CN111383288A - Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm - Google Patents

Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm Download PDF

Info

Publication number
CN111383288A
CN111383288A CN202010135361.2A CN202010135361A CN111383288A CN 111383288 A CN111383288 A CN 111383288A CN 202010135361 A CN202010135361 A CN 202010135361A CN 111383288 A CN111383288 A CN 111383288A
Authority
CN
China
Prior art keywords
camera
algorithm
value
fitness value
population
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010135361.2A
Other languages
Chinese (zh)
Other versions
CN111383288B (en
Inventor
宋佳音
池志祥
宋文龙
朱庆林
张晓鹏
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.)
Northeast Forestry University
Original Assignee
Northeast Forestry University
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 Northeast Forestry University filed Critical Northeast Forestry University
Priority to CN202010135361.2A priority Critical patent/CN111383288B/en
Publication of CN111383288A publication Critical patent/CN111383288A/en
Application granted granted Critical
Publication of CN111383288B publication Critical patent/CN111383288B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Studio Devices (AREA)

Abstract

The invention provides a novel method for optimizing camera internal parameters, which mixes a goblet sea squirt algorithm with strong global optimization capability and a self-adaptive differential evolution with strong local search capability, provides a concept of average fitness value to select individual fitness value, optimizes parameters by using the goblet sea squirt algorithm when the individual fitness value is smaller than the average fitness value, and otherwise optimizes the parameters by using the self-adaptive differential evolution algorithm when the individual fitness value is larger than the average fitness value, thereby simplifying the calculation complexity, improving the search efficiency, applying variation in the later iteration stage and increasing the population diversity by using a cross factor.

