阅读说明：本技术 一种低粒子密度的单相机粒子图像测速方法 (Single-camera particle image velocimetry method with low particle density ) 是由 单良 施飞杨 李浩然 孔明 熊俊哲 刘维 洪波 郭天太 赵军 于 2021-06-07 设计创作，主要内容包括：本发明提供一种低粒子密度的单相机粒子图像测速方法,包括以下步骤：在流场中均匀地撒入示踪粒子,白色光源发出一束准直的稳定白色光束,光束经过滤波片后得到彩色片状光射入流场,照亮流场中的示踪粒子,CCD摄像机拍摄得到低粒子密度的彩色粒子图片并传送给上位机,上位机对低粒子密度的彩色粒子图片进行滤波处理和数据降采样,对处理后的低粒子密度的彩色粒子图片进行分析重建得到高粒子密度的三维粒子分布概率场后进行互相关计算得到流体三维速度场。本发明解决了现有技术中三维互相关算法应用在低粒子密度条件时存在的问题。(The invention provides a single-camera particle image velocimetry method with low particle density, which comprises the following steps: uniformly scattering tracer particles in a flow field, enabling a white light source to emit a collimated stable white light beam, enabling the light beam to pass through a filter to obtain colored sheet light to be emitted into the flow field, illuminating the tracer particles in the flow field, shooting by a CCD (charge coupled device) camera to obtain a colored particle picture with low particle density and transmitting the colored particle picture to an upper computer, carrying out filtering processing and data reduction sampling on the colored particle picture with low particle density by the upper computer, analyzing and reconstructing the processed colored particle picture with low particle density to obtain a three-dimensional particle distribution probability field with high particle density, and then carrying out cross-correlation calculation to obtain a fluid three-dimensional velocity field. The invention solves the problem that the three-dimensional cross-correlation algorithm in the prior art is applied to the condition of low particle density.)

1. A single-camera particle image velocimetry method with low particle density is characterized in that: the method comprises the following steps: uniformly scattering tracer particles in a flow field, enabling a white light source to emit a collimated stable white light beam, enabling the light beam to pass through a filter to obtain colored sheet light to be emitted into the flow field, illuminating the tracer particles in the flow field, shooting by a CCD (charge coupled device) camera to obtain a colored particle picture with low particle density and transmitting the colored particle picture to an upper computer, carrying out filtering processing and data reduction sampling on the colored particle picture with low particle density by the upper computer, analyzing and reconstructing the processed colored particle picture with low particle density to obtain a three-dimensional particle distribution probability field with high particle density, and then carrying out cross-correlation calculation to obtain a fluid three-dimensional velocity field.

2. The method according to claim 1, wherein said method comprises: the depth of particles in the flow field is calibrated through visible light with different wavelengths, and three-dimensional information of the flow field can be obtained by combining two-dimensional pixel position information in a picture obtained by shooting of a CCD camera.

3. The method according to claim 1, wherein said method comprises: modeling the existence of particles at a certain point in a flow field as existence probability, establishing a corresponding image forming model according to the imaging process of an optical system to obtain the minimization problem of a linear system, and solving the minimization problem by using an ADMM algorithm; and obtaining a three-dimensional particle distribution probability field with high particle density through a color particle image with low particle density.

4. A method for single camera particle image velocimetry as claimed in claim 1 or 2 or 3, characterized in that: and carrying out three-dimensional cross-correlation analysis on the obtained three-dimensional particle distribution probability field with high particle density to further obtain a fluid three-dimensional velocity field.

5. A method according to claim 3, wherein said method comprises:

the wavelength lambda and the depth Z have a linear relation, and the position of a light plane of the particle in a three-dimensional space is determined through the wavelength lambda; representing the three-dimensional coordinates of the particle in association with the two-dimensional coordinates (x, y), (x, y, λ) of the particle in the light plane obtained by the color image; modeling the probability of particle presence at a point in three-dimensional space as P (X, λ); because incoherent light is used, the imaging process of the optical system can be modeled as a set of Point Spread Functions (PSFs), which are the impulse responses of the focusing optical system and the spatial domain representation of the imaging system transfer function; the PSF is divided into three color channels of red, green and blue, and corresponding PSF point spread functions are respectively established: g_C(x, λ), C ∈ { red, green, blue }; the image formation model can thus be modeled as:

