearization process is carried out to system-computed model；
D., constraints is set and is iterated calculating, directly generates height result.
Preferably, the specific implementation of the step A is：
Under the conditions of Lambert ideal images, by Lambert reflection equations it is found that gradation of image and the people of monocular image Image plane height is in binary nonlinear relationship, proposes that the computation model of the method for the invention, process are as follows based on this：
Assuming that the in-plane of original monocular image is denoted as (Dx,Dy), point light source incident direction is denoted as (DLx,DLy), image Gray scale be denoted as g (x, y),
Then there is following formula according to Lambert reflection equations：
To reduce the uncertainty for building result of calculation caused by above-mentioned model under a variety of ideal conditions, introduce following Compensation method：
Wherein, EnWhite Gaussian noise error is indicated, for compensating the error caused by noise；ErIt indicates roughness error, uses In compensation in-plane error；EgGradient similitude error is indicated, produced by compensating equation during sliding-model control Error；(Zx,Zy) indicate discrete picture in-plane；
Therefore, systematic error ErrIt may be defined as：
Preferably, the specific implementation of the step B is：
The process of sliding-model control is：
1. the minimum conditions of solving system error and gradient similitude error：
To formula (3) derivation, and solves derivative equation and can obtain：
2. can be obtained to formula (4) sliding-model control：
3. obtaining final discretization calculates formula：
Pull-type approximate discretization transformation factor is introduced,WhereinIndicate in-plane L's Mean direction, δ indicate the pixel of monocular image away from being converted to formula (5) with the transformation factor, human body body can be obtained High and gradient discretization calculates formula, as follows：
To sum up, the discretization for just having obtained height and gradient similarity calculates formula.
Preferably, the specific implementation of the step C is：
The problem of annual reporting law can not restrain in special circumstances is conciliate to improve computational efficiency, to system-computed model into line Propertyization processing：
Using Talor formula, using the result of calculation of laststate as reference data, the line of discretization formula (6) can be obtained Property approximation formula is as follows：
Substituting the above to (6) can obtain：
Wherein, σ indicates variable quantity of each variable relative to t state reference benchmark；
Based on the above, can be the variation for solving in-plane by the non-linear transfer that in-plane calculatesLinear equation, in-plane (D then can be calculated by following formulax,Dy) value：
Preferably, the specific implementation of the step D is：
1. to obtain faster calculating speed and less EMS memory occupation, using staggered-mesh method：
Build staggered-mesh：The pixel spot size of monocular figure is w × s, in-plane (Dx,Dy) and gradation of image g (x, y) Array dimension also be w × s；The pixel of human body contour outline and in-plane pixel are staggered 0.5 pixel, human body contour outline pixel Matrix dimensionality is (w+1) × (s+1)；In-plane is also (w+1) × (s+1) to the local derviation dimension of both direction；
To sum up, the Discrete Linear equation solved is as follows in the first-order partial derivative of its in-plane：
2. introducing the constraint of two classes, first, reference point constrains, second is that boundary condition constrains, method is as follows：
A. reference point constraint is the point of known in-plane and height to be added in camera lens visual angle, and ensure that the point can It is shown on monocular image；
B. boundary condition constrains, that is, the in-plane of boundary pixel point is added, uses interpolation method outside with central pixel point Interpolation, then the interpolation formula of four direction be：
Wherein, central pixel point refers to satisfactionImage-region is w × s pixels,
On the above basis, it is iterated using formula (8), then Human Height H=max (z1,z2…zn), wherein n is The dimension of portrait profile set of pixels in monocular image.
Beneficial effects of the present invention are：
The method of the invention, which is directly handled monocular image, obtains Human Height, without increasing auxiliary at the scene Detection device, deployment is flexible, can greatly save cost.Particularly, this method has the ability that batch calculates height, in magnanimity In information comparison, there is unique Efficiency and accuracy advantage.
Description of the drawings
Fig. 1 is a kind of overall flow figure of the Human Height Real-time Generation based on monocular image；
Fig. 2 is the staggered-mesh of Discrete Linear equation.
Specific implementation mode
The overall flow of Human Height Real-time Generation of the present invention based on monocular image is as shown in Figure 1, solution Certainly method used by its problem includes the following steps：
A. according to monocular image, computation model is built：