Description

Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm
Technical Field
The invention relates to a camera internal reference optimization algorithm mixed with a goblet sea squirt algorithm and a self-adaptive differential evolution algorithm.
Background
In machine vision, in order to solve the relationship between the three-dimensional geometric position of the surface of the space object and the corresponding point of the surface in the image, a geometric model of camera imaging needs to be established, and the parameters of the model are the camera parameters. The process of solving the parameters is camera calibration, the parameter calibration is an extremely important link in machine vision, and the accuracy of the solved parameters and the stability of the algorithm determine the accuracy of later-stage target measurement to a great extent.
The traditional camera calibration algorithm comprises a Direct Linear Transformation (DLT) method, a Tasi two-step method, a Zhang Zhengyou calibration method and the like. DLT is easy to realize, the algorithm is simple, but the precision is not high; the Tasi two-step method has high precision, but has high requirements on equipment and is complex to implement.
With the development of naturally inspired swarm intelligent optimization algorithms, some improved optimization algorithms are widely applied to visual calibration and detection, and the creep process and the like provides a camera internal parameter optimization method based on the improved particle swarm optimization, wherein the reprojection error of the method is 0.0811, the precision is high, but the particle swarm optimization is unstable and is easy to fall into local optimum; the method has the advantages that the reprojection error is 0.1398, the optimization precision is high, the whole optimization process is complex, the calculated amount is large, the optimization time is increased, and the optimization efficiency is reduced.
Zhangyingyou calibration method: the Zhangzhengyou calibration method is also called a checkerboard calibration method, and is a single-plane checkerboard camera calibration method proposed by Zhangzhengyou professor 1998. The method is between the traditional calibration method and the self-calibration method, adopts the orthogonal condition of the rotation matrix and the nonlinear optimization in the calibration of the camera, is simple to use and strong in practicability, and has the following advantages: no additional equipment is needed, and a printed checkerboard is obtained. The calibration is simple, and the camera and the calibration plate can be placed at will. Corner points are obvious, but the accuracy is not sufficient.
The Salp Swarm Algorithm (SSA) is a meta-heuristic algorithm, and is proposed by Mirjalili et al based on the gathering behavior of the sea squirt group, so as to simulate the process of cruising and foraging of the sea squirt in the ocean, establish a chain mathematical model of the sea squirt chain, and divide the individual sea squirt group into a leader and a follower. The leader is at the front of the chain of goblet sea squirts, the rest are considered followers. The leader guides the population and the followers follow each other. The method has the advantages of few parameters, high robustness, simple structure, fast convergence and the like. Compared with some traditional optimization algorithms, the method has better global optimization effect, and although the goblet sea squirt algorithm has obvious global optimization effect, the method is easy to fall into local optimization in the aspect of local optimization problem.
Adaptive Differential Evolution (ADE): the differential evolution algorithm is a simple and powerful algorithm proposed by storm and Price in 1997. The algorithm has three main operators, namely mutation, intersection and selection, the self-adaptive differential evolution changes parameters SF and CR in the differential evolution algorithm, so that the optimization effect is improved, and the differential evolution algorithm introducing the self-adaptive parameters is called as the self-adaptive differential evolution algorithm.
Disclosure of Invention
In order to avoid the camera calibration calculation from being trapped in local optimization, the invention fuses the adaptive differential evolution algorithm with strong local search capability and the goblet and sea squirt algorithm to improve the precision and stability of the camera calibration.
The invention provides a novel method for optimizing camera internal parameters, which mixes a goblet sea squirt algorithm with strong global optimization capability and a self-adaptive differential evolution algorithm with strong local search capability, provides a concept of average fitness value to select individual fitness value, and optimizes parameters by using the goblet sea squirt algorithm when the individual fitness value is smaller than the average fitness value; and otherwise, when the individual fitness value is larger than the average fitness value, optimizing the parameters by using a self-adaptive differential evolution algorithm. The calculation complexity is simplified, and the search efficiency is improved. And (3) applying variation and cross factors at the later stage of iteration to increase the population diversity. The method comprises the following steps:
1. and (3) reading pictures, namely shooting pictures of the calibration plate by using photographic equipment, wherein the used calibration template is checkerboards with black and white alternated, the number of the checkerboards is 10 × 8, the size of each checkerboard is 50mm × 50mm, 15 calibration pictures with different angles are selected, and the calibration pictures are read by using an immead function in MATLAB.
2. Graying treatment: and (4) applying an rgb2gray function in MATLAB to perform picture graying treatment.
3. Angular point extraction: and (3) extracting the angular points of the gray-scale pictures by using an MATLAB tool box, wherein the pixel coordinates of all the angular points can be extracted, and each picture has 63 angular points which are 945 in total.
4. Population initialization:
① calculation of camera parameters
Representing a point (x) in the world coordinate system in the linear model according to the Zhang-friend scaling methodw,yw,zw) Conversion to a point (x) in the pixel coordinate systemd,yd) The function relation of (a), namely the ideal camera imaging relation, is as follows:
Figure BDA0002397093420000031