i_C(X)＝∫_Λ∫_Xg_C(X-X′，λ)·i_r(X，λ)·P(X，λ)dX′dλ

in the formula i_C(X) color channel, i, corresponding to the captured RGB image_r(X, λ) is the corresponding spectral distribution incident on the image sensor, g_C(X-X ', lambda) is a PSF function corresponding to a certain point in a three-dimensional space, the probability that particles exist at the corresponding point in the P (X, lambda) three-dimensional space, and dX' is the wavelength corresponding to the light plane at the two-dimensional position d lambda on the light plane of the current point; the spatial integral corresponds to a convolution representing a potential imperfect focus; the spectral image encoding the 3D particle position (spectral distribution obtained on the image sensor) can be converted into an RGB image by integration over wavelength;

after discretization, the convolution of the PSF and the reflected light intensity can be formulated as a matrix A epsilon R^3N×NLWhere N is the number of pixels of the low particle density color particle image, L is the number of discretized stages along the wavelength coordinate direction, and the number 3 represents the three different color image channels; i.e. i_t∈R^3NRepresenting the particle image captured by the camera at time t; p_t∈[0，1]^NLRepresenting the probability of the particle being present at different points at time point t; the particle field distribution of each time point can be obtained by solving the linear system; AP (Access Point)_t＝i_tBut full spectral information encoding the position of the particleInformation exists in three different color channels, which causes the linear system solution to be a ill-conditioned inverse problem; therefore, some prior knowledge of particle distribution needs to be introduced as a regular term, and the inverse problem becomes the following minimization problem:

in the formula: (p)^*) Set of solutions, AP, of the size of the probability that a particle exists at each point in three-dimensional space, obtained for solving the minimization problem_tThe simulated particle picture is obtained according to the image forming model; i.e. i_tA low particle density color particle picture collected at time t; a [ p ]₁|...|p_T]A group of simulated particle pictures are obtained according to an image forming model when a group of shot pictures are analyzed; p is a radical of_tIs the particle existence probability at time t; k is a radical of₁To optimize the parameters;

operator II_[0，1]Projecting all volume occupancy probabilities to an effective probability 0, 1]^NLThe above step (1); the first row in the formula is the least squares data fit term to solve the linear system inverse problem; the first term defines a weighting term L₁The sparse distribution of particles in the flow field to be measured is encouraged, and because the sensitivity of the camera to light with different wavelengths is different, the weighting coefficient is iteratively changed according to the previous result through further weighting by the diagonal matrix diag (w), is fixed and unchanged in the iterative process but can be changed along with the change of the particle depth, so that the sparsity can be effectively enhanced; when wavelengths in the yellow or blue-green part of the spectrum cause strong responses in multiple color channels, light of wavelengths further away from the blue or red part triggers only one channel, which results in an uneven particle distribution; different particles can be placed at the depth where the particles are more likely to be located through the setting of the weighting terms; the weighting term eliminates this bias by compensating for the non-uniformity of the luminance; the second term of the formula controls the occupation probability of the indicator function to be 0, 1]To (c) to (d);

due to the weighting term L in the optimization problem described above₁Sum index function II_[0，1](p₁；...；p_t) The optimization problem is non-smooth; adopting an alternative direction multiplier method ADMM to solve the non-smooth problem; the alternating direction multiplier method ADMM decouples a larger global problem which is difficult to solve into a plurality of smaller local problems which are easy to obtain effective solutions in a decomposition coordination program mode, and obtains a final solution of the global problem by iteratively solving a plurality of subproblems.

6. The method according to claim 5, wherein said method comprises:

after the optimization problem is solved by using an ADMM algorithm, a probability field of particle distribution is obtained, and then a previous three-dimensional particle distribution field is divided into a plurality of search areas with fixed sizes according to the set size of a search window; and the searching area with the same position in the next particle field is scanned from the upper left corner to the lower right corner line by line in an inquiry window larger than the searching window according to a formula Performing 3D cross-correlation to obtain a matrix of correlation coefficients, wherein the coordinates of the peak values of the correlation coefficients can represent the distance of particle displacement; substituting different moments of two data sets taken into a formula A rough three-dimensional velocity field of the flow field can be obtained; and then removing gross errors according to the basic principle of hydrodynamics, and correcting wrong velocity vectors by interpolation fitting to obtain an accurate three-dimensional particle velocity field.