In monocular image, human body can be considered the uniform diffuse reflection body in certain marginal range.Meanwhile it being managed in Lambert Think under image-forming condition, by Lambert reflection equations it is found that the gradation of image of monocular image is non-thread in binary with portrait level Sexual intercourse proposes that the computation model of the method for the invention, process are as follows based on this：
Assuming that the in-plane of original monocular image is denoted as (Dx,Dy), point light source incident direction is denoted as (DLx,DLy), image Gray scale be denoted as g (x, y).
Then there is following formula according to Lambert reflection equations：
It is one in the process for calculating height using gradation of image since above-mentioned model is built upon under a variety of ideal conditions The solution procedure of a Ill-posed characteristic will greatly increase the uncertainty of result of calculation, and to solve this problem, this method introduces Following compensation method：
Wherein, EnWhite Gaussian noise error is indicated, for compensating the error caused by noise；ErIt indicates roughness error, uses In compensation in-plane error；EgGradient similitude error is indicated, produced by compensating equation during sliding-model control Error.(Zx,Zy) indicate discrete picture in-plane.
Therefore, systematic error ErrIt may be defined as：
B. the sliding-model control of system-computed model：
The present invention solves the problems, such as that method used by it is using the pixel of image as Computing Meta, therefore in establishment step A After the computation model of the system, it is also necessary to carry out sliding-model control, process is as follows：
1. the minimum conditions of solving system error and gradient similitude error：
To formula (3) derivation, and solves derivative equation and can obtain：
2. can be obtained to formula (4) sliding-model control：
3. obtaining final discretization calculates formula：
Pull-type approximate discretization transformation factor is introduced,WhereinIndicate in-plane L's Mean direction, δ indicate the pixel of monocular image away from being converted to formula (5) with the transformation factor, human body body can be obtained High and gradient discretization calculates formula, as follows：
To sum up, the discretization for just having obtained height and gradient similarity calculates formula.
C. it is to improve computational efficiency to conciliate the problem of annual reporting law can not restrain in special circumstances, system-computed model is carried out Linearization process：
Discretization formula can be obtained using the result of calculation of laststate as reference data using Talor formula in this method (6) linearization approximate formula is as follows：
Substituting the above to (6) can obtain：
Wherein, σ indicates variable quantity of each variable relative to t state reference benchmark.
Based on the above, can be the variation for solving in-plane by the non-linear transfer that in-plane calculatesLinear equation, in-plane (D then can be calculated by following formulax,Dy) value：
D., constraints is set and is iterated calculating：
1. being calculated by after step A, B, C, used by this method and being changed into the linear of the in-plane for solving discretization Equation to obtain faster calculating speed and less EMS memory occupation, therefore uses staggered-mesh method.
The structure of staggered-mesh is as shown in Fig. 2, solid line point indicates that the pixel of monocular figure, size are w × s, in-plane (Dx,Dy) and the array dimension of gradation of image g (x, y) be all w × s；Dotted line point indicates human body contour outline relative value, the pixel of profile Point is staggered 0.5 pixel with in-plane pixel, and human body contour outline picture element matrix dimension is (w+1) × (s+1)；In-plane pair The local derviation dimension of both direction is also (w+1) × (s+1).
To sum up, the Discrete Linear equation solved is as follows in the first-order partial derivative of its in-plane：
2. to ensure that iterative algorithm can be expected to restrain under the premise of all kinds of inputs, constraints is introduced.In addition, in order to So that algorithm directly export Human Height value, rather than the relative value that staggered-mesh is calculated, the constraint of two classes is introduced, first, joining Examination point constrains, second is that boundary condition constrains, method is as follows：
A. reference point constraint is the point of known in-plane and height to be added in camera lens visual angle, and ensure that the point can It is shown on monocular image.
B. boundary condition constrains, that is, the in-plane of boundary pixel point is added, uses interpolation method outside with central pixel point Interpolation, then the interpolation formula of four direction be：
Wherein, central pixel point refers to satisfactionImage-region is w × s pixels.
On the above basis, it is iterated using formula (8), then Human Height H=max (z1,z2…zn), wherein n is The dimension of portrait profile set of pixels in monocular image.
A kind of Human Height Real-time Generation based on monocular image of the present invention is completed in summary, it should Method, which is directly handled monocular image, obtains Human Height, and without increasing aided-detection device at the scene, deployment is flexible, Cost can greatly be saved.Particularly, this method has the ability that batch calculates height, in magnanimity information comparison, has only Special Efficiency and accuracy advantage.