wherein R is a rotation matrix of 3 × 3, T is a translation matrix of 3 × 1, MR，TIs a combination of a rotation matrix and a translation matrix; mAFor internal parameters, representing the internal geometric characteristics of the camera, the mathematical model is:
Figure BDA0002397093420000032
in the formula (f)x＝fc1/dx,fy＝fc2/dy，fc1、fc2Is the camera focal length; dxAnd dyIs the physical length of the pixel, u0And voIs the pixel coordinate value of the intersection of the camera optical axis and the image plane.
Calculating the internal parameter f of the ideal camera according to the formulas (1) and (2)x、fy、u0、v0。
② distortion coefficient calculation
The actual camera lens has distortion and has a nonlinear imaging relation, and a point (x) in an image coordinate system of an ideal camera model is established by considering the radial distortion and the tangential distortion of the lensd,yd) And a point (x) in the actual non-linear camera modelu,yu) A mathematical model of distortion of (a).
The mathematical model of radial distortion is:
Figure BDA0002397093420000041
the tangential distortion mathematical model is:
Figure BDA0002397093420000042
wherein r is2＝xu 2+yu 2(8)
The joint type (6), (7) and (8) are obtained, and the functional relation between the image coordinates of the ideal camera model and the image coordinates in the actual camera model is as follows:
Figure BDA0002397093420000043
in the formula (x)d,yd) Image coordinates of an ideal camera model, (x)u,yu) Actual image coordinates in the nonlinear camera model taking lens distortion into account; k is a radical of1，k2，k3As radial distortion coefficient, p1，p2Is a tangential distortion coefficient; the initial value of the distortion coefficient is obtained by equation (3).
③ population initialization
The initial parameter values include camera parameters fx、fy、u0、v0Coefficient of radial distortion k1、k2、k3And tangential distortion coefficient p1、p2The unit of the parameter is a pixel. With the calculated camera intrinsic parameter fx、fy、u0、v0The initial value of (1) is the value of (8), and the initialization is performed. Coefficient of radial distortion k1、k2、k3And tangential distortion coefficient p1、p2Initialization is performed with an initial value of 0 and a variance of 1, thereby completing population initialization. The variance is determined from empirical values.
5. Boundary detection:
each dimension is initialized at population initialization according to the boundary requirements in the cask sea squirt algorithm. And (3) taking the maximum value of each dimension as the upper bound of the current dimension, taking the minimum value as the lower bound of the current dimension, representing the upper bound of the search space as ubi, and representing the lower bound of the search space as lbi, wherein i is more than or equal to 1 and less than or equal to 9, removing the value of the jumping-out boundary in each population iteration, and selecting the numerical value in the boundary.
6. Calculating individual fitness value f of goblet sea squirt population
Let the fitness function be:
Figure BDA0002397093420000051
wherein i is the image number, j is the calibration point number, pijIs the actual pixel coordinate of the jth calibration point in the ith image, N is the total number of images 15, M is the total number of calibration points 945, p (f)x,fy,u0,v0,k1,k2,k3,p1,p2,Pj) Calculating the obtained pixel coordinates; f. ofx、fy、u0、v0For the in-camera parameters to be optimized, k1、k2、k3、p1、p2For the distortion coefficient to be optimized, PjThe coordinates of the jth point in the ith image in the world coordinate system are obtained; the fitness function is used to calculate individual fitness values in each population iteration.
7. Calculating mean fitness value of goblet sea squirt population
Figure BDA0002397093420000052
Since there are 30 individuals in each generation, after the fitness of 30 individuals is obtained, the average value is calculated:
Figure BDA0002397093420000053
note: the number of iterations and the number of individuals are determined based on empirical values. If the number of individuals is larger, the population is more diverse, the test effect is better, but the optimization time is long; if the setting is smaller, the optimal solution may not be obtained; therefore 30 individuals were selected based on empirical values.
8. Finding an optimal fitness value
Judging individual fitness value fi and population average fitness value
Figure BDA0002397093420000054
The size of (d); when in use
Figure BDA0002397093420000055
Then, optimizing by adopting a self-adaptive differential evolution algorithm; when in use
Figure BDA0002397093420000056
In time, optimization is carried out by adopting a goblet sea squirt algorithm; and after optimizing, updating the optimal position, and judging according to the optimal position.
(1) When in use
Figure BDA0002397093420000057
And then, solving by using a self-adaptive differential evolution algorithm:
① update control parameters SF and CR
Figure BDA0002397093420000058
Figure BDA0002397093420000059
In the formula, rand1、rand2、rand3、rand4Are all [0,1]Random number of between, τ1And τ2Representing transition probability, SFlAnd SFuA boundary scaling factor; let τ be based on empirical values1＝τ2When the value of SF is 0.1, the initial value of SF is 0.5, and the initial value of CR is 0.9.
② mutation individuals are generated by mutation operations, i.e. mutation operators.
The mutation operator is represented as:
Figure BDA0002397093420000061
in the formula (I), the compound is shown in the specification,
Figure BDA0002397093420000062
is the mutant individual in the t +1 th iteration;
Figure BDA00023970934200000612
each representing three different individuals in the population, and r1≠r2≠r3SF denotes a scaling factor and is constant.
③ in the crossing process, in order to improve the diversity of the population, the existing individuals are selected
Figure BDA0002397093420000063
Or a mutant individual
Figure BDA0002397093420000064
Selecting test individuals
Figure BDA0002397093420000065
I.e. the crossover operator.
The crossover operator is represented as:
Figure BDA0002397093420000066
wherein rand is a random number between [0,1 ]; CR represents the crossover probability and is a constant.
④ in the selection process, the test individuals are compared
Figure BDA0002397093420000067
And the current individual
Figure BDA0002397093420000068
Is to obtain the individuals of the t +1 th generation, i.e. the selection operator.
The selection operator is represented as:
Figure BDA0002397093420000069
and updating the optimal position and judging.