7. The method according to claim 1, wherein said method comprises: the filter is a linear variable band-pass filter.

8. The method according to claim 1, wherein said method comprises: the volume ratio of the particle solution to the fluid solution is less than 100.

Technical Field

The invention relates to the technical field of measurement, in particular to a single-camera particle image velocity measurement method with low particle density.

Background

Fluid motion is very common in nature, hydrodynamics involved in the design of airplanes and underwater vehicles is closely related to fluid measurement, so that the specific motion condition of fluid is a precondition for deep research of hydrodynamics, and the measurement and analysis of the motion state of fluid are significant and are key problems in the field of fluid. The flow field measurement technology appears in 1904 at the earliest, prandtl invents a hand-operated water tank to observe flow fields around different models, uses a flow field display technology to qualitatively analyze and describe the flow fields, and researches the motion state of the flow fields. With the development of scientific technology, the flow field speed can be quantitatively measured by different measuring means such as temperature, laser, ultrasonic waves and the like.

In a traditional flow field measurement mode, a hot wire/hot film velocimeter (HWFA) can calculate the speed of a flow field but has certain influence on the movement of the flow field by utilizing a heat balance principle according to the difference of heat values taken away from a constant-temperature thermosensitive element by flow of the flow field with different speeds. The traditional non-contact fluid measurement method includes that a Laser Doppler Velocimeter (LDV) calculates the velocity of particles through the influence of fluid flowing through the intersection of two beams of laser on interference fringes, and an ultrasonic doppler velocimeter (ADV) judges the velocity of a flow field by using the doppler frequency shift of an echo after an acoustic pulse generated by an acoustic wave transmitting transducer passes through a measurement point. Although the fluid measurement modes have high measurement accuracy and higher resolution, they can only perform single-point measurement and cannot analyze the overall motion condition of the flow field, and it is difficult to obtain a transient image of the flow field and an overall result of the flow field.

The particle image velocimetry developed in the 80 th 19 th century, integrated with the development results of modern materials, digital imaging, laser technology and image analysis, is a transient, multi-point and non-contact fluid mechanics velocimetry, and can accurately measure the transient flow field in a plane. Compared with the traditional velocimetry, the particle image velocimetry can provide relatively ideal data base for qualitative description and quantitative research of fluid movement under the condition of not interfering the fluid movement. The particle image velocimetry mainly reflects the motion condition of a flow field by scattering particles with good followability in the flow field and measuring the velocity of the particles. And illuminating the particles in the flow field by using an illuminating device, shooting a particle picture of multiple exposures by using an imaging system, and processing the particle picture by using a PIV image processing method to obtain the motion of the particles. Particle image velocimetry is the main measurement method for the current flow field velocity measurement. The most critical of which is the particle image analysis algorithm. Therefore, the method has very important significance for the research of the PIV image processing algorithm.

In the prior art, most of three-dimensional PIV algorithms reconstruct and obtain corresponding three-dimensional particle fields according to shot particle images when restoring velocity fields, and then carry out cross-correlation on the three-dimensional particle fields through a three-dimensional cross-correlation algorithm to obtain the velocity fields of particles. When the three-dimensional cross-correlation operation is carried out, the accurate speed result can be obtained only by detecting that a sufficient number of particles exist in a window. Therefore, the error is large when the flow field with low particle density is measured, and the prior art needs to be improved.

Disclosure of Invention

The invention aims to provide an efficient single-camera particle image velocimetry method with low particle density.

In order to solve the technical problem, the invention provides a single-camera particle image velocimetry method with low particle density, which comprises the following steps: uniformly scattering tracer particles in a flow field, enabling a white light source to emit a collimated stable white light beam, enabling the light beam to pass through a filter to obtain colored sheet light to be emitted into the flow field, illuminating the tracer particles in the flow field, shooting by a CCD (charge coupled device) camera to obtain a colored particle picture with low particle density and transmitting the colored particle picture to an upper computer, carrying out filtering processing and data reduction sampling on the colored particle picture with low particle density by the upper computer, analyzing and reconstructing the processed colored particle picture with low particle density to obtain a three-dimensional particle distribution probability field with high particle density, and then carrying out cross-correlation calculation to obtain a fluid three-dimensional velocity field.