Claims (5)

1. a kind of Human Height Real-time Generation based on monocular image, it is characterised in that：It the described method comprises the following steps：
A. according to monocular image, computation model is built；
B. sliding-model control is carried out to computation model；
C. linearization process is carried out to system-computed model；
D., constraints is set and is iterated calculating, directly generates height result.
2. the Human Height Real-time Generation based on monocular image as described in claim 1, it is characterised in that：The step The specific implementation of A is：
Under the conditions of Lambert ideal images, by Lambert reflection equations it is found that the gradation of image of monocular image is put down with portrait Face height is in binary nonlinear relationship, proposes that the computation model of the method for the invention, process are as follows based on this：
Assuming that the in-plane of original monocular image is denoted as (Dx,Dy), point light source incident direction is denoted as (DLx,DLy), the ash of image Degree is denoted as g (x, y),
Then there is following formula according to Lambert reflection equations：
To reduce the uncertainty for building result of calculation caused by above-mentioned model under a variety of ideal conditions, following compensation is introduced Method：
Wherein, EnWhite Gaussian noise error is indicated, for compensating the error caused by noise；ErRoughness error is indicated, for mending Repay in-plane error；EgGradient similitude error is indicated, for compensating equation generated mistake during sliding-model control Difference；(Zx,Zy) indicate discrete picture in-plane；
Therefore, systematic error ErrIt may be defined as：
Err=En+μ·Er+θ·Eg
=∫ ∫ [(g (x, y)-F (Dx,Dy))2+μ·(Dx 2+Dy 2+DLx 2+DLy 2)+θ·((Zx-Dx)2+(Zy-Dy)2)]dxdy； (3)
3. the Human Height Real-time Generation based on monocular image as claimed in claim 2, it is characterised in that：The step The specific implementation of B is：
The process of sliding-model control is：
1. the minimum conditions of solving system error and gradient similitude error：
To formula (3) derivation, and solves derivative equation and can obtain：
2. can be obtained to formula (4) sliding-model control：
3. obtaining final discretization calculates formula：
Pull-type approximate discretization transformation factor is introduced,WhereinIndicate being averaged for in-plane L Direction, δ indicate monocular image pixel away from, formula (5) is converted with the transformation factor, can be obtained Human Height and The discretization of gradient calculates formula, as follows：
To sum up, the discretization for just having obtained height and gradient similarity calculates formula.
4. the Human Height Real-time Generation based on monocular image as claimed in claim 3, it is characterised in that：The step The specific implementation of C is：
The problem of annual reporting law can not restrain in special circumstances is conciliate to improve computational efficiency, system-computed model is linearized Processing：
Using Talor formula, using the result of calculation of laststate as reference data, the linearisation of discretization formula (6) can be obtained Approximate formula is as follows：
Substituting the above to (6) can obtain：
Wherein, σ indicates variable quantity of each variable relative to t state reference benchmark；
Based on the above, can be the variation for solving in-plane by the non-linear transfer that in-plane calculates's Then linear equation can calculate in-plane (D by following formulax,Dy) value：
5. the Human Height Real-time Generation based on monocular image as claimed in claim 4, it is characterised in that：The step The specific implementation of D is：
1. to obtain faster calculating speed and less EMS memory occupation, using staggered-mesh method：
Build staggered-mesh：The pixel spot size of monocular figure is w × s, in-plane (Dx,Dy) and gradation of image g (x, y) battle array Row dimension is also w × s；The pixel of human body contour outline and in-plane pixel are staggered 0.5 pixel, human body contour outline picture element matrix Dimension is (w+1) × (s+1)；In-plane is also (w+1) × (s+1) to the local derviation dimension of both direction；
To sum up, the Discrete Linear equation solved is as follows in the first-order partial derivative of its in-plane：
2. introducing the constraint of two classes, first, reference point constrains, second is that boundary condition constrains, method is as follows：
A. reference point constraint is the point of known in-plane and height to be added in camera lens visual angle, and ensure that the point can be in list It is shown on mesh image；
B. boundary condition constrains, that is, the in-plane of boundary pixel point is added, and uses interpolation method with central pixel point to extrapolated value, Then the interpolation formula of four direction is：
Wherein, central pixel point refers to satisfactionImage-region is w × s pixels,
On the above basis, it is iterated using formula (8), then Human Height H=max (z1,z2…zn), wherein n is monocular The dimension of portrait profile set of pixels in image.
CN201810356143.4A 2018-04-19 2018-04-19 A kind of Human Height Real-time Generation based on monocular image Withdrawn CN108596888A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810356143.4A CN108596888A (en) 2018-04-19 2018-04-19 A kind of Human Height Real-time Generation based on monocular image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810356143.4A CN108596888A (en) 2018-04-19 2018-04-19 A kind of Human Height Real-time Generation based on monocular image