(2) When in use
Figure BDA00023970934200000610
And then, the calculation process by using a goblet sea squirt algorithm is as follows:
① calculating parameter c1：
c1Plays a role in global balance and local development and is the most important parameter.
Figure BDA00023970934200000611
In the formula: t is the current iteration number, T ═ TmaxIs the maximum number of iterations.
② update leader position of goblet sea squirt
Figure BDA0002397093420000071
In the formula (I), the compound is shown in the specification,
Figure BDA0002397093420000072
for the 1 st leader position in the j-dimensional space, c2 and c3 are both [0,1]]Random numbers in between, which determine the direction of movement of the next location, enhance
Figure BDA0002397093420000073
The randomness of (2) increases the individual diversity. ubj、lbjThe number of parameters to be optimized is 9 respectively at the upper bound and the lower bound of the jth dimension search space, the dimension is equal to 9, and the value of j is more than or equal to 1 and less than or equal to 9.
③ update tracing position of ascidian
Figure BDA0002397093420000074
When i is more than or equal to 2,
Figure BDA0002397093420000075
denotes the location of the ith follower in dimension j, t is time, v0Is the initial velocity.
And updating the optimal position and judging.
9. Judging whether a preset maximum iteration number or an expected fitness value is reached; if yes, outputting an optimal value of the camera internal parameter and the distortion coefficient; if not, go back to step 6 to continue the calculation.
The photographic equipment can be any digital camera or electronic equipment with a camera shooting function.
The flow of adopting the goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm is shown in fig. 1.
Drawings
FIG. 1 algorithm flow chart
FIG. 2 is a re-projection error diagram of the Zhang Yong calibration method
FIG. 3 goblet sea squirt algorithm reprojection error map
FIG. 4 goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm reprojection error map
Detailed Description
A group of pictures are taken by using a digital camera device, and camera internal parameters are optimized and calculated by using a Zhang Yongyou calibration method, a goblet ascidian algorithm and the goblet ascidian-adaptive differential evolution hybrid algorithm. And comparing and analyzing the feasibility, stability and calibration precision of the algorithm.
The camera used in the experiment is a glorious 10 mobile phone self-contained camera, the effective pixel of a shot picture is 4608pixel × 3456pixel, the experiment is carried out on a windows10 system platform, the processing software is Matlab2017, a goblet sea squirt population N is set to be 30, the iteration total number is 300, the experiment uses a standard checkerboard with 50mm × 50mm of each grid as a calibration board, the number of the checkerboard X directions is 10, the number of the Y directions is 8, 15 calibration pictures with different angles are selected for verification, the camera internal parameters are optimized by a Zhang friend calibration method and a goblet sea squirt algorithm respectively, the camera internal parameters and distortion coefficients are obtained after 300 times of iteration, and the reprojection error is calculated.
And (3) analyzing an experimental result:
the camera internal reference calibration results of Zhangzhen friend calibration method, Zun ascidian algorithm (SSA) and Zun ascidian-adaptive differential evolution hybrid camera internal reference optimization algorithm (SSA-ADE) are listed in Table 1. In order to verify the calibration accuracy under different methods, the reprojection error is calculated, fig. 2 is a reprojection error graph obtained by a Zhang Zhengyou calibration method, fig. 3 is a reprojection error graph obtained by a goblet sea squirt algorithm, and fig. 4 is a reprojection error graph obtained by a goblet sea squirt-adaptive differential evolution hybrid camera internal parameter optimization algorithm. Table 2 shows the average error of the three algorithms, the calculated average reprojection error of the zhangying scaling method is 0.259363, the average reprojection error under the goblet ascidian algorithm is 0.157219, and the average reprojection error under the goblet ascidian-adaptive differential evolution hybrid camera internal reference optimization algorithm is 0.064030. Compared with a reprojection error map, the SSA-ADE effect of the hybrid algorithm is better; compared with the Zhang Zhengyou calibration method, the accuracy of the hybrid algorithm is improved by 75.313% compared with the Zhang Zhengyou calibration method, and is improved by 35.930% compared with the Zun Haisha algorithm, so that the algorithm of the invention has feasibility and is more accurate in calibration accuracy.
TABLE 1 calibration results of camera internal parameters
Figure BDA0002397093420000081
Figure BDA0002397093420000091
TABLE 2 average error comparison
Figure BDA0002397093420000092
The invention provides a novel method for optimizing camera internal parameters, which mixes a goblet sea squirt algorithm with strong global optimization capability and a self-adaptive differential evolution with strong local search capability, provides a concept of average fitness value to select individual fitness value, optimizes parameters by using the goblet sea squirt algorithm when the individual fitness value is smaller than the average fitness value, and otherwise optimizes the parameters by using the self-adaptive differential evolution algorithm when the individual fitness value is larger than the average fitness value, thereby simplifying the calculation complexity, improving the search efficiency, applying variation in the later iteration stage and increasing the population diversity by using a cross factor. The reprojection error comparison is carried out by using Zhangfriend calibration method, a goblet ascidian algorithm and the goblet ascidian-self-adaptive differential evolution hybrid camera internal reference optimization algorithm, and the algorithm has good feasibility; in the aspect of calibration precision, the calibration precision is improved by 75.313% compared with that of a Zhang Zhengyou calibration method, the calibration precision is improved by 35.930% compared with that of a goblet ascidian algorithm, and experimental results show that the goblet ascidian-adaptive differential evolution hybrid camera internal reference optimization algorithm has good feasibility and accuracy for camera internal reference optimization.