As an improvement of the single-camera particle image velocimetry method with low particle density of the invention: the depth of particles in the flow field is calibrated through visible light with different wavelengths, and three-dimensional information of the flow field can be obtained by combining two-dimensional pixel position information in a picture obtained by shooting of a CCD camera.

As an improvement of the single-camera particle image velocimetry method with low particle density of the invention: modeling the existence of particles at a certain point in a flow field as existence probability, establishing a corresponding image forming model according to the imaging process of an optical system to obtain the minimization problem of a linear system, and solving the minimization problem by using an ADMM algorithm; and obtaining a three-dimensional particle distribution probability field with high particle density through a color particle image with low particle density.

As an improvement of the single-camera particle image velocimetry method with low particle density of the invention: and carrying out three-dimensional cross-correlation analysis on the obtained three-dimensional particle distribution probability field with high particle density to further obtain a fluid three-dimensional velocity field.

As an improvement of the single-camera particle image velocimetry method with low particle density of the invention:

the wavelength lambda and the depth Z have a linear relation, and the position of a light plane of the particle in a three-dimensional space is determined through the wavelength lambda; representing the three-dimensional coordinates of the particle in association with the two-dimensional coordinates (x, y), (x, y, λ) of the particle in the light plane obtained by the color image; modeling the probability of particle presence at a point in three-dimensional space as P (X, λ); because incoherent light is used, the imaging process of the optical system can be modeled as a set of Point Spread Functions (PSFs), which are the impulse responses of the focusing optical system and the spatial domain representation of the imaging system transfer function; the PSF is divided into three color channels of red, green and blue, and corresponding PSF point spread functions are respectively established: g_C(x, λ), C ∈ { red, green, blue }; the image formation model can thus be modeled as:

i_C(X)＝∫_Λ∫_Xg_C(X-X′，λ)·i_r(X，λ)·P(X，λ)dX′dλ

in the formula i_C(X) color channel, i, corresponding to the captured RGB image_r(X, λ) is the corresponding spectral distribution incident on the image sensor, g_C(X-X' lambda) is a PSF function corresponding to a certain point in a three-dimensional space, the probability that particles exist at the corresponding point in the P (X, lambda) three-dimensional space, and dX is the wavelength corresponding to the light plane at the two-dimensional position d lambda on the light plane of the current point; the spatial integral corresponds to a convolution representing a potential imperfect focus; the spectral image encoding the 3D particle position (spectral distribution obtained on the image sensor) can be converted into an RGB image by integration over wavelength;

after discretization, the convolution of the PSF and the reflected light intensity can be formulated as a matrix A epsilon R^3N×NLWhere N is the number of pixels of the low particle density color particle image, L is the number of discretized stages along the wavelength coordinate direction, and the number 3 represents the three different color image channels; i.e. i_t∈R^3NRepresenting the particle image captured by the camera at time t; p_t∈[0，1]^NLRepresenting the probability of the particle being present at different points at time point t; the particle field distribution of each time point can be obtained by solving the linear system; AP (Access Point)_t＝i_tHowever, the complete spectral information of the encoded particle positions exists in three different color channels, resulting in the linear system solving being a ill-conditioned inverse problem; therefore, some prior knowledge of particle distribution needs to be introduced as a regular term, and the inverse problem becomes the following minimization problem:

in the formula: (p)^*) Set of solutions, AP, of the size of the probability that a particle exists at each point in three-dimensional space, obtained for solving the minimization problem_tThe simulated particle picture is obtained according to the image forming model; i.e. i_tA low particle density color particle picture collected at time t; a [ p ]₁|...|p_T]For a group obtained from an image forming model when analyzing a group of taken picturesSimulating a particle picture; p is a radical of_tIs the particle existence probability at time t; k is a radical of₁To optimize the parameters;

operator II_[0，1]Projecting all volume occupancy probabilities to an effective probability 0, 1]^NLThe above step (1); the first row in the formula is the least squares data fit term to solve the linear system inverse problem; the first term defines a weighting term L₁The sparse distribution of particles in the flow field to be measured is encouraged, and because the sensitivity of the camera to light with different wavelengths is different, the weighting coefficient is iteratively changed according to the previous result through further weighting by the diagonal matrix diag (w), is fixed and unchanged in the iterative process but can be changed along with the change of the particle depth, so that the sparsity can be effectively enhanced; when wavelengths in the yellow or blue-green part of the spectrum cause strong responses in multiple color channels, light of wavelengths further away from the blue or red part triggers only one channel, which results in an uneven particle distribution; different particles can be placed at the depth where the particles are more likely to be located through the setting of the weighting terms; the weighting term eliminates this bias by compensating for the non-uniformity of the luminance; the second term of the formula controls the occupation probability of the indicator function to be 0, 1]To (c) to (d);