Publications (1)

Publication Number Publication Date
CN108596888A true CN108596888A (en) 2018-09-28

Family

ID=63613254

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810356143.4A Withdrawn CN108596888A (en) 2018-04-19 2018-04-19 A kind of Human Height Real-time Generation based on monocular image

Country Status (1)

Country Link
CN (1) CN108596888A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111611928A (en) * 2020-05-22 2020-09-01 杭州智珺智能科技有限公司 Height and body size measuring method based on monocular vision and key point identification

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111611928A (en) * 2020-05-22 2020-09-01 杭州智珺智能科技有限公司 Height and body size measuring method based on monocular vision and key point identification
CN111611928B (en) * 2020-05-22 2023-07-28 郑泽宇 Height and body size measuring method based on monocular vision and key point identification

Similar Documents

Publication Publication Date Title
CN105096329B (en) Method for accurately correcting image distortion of ultra-wide-angle camera
CN111243033B (en) Method for optimizing external parameters of binocular camera
CN111784778B (en) Binocular camera external parameter calibration method and system based on linear solving and nonlinear optimization
CN107025670A (en) A kind of telecentricity camera calibration method
CN109474814A (en) Two-dimensional calibration method, projector and the calibration system of projector
CN107633533B (en) High-precision circular mark point center positioning method and device under large-distortion lens
CN110956661A (en) Method for calculating dynamic pose of visible light and infrared camera based on bidirectional homography matrix
CN111986106B (en) High-dynamic image reconstruction method based on neural network
CN103226819A (en) Segmental counting-based relative radiation correction method
WO2012122788A1 (en) Method for obtaining image correction coefficient, and method and system for correcting non-uniform image
CN110910456B (en) Three-dimensional camera dynamic calibration method based on Harris angular point mutual information matching
CN105869129B (en) For the thermal infrared images residue non-uniform noise minimizing technology after nonuniformity correction
CN114359405A (en) Calibration method of off-axis Samm 3D line laser camera
CN115266779B (en) Non-linear correction method for pixel response of flat panel detector for CT scanning
CN106296627B (en) A kind of heteropical method correcting infrared imaging system
CN108596888A (en) A kind of Human Height Real-time Generation based on monocular image
CN113409399B (en) Dual-camera combined calibration method, system and device
CN112465918B (en) Microscopic vision calibration method based on Tsai calibration
CN114184284B (en) Real-time scene correction method and system for longitudinal scanning thermal infrared imager
WO2022252697A1 (en) Camera image fusion method and camera image fusion system
CN105046674A (en) Nonuniformity correction method of multi-pixel parallel scanning infrared CCD images
CN109799731B (en) Multi-band optical radiation type semi-physical simulation method and system
CN113295701A (en) Method, device and medium for measuring damage of shear wall based on digital image
CN111076820B (en) Infrared real-time non-uniformity correction method
TW200931000A (en) Neural network-based two-dimensional luminance measurement method

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20180928