Claims (5)

1. A goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm is characterized in that: mixing a goblet sea squirt algorithm with strong global optimization capability and a self-adaptive differential evolution algorithm with strong local search capability, providing a concept of average fitness value to select the individual fitness value, and optimizing parameters by using the goblet sea squirt algorithm when the individual fitness value is smaller than the average fitness value; otherwise, when the individual fitness value is larger than the average fitness value, the parameters are optimized by using a self-adaptive differential evolution algorithm; variation and cross factors are applied in the later iteration stage to increase the diversity of the population; the method comprises the following steps:
1) taking a picture by using photographic equipment, wherein the used calibration template is checkerboard with black and white alternated, the number of the checkerboard is 10 × 8, the size of each checkerboard is 50mm × 50mm, 15 calibration pictures with different angles are selected, and the calibration pictures are read by using an immead function in MATLAB;
2) graying treatment: performing image graying by applying an rgb2gray function in MATLAB;
3) angular point extraction: performing corner extraction on the gray-scale picture by using an MATLAB tool box, wherein pixel coordinates of all corners can be extracted, and each picture has 63 corners and 945 corners in total;
4) population initialization:
① calculation of camera parameters
Representing a point (x) in the world coordinate system in the linear model according to the Zhang-friend scaling methodw,yw,zw) Conversion to a point (x) in the pixel coordinate systemd,yd) Is composed ofThe numerical relation, i.e. the ideal camera imaging relation, is:
Figure FDA0002397093410000011
wherein R is a rotation matrix of 3 × 3, T is a translation matrix of 3 × 1, MR，TIs a combination of a rotation matrix and a translation matrix; mAFor internal parameters, representing the internal geometric characteristics of the camera, the mathematical model is:
Figure FDA0002397093410000012
in the formula (f)x＝fc1/dx,fy＝fc2/dy，fc1、fc2Is the camera focal length; dxAnd dyIs the physical length of the pixel, u0And voIs the pixel coordinate value of the intersection point of the camera optical axis and the image plane;
calculating the internal parameter f of the ideal camera according to the formulas (1) and (2)x、fy、u0、v0；
② distortion coefficient calculation
The actual camera lens has distortion and has a nonlinear imaging relation; considering the radial distortion and the tangential distortion of a lens, establishing a point (x) in an image coordinate system of an ideal camera modeld,yd) With a point (x) in the actual camera modelu,yu) The function relation between the image coordinates of the ideal camera model and the image coordinates in the actual camera model is obtained by the mathematical model of (1):
Figure FDA0002397093410000021
in the formula (x)d,yd) Image coordinates of an ideal camera model, (x)u,yu) Actual image coordinates in the actual non-linear camera model taking lens distortion into account; k is a radical of1、k2、k3Is a system of radial distortionNumber, p1、p2Is a tangential distortion coefficient; obtaining an initial value of a distortion coefficient through an equation (3);
③ population initialization
The initial parameter values include camera parameters fx、fy、u0、v0Coefficient of radial distortion k1、k2、k3And tangential distortion coefficient p1、p2The unit of the parameter is a pixel; with the calculated camera intrinsic parameter fx、fy、u0、v0Initializing the value of (1) as an initial value and the variance of (8); coefficient of radial distortion k1、k2、k3And tangential distortion coefficient p1、p2Initializing with an initial value of 0 and a variance of 1, and further completing population initialization;
5) boundary detection:
according to the boundary requirement in the goblet sea squirt algorithm, when a population is initialized, initializing each dimension, wherein the maximum value of each dimension is the upper bound of the current dimension, the minimum value is the lower bound of the current dimension, the upper bound of a search space is represented as ubi, the lower bound of the search space is represented as lbi, wherein i is more than or equal to 1 and less than or equal to 9, in each population iteration, the value of jumping out of the boundary is removed, and the numerical value in the boundary is selected;
6) calculating individual fitness value f of goblet sea squirt population
Let the fitness function be:
Figure FDA0002397093410000031
wherein i is the image number, j is the calibration point number, pijIs the actual pixel coordinate of the jth calibration point in the ith image, N is the total number of images 15, M is the total number of calibration points 945, p (f)x,fy,u0,v0,k1,k2,k3,p1,p2,Pj) Calculating the obtained pixel coordinates; f. ofx、fy、u0、v0For the in-camera parameters to be optimized, k1、k2、k3、p1、p2For the distortion coefficient to be optimized, PjThe coordinates of the jth point in the ith image in the world coordinate system are obtained; in each population iteration, calculating an individual fitness value by using the fitness function;
7) calculating mean fitness value of goblet sea squirt population