due to the weighting term L in the optimization problem described above₁Sum index function II_[0，1](p₁；...；p_t) The optimization problem is non-smooth; adopting an alternative direction multiplier method ADMM to solve the non-smooth problem; the alternating direction multiplier method ADMM decouples a larger global problem which is difficult to solve into a plurality of smaller local problems which are easy to obtain effective solutions in a decomposition coordination program mode, and obtains a final solution of the global problem by iteratively solving a plurality of subproblems.

As an improvement of the single-camera particle image velocimetry method with low particle density of the invention:

after the optimization problem is solved by using an ADMM algorithm, a probability field of particle distribution is obtained, and then a previous three-dimensional particle distribution field is divided into a plurality of search areas with fixed sizes according to the set size of a search window; and comparing the search region of the same position in the subsequent particle fieldScanning the upper left corner line by line to the lower right corner in the inquiry window with larger search window according to a formula Performing 3D cross-correlation to obtain a matrix of correlation coefficients, wherein the coordinates of the peak values of the correlation coefficients can represent the distance of particle displacement; substituting different moments of two data sets taken into a formula A rough three-dimensional velocity field of the flow field can be obtained; and then removing gross errors according to the basic principle of hydrodynamics, and correcting wrong velocity vectors by interpolation fitting to obtain an accurate three-dimensional particle velocity field.

As an improvement of the single-camera particle image velocimetry method with low particle density of the invention: the filter is a linear variable band-pass filter.

As an improvement of the single-camera particle image velocimetry method with low particle density of the invention: the volume ratio of the particle solution to the fluid solution is less than 100.

In the prior art, most of three-dimensional PIV algorithms reconstruct and obtain corresponding three-dimensional particle fields according to shot particle images when restoring velocity fields, and then carry out cross-correlation on the three-dimensional particle fields through a three-dimensional cross-correlation algorithm to obtain the velocity fields of particles. When the three-dimensional cross-correlation operation is carried out, the accurate speed result can be obtained only by detecting that a sufficient number of particles exist in a window. Therefore, the error is larger when the low particle density flow field is measured, the invention provides the single-camera particle image velocimetry method with low particle density, and the accurate measurement result can be obtained under the condition of low particle density.

The invention discloses a single-camera particle image velocimetry method with low particle density, which has the technical advantages that:

the method models the possibility of the existence of a certain point of the particles in the flow field as the existence probability, establishes a corresponding image forming model according to the imaging process of the optical system, obtains the miniaturization problem of the linear system and solves the problem by using an ADMM algorithm. And obtaining a three-dimensional particle distribution probability field with high particle density through a color particle image with low particle density. The three-dimensional cross-correlation algorithm is improved, and a good experimental result can be obtained under the condition of low particle density.

The invention solves the problem that the three-dimensional cross-correlation algorithm in the prior art is applied to the condition of low particle density. The invention has the beneficial effects that: the color light source is used for replacing a laser light source, the depths of the particles are calibrated by using visible light with different wavelengths, and three-dimensional information of a flow field can be obtained by using one camera by combining two-dimensional pixel information shot by a CCD camera, so that the limitation of practical application is reduced. And analyzing and solving the low-particle-density color particle picture shot by the CCD to obtain a high-density three-dimensional particle distribution probability field, and analyzing the probability field by a three-dimensional cross-correlation algorithm to obtain a three-dimensional fluid velocity field.

Detailed Description

The invention will be further described with reference to specific examples, but the scope of the invention is not limited thereto.

Embodiment 1, a method for measuring a velocity of a particle image with a single camera having a low particle density, mainly comprising two parts:

one is an optical system that encodes the depth of the particles in three-dimensional space into corresponding different colors and collects color images of the particles in the fluid.

The other part is a reconstruction algorithm which is used for reducing the three-dimensional particle field and the three-dimensional velocity field of the fluid step by analyzing the color image shot by the optical system.

An optical device: compared with the conventional Particle Image Velocimetry (PIV), the particle in the fluid is illuminated by adopting the sheet light, and the particle flowing in the fluid is shot by the camera to obtain a color image with low particle density. The difference is that the conventional Particle Image Velocimetry (PIV) usually uses a laser light source to obtain sheet light perpendicular to the sight line of a camera through a cylindrical lens. The camera may be focused on the illumination plane and observing the moving particles in the tracking illumination plane may result in two components of the velocity field on the 2D slice of the current volume. And a white light source is used for replacing a laser light source, and sheet light with the wavelength linearly changing with the depth can be obtained after collimation and filtering, so that three components of the velocity field can be obtained on the same 2D slice at the same time.

A small amount of tracer particles (the volume ratio of a particle solution to a fluid solution is less than 100) are uniformly scattered in a flow field, a white light source emits a collimated stable white light beam, the light beam passes through a filter (a linear variable band-pass filter) to obtain colored sheet light, the colored sheet light is emitted into the flow field, and a tracer particle CCD camera in the flow field is illuminated to obtain a picture which is then transmitted to an upper computer.

And (3) reconstruction algorithm: after Gaussian filtering and data down-sampling are carried out on the color image of the optical device, which obtains low particle density, calculation analysis is carried out, and a three-dimensional particle distribution probability field with high particle density is obtained according to the positions and colors of particles on the image. And obtaining two three-dimensional particle fields separated by a fixed time T, and analyzing and reconstructing the two three-dimensional particle fields by using a 3D cross-correlation algorithm to obtain a fluid three-dimensional velocity field.

Reconstructing the three-dimensional particle distribution probability field of the particles requires modeling the image formation process and correlating the particle positions obtained by the model with the captured color image with low particle density. Solving the image forming model can obtain the probability that each point in the three-dimensional space of the particles has the particles, and obtaining a three-dimensional particle distribution probability field with high particle density from a three-dimensional particle field with low density. But an ill-posed inverse problem is obtained, and two regular terms are introduced to optimize the problem, so that convergence can be guaranteed to effectively solve the problem. One of the regularization terms is k to ensure sparse distribution of particles₁||diag(w)(p₁；...；p_T)||₁The other is to ensure that the probability of the particle existing at a certain point in the three-dimensional space is [0, 1 ]]On the effective convex set.

Because the illumination used by the optical device is a continuous narrow-band incoherent spectral light sheet, its wavelength λ varies with depth. Wavelength lambdaAnd depth Z is linear. The position of the light plane in the three-dimensional volume of the particle can be determined by the wavelength λ. The three-dimensional coordinates of the particles are represented in conjunction with the two-dimensional coordinates (x, y), (x, y, λ) of the particles in the light plane, which are obtained by the color image. The probability of the presence of a particle at a point in three-dimensional space is modeled as P (X, λ). Because incoherent light is used, the imaging process of the optical system can be modeled as a set of Point Spread Functions (PSFs), which are the impulse response of the focusing optical system and the spatial domain representation of the imaging system transfer function. The PSF is divided into three color channels of red, green and blue, and corresponding PSF point spread functions are respectively established: g_C(x, λ), C ∈ { red, green, blue }. The image formation model can thus be modeled as:

i_C(X)＝∫_Λ∫_Xg_C(X-X′，λ)·i_r(X，λ)·P(X，λ)dX′dλ

in the formula i_C(X) color channel, i, corresponding to the captured RGB image_r(X, λ) is the corresponding spectral distribution incident on the image sensor, g_C(X-X', λ) is a PSF function corresponding to a point in the three-dimensional space, P (X, λ) is a probability that a particle exists at a corresponding point in the three-dimensional space, and dX is a two-dimensional position d λ on the light plane of the current point is a wavelength corresponding to the light plane. The spatial integration corresponds to a convolution representing a potential imperfect focus. The spectral image encoding the 3D particle position (the spectral distribution obtained on the image sensor) can be converted into an RGB image by integration over wavelength.

After discretization, the convolution of the PSF and the reflected light intensity can be formulated as a matrix A epsilon R^3N×NLWhere N is the number of pixels of the low particle density color particle image, L is the number of discretized stages along the wavelength coordinate direction, and the number 3 represents the three different color image channels. i.e. i_t∈R^3NRepresenting the image of the particles taken by the camera at time t. P_t∈[0，1]^NLIndicating the probability that a particle exists at different points at time t. The particle field distribution at each time point can be obtained by solving a linear system. AP (Access Point)_t＝i_tBut the complete spectral information of the encoded particle position existsIn three different color channels, the solution that results in this linear system is a ill-conditioned inverse problem. Therefore, some prior knowledge of particle distribution needs to be introduced as a regular term, and the inverse problem becomes the following minimization problem:

in the formula: (p)^*) Set of solutions, AP, of the size of the probability that a particle exists at each point in three-dimensional space, obtained for solving the minimization problem_tThe simulated particle picture is obtained according to the image forming model; i.e. i_tA low particle density color particle picture collected at time t; a [ p ]₁|...|p_T]A group of simulated particle pictures are obtained according to an image forming model when a group of shot pictures are analyzed; p is a radical of_tIs the particle existence probability at time t; k is a radical of₁To optimize the parameters.

Operator II_[0，1]Projecting all volume occupancy probabilities to an effective probability 0, 1]^NLThe above. The first line in the equation is the least squares data fit term to solve the linear system inverse problem. The first term defines a weighting term L₁The sparse distribution of the particles in the flow field to be measured is encouraged, and the sensitivity of the camera to light with different wavelengths is different, so that the weighting coefficient is further weighted through the diagonal matrix diag (w), and is iteratively changed according to the previous result, the weighting coefficient is fixed and unchanged in the iteration process but is changed along with the change of the particle depth, and the sparsity can be effectively enhanced. When wavelengths in the yellow or blue-green part of the spectrum cause strong responses in multiple color channels, light of wavelengths further away from the blue or red part triggers only one channel, which results in an uneven particle distribution. Different particles can be placed at depths where they are more likely to be by setting the weighting term. The weighting term eliminates this bias by compensating for photometric non-uniformities. The second term of the formula controls the occupation probability of the indicator function to be 0, 1]In the meantime.

Due to the weighting term L in the optimization problem described above₁And index letterCounting II_[0，1](p₁；...；p_t) The optimization problem is not smooth. And therefore cannot be solved by conventional optimization methods such as gradient descent methods and the like. To solve the non-smooth optimization problem, the non-smooth item can be decoupled from the original optimization problem, so that different parts can be processed respectively. The present invention uses an Alternating Direction Method of Multipliers (ADMM) to solve the non-smoothness problem. The ADMM decouples a larger global problem which is difficult to solve into a plurality of smaller local problems which are easy to obtain effective solutions in a decomposition coordination program mode, and iterative solution is carried out on a plurality of sub-problems to obtain a final solution of the global problem.

ADMM framework reconstruction of particle distribution fields

Inputting: an objective function F1, H1, the number of iterations j, a relaxation variable z, a maximum number of iterations max, p

And (3) outputting: dual variable q

Wherein: f₁And H₁Is defined as:H₁(p)＝k₁||diag(w)(p₁；...；p_T)||₁+П_[0，1](p₁；...；p_t)

derivation of the near-end operator: simplifying the formula to convert z^j-q^jReduced to d^j，p^j+1-q^jReduced to e^j. Near-end operatorIs a conjugate gradient solving process:

near-end operatorIs a point-by-point contraction operation and then projected to [0, 1 ]]On the domain of (c): z ═ II_[0，1]((e^j-τ₁k₁w))-(-e^j-τ₁k₁w))

After the optimization problem is solved by using an ADMM algorithm, a probability field of particle distribution is obtained, and then the previous three-dimensional particle distribution field is divided into a plurality of search areas with fixed sizes according to the set size of a search window. And the searching area with the same position in the next particle field is scanned from the upper left corner to the lower right corner line by line in an inquiry window larger than the searching window according to a formula Performing 3D cross-correlation, a matrix of correlation coefficients can be obtained, wherein the coordinates of the peak of the correlation coefficients can represent the distance of particle displacement. Substituting different moments of two data sets taken into a formula A rough three-dimensional velocity field of the flow field can be obtained. And then removing gross errors according to the basic principle of fluid mechanics, and correcting wrong velocity vectors to a certain extent through interpolation fitting to obtain an accurate fluid three-dimensional velocity field.

Finally, it is also noted that the above-mentioned lists merely illustrate a few specific embodiments of the invention. It is obvious that the invention is not limited to the above embodiments, but that many variations are possible. All modifications which can be derived or suggested by a person skilled in the art from the disclosure of the present invention are to be considered within the scope of the invention.