Figure FDA0002397093410000032
Since there are 30 individuals in each generation, after the fitness of 30 individuals is obtained, the average value is calculated:
Figure FDA0002397093410000033
8) finding an optimal fitness value
Judging individual fitness value fi and population average fitness value
Figure FDA0002397093410000034
The size of (d); when in use
Figure FDA0002397093410000035
Then, optimizing by adopting a self-adaptive differential evolution algorithm; when in use
Figure FDA0002397093410000036
In time, optimization is carried out by adopting a goblet sea squirt algorithm; after optimizing, updating the optimal position, and judging according to the optimal position;
9) judging whether a preset maximum iteration number or an expected fitness value is reached; if yes, outputting an optimal value of the camera internal parameter and the distortion coefficient; if not, returning to the step 6) to continue the calculation.
2. The ascidian-adaptive differential evolution hybrid camera context optimization algorithm of claim 1, wherein: establishing one point (x) in the image coordinate system of the ideal camera model in the step 4)d,yd) And a point (x) in the actual non-linear camera modelu,yu) The mathematical model of (a) for the distortion,
the mathematical model of radial distortion is:
Figure FDA0002397093410000037
the tangential distortion mathematical model is:
Figure FDA0002397093410000041
wherein r is2＝xu 2+yu 2(8)
The relational expressions between the image coordinates of the ideal camera model and the image coordinates of the actual camera model are expressed by the following equations (3), which are obtained by the joint models (6), (7) and (8).
3. The ascidian-adaptive differential evolution hybrid camera context optimization algorithm of claim 1, wherein: in step 8) when
Figure FDA0002397093410000042
And then, solving by using a self-adaptive differential evolution algorithm:
① update control parameters SF and CR
Figure FDA0002397093410000043
Figure FDA0002397093410000044
In the formula, rand1、rand2、rand3、rand4Are all [0,1]Random number of between, τ1And τ2Representing transition probability, SFlAnd SFuA boundary scaling factor; let τ be1＝τ20.1, the initial value of SF is set to 0.5, and the initial value of CR is 0.9;
② generating mutation individuals through mutation operation, namely mutation operators;
the mutation operator is represented as:
Figure FDA0002397093410000045
in the formula:
Figure FDA0002397093410000046
is the mutant individual in the t +1 th iteration; x is the number oft r1、xt r2、xt r3Each representing three different individuals in the population, and r1≠r2≠r3SF denotes a scaling factor, which is constant;
③ in the crossing process, in order to improve the diversity of the population, the existing individuals are selected
Figure FDA0002397093410000047
Or a mutant individual
Figure FDA0002397093410000048
Selecting test individuals
Figure FDA0002397093410000049
Namely, a crossover operator;
the crossover operator is represented as:
Figure FDA00023970934100000410
wherein rand is a random number between [0,1 ]; CR represents the cross probability, which is constant;
④ in the selection process, the test individuals are compared
Figure FDA0002397093410000051
And the current individual
Figure FDA0002397093410000052
Is to makeObtaining individuals of the t +1 th generation, namely selecting an operator;
the selection operator is represented as:
Figure FDA0002397093410000053
and updating the optimal position and judging.
4. The ascidian-adaptive differential evolution hybrid camera context optimization algorithm of claim 1, wherein: in step 8) when
Figure FDA0002397093410000054
And then, the calculation process by using a goblet sea squirt algorithm is as follows:
① calculating parameter c1：
c1Plays the roles of global balance and local development and is the most important parameter;
Figure FDA0002397093410000055
in the formula: t is the current iteration number, T ═ TmaxIs the maximum iteration number;
② update leader position of goblet sea squirt
Figure FDA0002397093410000056
In the formula (I), the compound is shown in the specification,
Figure FDA0002397093410000057
for the 1 st leader position in the j-dimensional space, c2 and c3 are both [0,1]]Random numbers in between, which determine the direction of movement of the next location, enhance
Figure FDA0002397093410000058
The randomness of the method increases the diversity of individuals; ubj、lbjRespectively j-dimension search spaceThe upper and lower bounds of (1) have 9 parameters to be optimized, the dimensionality is 9, and the value of j is more than or equal to 1 and less than or equal to 9;
③ update tracing position of ascidian
Figure FDA0002397093410000059
When i is more than or equal to 2,
Figure FDA00023970934100000510
denotes the location of the ith follower in dimension j, t is time, v0Is the initial speed;
and updating the optimal position and judging.
5. The camera device of claim 1 may be any digital camera or electronic device with camera function.
CN202010135361.2A 2020-03-02 2020-03-02 Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm Active CN111383288B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010135361.2A CN111383288B (en) 2020-03-02 2020-03-02 Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010135361.2A CN111383288B (en) 2020-03-02 2020-03-02 Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm

Publications (2)

Publication Number Publication Date
CN111383288A true CN111383288A (en) 2020-07-07
CN111383288B CN111383288B (en) 2022-11-11

Family

ID=71219731

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010135361.2A Active CN111383288B (en) 2020-03-02 2020-03-02 Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm

Country Status (1)

Country Link
CN (1) CN111383288B (en)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100303349A1 (en) * 2009-05-28 2010-12-02 Bechtel Jon H Digital image processing and systems incorporating the same
CN104059936A (en) * 2014-04-30 2014-09-24 唐星 Preparation method of genetically engineered bacterium for synthesizing glutathione and product thereof
CN108594654A (en) * 2018-03-21 2018-09-28 南京邮电大学 A kind of overhead crane anti-swing control method based on two type fuzzy sliding modes
CN108984917A (en) * 2018-07-20 2018-12-11 北京航空航天大学 Large aircraft flies control actuating system intelligent design and evaluation method
CN109284860A (en) * 2018-08-28 2019-01-29 温州大学 A kind of prediction technique based on orthogonal reversed cup ascidian optimization algorithm
CN109448060A (en) * 2018-09-04 2019-03-08 惠州市德赛西威智能交通技术研究院有限公司 A kind of camera calibration parameter optimization method based on bat algorithm
CN109472320A (en) * 2018-11-30 2019-03-15 南京农业大学 Crop growing stage model Cultivar parameter automatically corrects frame under condition of uncertainty
CN109657869A (en) * 2018-12-26 2019-04-19 国网内蒙古东部电力有限公司经济技术研究院 High latitude severe cold area substation LCC prediction model based on SSA-LSSVM
CN109873810A (en) * 2019-01-14 2019-06-11 湖北工业大学 A kind of phishing detectin method based on cup ascidian group's algorithm support vector machines

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100303349A1 (en) * 2009-05-28 2010-12-02 Bechtel Jon H Digital image processing and systems incorporating the same
CN104059936A (en) * 2014-04-30 2014-09-24 唐星 Preparation method of genetically engineered bacterium for synthesizing glutathione and product thereof
CN108594654A (en) * 2018-03-21 2018-09-28 南京邮电大学 A kind of overhead crane anti-swing control method based on two type fuzzy sliding modes
CN108984917A (en) * 2018-07-20 2018-12-11 北京航空航天大学 Large aircraft flies control actuating system intelligent design and evaluation method
CN109284860A (en) * 2018-08-28 2019-01-29 温州大学 A kind of prediction technique based on orthogonal reversed cup ascidian optimization algorithm
CN109448060A (en) * 2018-09-04 2019-03-08 惠州市德赛西威智能交通技术研究院有限公司 A kind of camera calibration parameter optimization method based on bat algorithm
CN109472320A (en) * 2018-11-30 2019-03-15 南京农业大学 Crop growing stage model Cultivar parameter automatically corrects frame under condition of uncertainty
CN109657869A (en) * 2018-12-26 2019-04-19 国网内蒙古东部电力有限公司经济技术研究院 High latitude severe cold area substation LCC prediction model based on SSA-LSSVM
CN109873810A (en) * 2019-01-14 2019-06-11 湖北工业大学 A kind of phishing detectin method based on cup ascidian group's algorithm support vector machines

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
GHISLAIN PICARD等: "Development and calibration of an automatic spectral albedometer to estimate near-surface snow SSA time series", 《TC》 *
张德成等: "MHCSSA:一种基于多群分层协作策略的社会蜘蛛算法", 《长春师范大学学报》 *
李强等: "基于参数自适应差分进化算法的Web服务组合", 《计算机技术与发展》 *
王梦秋 等: "基于改进樽海鞘群算法的PMSM多参数辨识", 《系统仿真学报》 *

Also Published As

Publication number Publication date
CN111383288B (en) 2022-11-11

Similar Documents

Publication Publication Date Title
JP6011102B2 (en) Object posture estimation method
CN109934847B (en) Method and device for estimating posture of weak texture three-dimensional object
CN110634161A (en) Method and device for quickly and accurately estimating pose of workpiece based on point cloud data
CN109325995B (en) Low-resolution multi-view hand reconstruction method based on hand parameter model
JPWO2005038716A1 (en) Image collation system and image collation method
CN112329726B (en) Face recognition method and device
WO2023116430A1 (en) Video and city information model three-dimensional scene fusion method and system, and storage medium
CN112016454A (en) Face alignment detection method
CN112488067A (en) Face pose estimation method and device, electronic equipment and storage medium
CN109816634A (en) Detection method, model training method, device and equipment
CN113593043B (en) Point cloud three-dimensional reconstruction method and system based on generation countermeasure network
US7486815B2 (en) Method and apparatus for scene learning and three-dimensional tracking using stereo video cameras
CN111260706B (en) Dense depth map calculation method based on monocular camera
CN111383288B (en) Goblet ascidian-adaptive differential evolution hybrid camera internal parameter optimization algorithm
JP2019185295A (en) Image processing device and program for generating two-dimensional image
CN116862984A (en) Space pose estimation method of camera
CN112950787B (en) Target object three-dimensional point cloud generation method based on image sequence
CN113269831B (en) Visual repositioning method, system and device based on scene coordinate regression network
CN114821228A (en) Depth image output model training method, depth image obtaining method and device
CN112906432A (en) Error detection and correction method applied to human face key point positioning task
CN117132501B (en) Human body point cloud cavity repairing method and system based on depth camera
CN113496521B (en) Method and device for generating depth image and camera external parameter by using multiple color pictures
CN114821203B (en) Fine-grained image model training and identifying method and device based on consistency loss
Xing et al. Research on curved Chinese document correction based on deep neural network
JP2024032055A (en) Method for recognizing position and orientation of object, and computer program

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant