CN113744358A - Three-dimensional particle field and velocity field reconstruction method based on GPU acceleration - Google Patents
Three-dimensional particle field and velocity field reconstruction method based on GPU acceleration Download PDFInfo
- Publication number
- CN113744358A CN113744358A CN202111003737.5A CN202111003737A CN113744358A CN 113744358 A CN113744358 A CN 113744358A CN 202111003737 A CN202111003737 A CN 202111003737A CN 113744358 A CN113744358 A CN 113744358A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- field
- gpu
- reconstruction
- velocity
- 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
Links
- 239000002245 particle Substances 0.000 title claims abstract description 87
- 238000000034 method Methods 0.000 title claims abstract description 58
- 230000001133 acceleration Effects 0.000 title claims abstract description 9
- 239000011159 matrix material Substances 0.000 claims abstract description 43
- 238000004364 calculation method Methods 0.000 claims abstract description 37
- 230000008569 process Effects 0.000 claims abstract description 27
- 238000003745 diagnosis Methods 0.000 claims abstract description 25
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000000638 solvent extraction Methods 0.000 claims abstract description 5
- 238000005259 measurement Methods 0.000 claims description 12
- 238000005314 correlation function Methods 0.000 claims description 10
- 238000003325 tomography Methods 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 6
- 238000013500 data storage Methods 0.000 claims description 4
- 239000000700 radioactive tracer Substances 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 22
- 238000000917 particle-image velocimetry Methods 0.000 description 13
- 238000004587 chromatography analysis Methods 0.000 description 7
- 238000011160 research Methods 0.000 description 5
- 239000012530 fluid Substances 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002591 computed tomography Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a three-dimensional particle field and velocity field reconstruction method based on GPU acceleration, which comprises the following steps: acquiring two groups of multi-camera image data with a certain time interval and a weight matrix search curve of each camera; adopting a GPU (graphics processing unit), carrying out three-dimensional particle field chromatographic reconstruction based on the multi-camera image data and the weight matrix search curve to obtain a reconstructed three-dimensional particle field, and in the chromatographic reconstruction process, using the thread number equal to the number of the voxels to calculate the light intensity reconstruction of all the voxels in parallel; and partitioning the reconstructed three-dimensional particle field, extracting two diagnosis window data of each data block, performing three-dimensional cross-correlation calculation by adopting a GPU concurrent process to obtain the average speed of all windows, and reconstructing to obtain a three-dimensional speed field. Compared with the prior art, the method has the advantages of effectively reducing the calculation time overhead, improving the efficiency and the like.
Description
Technical Field
The invention relates to the technical field of particle image speed measurement, in particular to a three-dimensional particle field based on GPU acceleration and a velocity field reconstruction method.
Background
Particle Image Velocimetry (PIV) technology is a non-contact interference-free quantitative measurement means for a space flow field. The PIV technology utilizes uniformly dispersed trace particles with small particle sizes to follow the motion of fluid, records the motion of the trace particles in a continuous 'lighting' and 'photographing' mode, extracts the motion information of the particles from continuous trace particle images based on an image processing algorithm, and obtains the approximation of the fluid space motion speed. The Tomography Particle Image Velocimetry (Tomo-PIV) technology combines the PIV technology with the Computed Tomography diagnostic technology (CT), uses a multi-view (at least three cameras) to shoot and record the illuminated Particle field simultaneously, and reconstructs the three-dimensional distribution of light intensity by an optical Tomography method to obtain the three-dimensional space distribution of the tracing particles; and then reconstructing the particle field through a cross-correlation algorithm to obtain three-dimensional velocity field distribution of the flow field.
Compared with the traditional two-dimensional PIV technology, the Tomo-PIV technology can realize instantaneous three-dimensional flow field velocity measurement and full-field quantitative measurement of a space flow field, and can provide great convenience for the research of complex flow field structures (such as flow fields with turbulent flow and three-dimensional vortex structures). The Tomo-PIV technology has the advantages of simple operation, high measurement precision, simple equipment, wide application range, more acquired information quantity and the like; however, there are limitations such as requiring higher volume illumination energy, huge data storage capacity, and low computational efficiency.
The core step of obtaining the three-dimensional particle field in the Tomo-PIV technology is a particle field chromatography reconstruction algorithm based on an image, the calculation of a weight matrix in the calculation process needs more than 12h, and the storage space of 200-500GB needs to be occupied, so that the extremely high calculation cost is consumed, and the method becomes a main technical bottleneck of measuring a flow field by the PIV technology. Secondly, the velocity field reconstruction after obtaining the particle distribution is also a core step of the Tomo-PIV technology, and since each particle (each particle is approximately a cubic grid and is called a voxel) of the two frames of particle field distribution needs to be compared in detail, the number of voxels obtained by tomographic reconstruction is often extremely large, so that the efficiency problem of velocity field reconstruction is also one of the application obstacles of the Tomo-PIV technology.
In recent years, research on particle field tomography Reconstruction at home and abroad focuses on improving the calculation precision of the Tomo-PIV technology, and numerous Reconstruction algorithms including a Multiplicative Algebraic Reconstruction Technique (MART), a body self-calibration technique, a Multiplicative First visit (MFG) algorithm, an MLOS-MART algorithm, an IntE-MART algorithm and the like are provided, so that the quality of three-dimensional particle field Reconstruction and the precision of three-dimensional velocity field Reconstruction are improved. However, since these algorithms run on the CPU, the efficiency improvement is not significant. In the velocity field reconstruction of the PIV technology, scholars at home and abroad begin to use a GPU (Graphics Processing Unit, image processor) to accelerate the cross-correlation algorithm of the traditional two-dimensional PIV or three-dimensional PIV to reconstruct the velocity field; however, the researches do not deeply optimize the GPU thread design, only the CUFFT module in the GPU function library is relied on to replace the fast Fourier transform in the cross-correlation algorithm, the multithreading concurrency mode is not deeply researched, and the velocity field reconstruction module cannot fully exert the advantages of the GPU.
In summary, the problem of the efficiency of particle field reconstruction and velocity field reconstruction has severely hampered the large-scale application of the Tomo-PIV technique, and improvements to the prior art are needed.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provide a three-dimensional particle field and velocity field reconstruction method based on GPU acceleration, which effectively reduces the calculation time overhead and improves the efficiency.
The purpose of the invention can be realized by the following technical scheme:
a three-dimensional particle field and velocity field reconstruction method based on GPU acceleration comprises the following steps:
acquiring two groups of multi-camera image data with a certain time interval and a weight matrix search curve of each camera;
adopting a GPU (graphics processing unit), carrying out three-dimensional particle field chromatographic reconstruction based on the multi-camera image data and the weight matrix search curve to obtain a reconstructed three-dimensional particle field, and in the chromatographic reconstruction process, using the thread number equal to the number of the voxels to calculate the light intensity reconstruction of all the voxels in parallel;
and partitioning the reconstructed three-dimensional particle field, extracting two diagnosis window data of each data block, performing three-dimensional cross-correlation calculation by adopting a GPU concurrent process to obtain the average speed of all windows, and reconstructing to obtain a three-dimensional speed field.
Further, the weight matrix search curve is obtained by the following steps:
calibrating multiple cameras to obtain a camera calibration matrix;
mapping the coordinates of the experimental image to a space three-dimensional real coordinate system based on the camera calibration matrix;
according to the space three-dimensional real coordinate obtained by mapping, the sight line of each image pixel point in the space three-dimensional real coordinate system is approximate to a cylinder, each voxel in the three-dimensional real coordinate system is approximate to a sphere, the contribution value of each voxel corresponding to the image pixel point is approximate to the ratio of the volume of the intersection of the cylinder and the voxel to the volume of the whole voxel, and the contribution value of each voxel corresponding to each pixel, namely a weight matrix searching curve, is fitted.
Further, the camera calibration matrix is obtained based on control constants and calibration data, wherein the control constants comprise particle field tomography reconstruction parameters and velocity field reconstruction parameters.
Further, the particle field tomography reconstruction parameters comprise an image data number, an image data type, an image data size, an image data storage path, a number of cameras, a measurement region size and a number of voxels per millimeter,
the velocity field reconstruction parameters include a diagnostic window size and a window overlap ratio.
Furthermore, in the three-dimensional particle field tomographic reconstruction, spatial distribution of the three-dimensional particle field is obtained by using a parallel MART algorithm based on a GPU.
Further, the calculation formula of the MART algorithm is as follows:
where k is the number of iterations, μ is the relaxation parameter, E (x)j,yj,zj) Is the voxel gray value, I (x)i,yj) Is a camera image pixel value, Wi,jA contribution weight matrix for each voxel j to the gray value of each pixel i, the contribution weight matrixObtaining by searching a curve according to the multi-camera image data and the weight matrix, NiIs the total number of voxels.
Further, the reconstructed three-dimensional particle field is partitioned based on the size of the diagnosis window and the window overlapping rate, and two pieces of diagnosis window data of each data block are extracted through an adaptive window selection technology.
Further, the three-dimensional cross-correlation calculation specifically includes:
performing three-dimensional Fourier transform on the two diagnosis window data based on CUFFT fast Fourier transform of the GPU respectively to obtain frequency domain transform results of the two diagnosis windows;
performing CUFFT convolution calculation based on a GPU to obtain energy spectrums of particle fields of two diagnosis windows;
performing inverse Fourier transform based on CUFFT inverse fast Fourier transform of the GPU to obtain a cross-correlation function of particle fields of two diagnosis windows;
and solving the size of the peak value of the cross-correlation function and the index of the three-dimensional array where the peak value is located, and calculating to obtain the average velocity of the tracer particles in the diagnosis window.
Further, after obtaining the average speed of all windows, searching whether an error speed vector exists, and if so, performing replacement processing.
Further, the time interval is 0.001 s.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention uses the novel GPU hardware to accelerate the particle field chromatography reconstruction algorithm in the Tomo-PIV technology and the three-dimensional cross-correlation algorithm in the velocity field reconstruction process, can greatly improve the calculation efficiency of the Tomo-PIV technology, reduces the complex data processing work of a user when using the Tomo-PIV technology related hardware equipment, saves a large amount of time, accelerates the experimental fluid mechanics research speed, provides a basis for the wide application of the Tomo-PIV technology in the flow field reconstruction, and has important significance for accelerating the experimental fluid research.
2. The method can greatly improve the particle field reconstruction efficiency and reduce most of time overhead in the chromatographic particle image velocimetry technology based on GPU parallel chromatographic reconstruction.
3. The method is based on the multi-task concurrent thinking, the three-dimensional cross-correlation calculation of a large number of particle field diagnosis windows is decomposed into multi-process concurrent calculation, and the calculation time overhead of speed field reconstruction is greatly reduced.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a flow chart of a three-dimensional particle field tomographic reconstruction of the present invention;
FIG. 3 is a flow chart of the three-dimensional velocity field reconstruction of the present invention;
FIG. 4 is a graph of a weight matrix lookup computed in an embodiment of the present invention;
FIG. 5 is a graph of 8 images read in accordance with an embodiment of the present invention;
fig. 6 is a three-dimensional particle field at two moments obtained based on accelerated reconstruction by a GPU in the embodiment of the present invention;
FIG. 7 is two time diagnostic window data extracted based on a three-dimensional particle field in an embodiment of the present invention;
fig. 8 is 4 sets of two-time diagnostic window data of 4 concurrent processes extracted based on a three-dimensional particle field in an embodiment of the present invention;
FIG. 9 is a cross-correlation function distribution obtained based on GPU-accelerated three-dimensional cross-correlation in an embodiment of the present invention;
FIG. 10 is a three-dimensional velocity field result in an embodiment of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
The invention provides a three-dimensional particle field and velocity field reconstruction method based on GPU acceleration, which adopts a GPU to search a curve based on multi-camera image data and a weight matrix to carry out three-dimensional particle field chromatography reconstruction to obtain a reconstructed three-dimensional particle field, and in the chromatography reconstruction process, the light intensity reconstruction of all voxels is calculated in parallel by using the number of threads equal to the number of the voxels; and partitioning the reconstructed three-dimensional particle field, extracting two diagnosis window data of each data block, performing three-dimensional cross-correlation calculation by adopting a GPU concurrent process to obtain the average speed of all windows, and reconstructing to obtain a three-dimensional speed field. The method can greatly improve the particle field reconstruction efficiency based on GPU parallel chromatography reconstruction, and reduce most of time overhead in the chromatography particle image velocimetry technology; meanwhile, based on the multi-task concurrency thinking, the three-dimensional cross-correlation calculation of a large number of particle field diagnosis windows is decomposed into multi-process concurrent calculation, and the calculation time overhead of speed field reconstruction is greatly reduced.
Referring to fig. 1 to fig. 3, a flow chart of a three-dimensional particle field and velocity field reconstruction method for accelerating tomographic particle image velocimetry based on a GPU of the present invention specifically includes the following steps:
1) and reading the calculated control constant and calibration data from the input file, and calibrating the camera to obtain a camera calibration matrix.
11) Reading control constants of the embodiment, including an image data number, an image data type, an image data size, an image data storage path, the number of cameras, an actual size of a measurement area, the number of voxels per millimeter and the like required by tomographic reconstruction; parameters including diagnostic window size, window overlap ratio, etc. required for velocity field reconstruction are shown in table 1.
TABLE 1 control parameters
12) And reading calibration data, including all calibration data required for calibrating the camera and calculating the calibration matrix A.
13) Calibrating the camera, and solving to obtain 12 unknown numbers a in the camera calibration matrix A according to the control constant data and the calibration data in the step 11) and the step 12)11，a12，a13，a14，a21，a22，a23，a24，a31，a32，a33，a34Construction ofThe calibration function relationship of the space three-dimensional real coordinate system and the image plane pixel coordinate system is as follows:
wherein, λ is a scale factor, (u, v) is a pixel coordinate of an image point P, and (X, Y, Z) is a three-dimensional real coordinate of a three-dimensional real space body pixel point P; and A is a camera projective transformation matrix.
2) According to the camera calibration matrix obtained by calculation in the step 1), mapping the coordinates of the experimental image to the three-dimensional real coordinates of the space through a calibration function relationship, and establishing a weight matrix search curve.
21) Reading two experimental images according to the camera calibration matrix A obtained in the step 1), and mapping the experimental image coordinates to a space three-dimensional real coordinate system through a calibration function relation.
22) According to the space three-dimensional real coordinate result obtained in the step 21), approximating the sight line of each image pixel point in the space three-dimensional real coordinate system to be a cylinder, approximating each voxel in the three-dimensional real coordinate system to be a sphere, approximating the contribution value of each voxel corresponding to the image pixel point to the ratio of the volume of the intersection of the sight line cylinder and the voxel to the volume of the whole voxel, and fitting the contribution value of each voxel corresponding to each pixel, namely a weight matrix search curve.
In this embodiment, the number of cameras is 4, each camera needs to calculate a weight matrix search curve, 4 weight matrix search curves are obtained through calculation, the data length of each table is 18000 points, and the results of the 4 weight matrix search curves are shown in fig. 4.
3) And (3) performing three-dimensional particle field tomography reconstruction based on the GPU.
As shown in fig. 2, the three-dimensional particle field tomography reconstruction specifically includes:
31) two groups of multi-camera image data with the time interval of 0.001s, including the number of cameras required by the tomography reconstruction process, the images recorded by each camera and the coordinates and gray values of each pixel of each image, are read and transferred to the memory of the GPU device.
In this embodiment, the number of cameras is 4, the shooting time interval is 0.001s, four cameras shoot one image at two moments respectively, the total number of the images is 8, the size of the image is 1000 × 2000 pixels, the numbers of the images are 1 to 4 and 5 to 8, 1 to 4 are 4 images at the first moment, 5 to 8 are 4 images at the second moment, and all the read images are as shown in fig. 5.
32) Copying and transmitting the weight matrix lookup curve of each camera obtained in the step 2) to a GPU memory.
33) Voxel gray value E (X) based on all voxel center coordinates in three-dimensional spacej,Yj,Zj) All pixels I (x) of the image taken with each camerai,yi) According to the camera image data transmitted into the GPU in the step 31) and the weight matrix searching curve transmitted into the GPU in the step 32), a contribution weight matrix W of each voxel j to the gray value of each pixel i of each camera image is obtained through calculationi,j。
In this example, the measurement area size is 50mm × 100mm × 100mm, and the number of voxels per mm is 11 voxels. Simultaneously calculating the contribution weight matrix W of all voxels j at two moments to all image pixels i with image numbers of 1-8 by using GPU multithreadingi,j。
34) According to the contribution weight matrix W calculated in step 33)i,jAnd carrying out chromatography reconstruction by using a MART algorithm based on GPU parallel to obtain the gray value space distribution of the three-dimensional particle field. And (3) performing chromatographic reconstruction calculation based on the GPU, calculating the light intensity reconstruction process of each voxel j and all camera pixels by adopting a GPU thread, and calculating the light intensity reconstruction process of all voxels in parallel by using the thread quantity equal to the voxel quantity by utilizing the characteristic of huge thread quantity in GPU equipment. And (3) accelerating the chromatographic reconstruction calculation process based on the CPU before, and reconstructing to obtain the three-dimensional particle field with higher calculation efficiency.
The calculation formula of the MART algorithm is as follows:
where k is the number of iterations, μ is the relaxation parameter, E (x)j,yj,zj) Is the voxel gray value, I (x)i,yj) Is a camera image pixel value, Wi,jA contribution weight matrix for each voxel j to the gray value of each pixel i, the contribution weight matrix being obtained from the multi-camera image data and a weight matrix look-up curve, NiIs the total number of voxels.
In the embodiment, the number of the cameras is 4, the shooting time interval is 0.001s, four cameras shoot one image at two moments respectively, the total number of the images is 8, the size of the image is 1000 × 2000 pixels, the number of the images is 1-4 and 5-8, 1-4 are 4 images at a first moment, and 5-8 are 4 images at a second moment; the measurement area size was 50mm x 100mm, and the number of voxels per mm was 11 voxels. Simultaneous multithreading using the GPU simultaneously computes gray values for all particles (voxels) at two times, the three-dimensional particle fields at the two times being shown in fig. 6.
4) GPU-based three-dimensional velocity field reconstruction.
As shown in fig. 2, the three-dimensional velocity field reconstruction specifically includes:
41) partitioning the particle field data according to the three-dimensional particle field data of the two moments obtained in the step 3) and based on the size of the diagnostic window obtained in the step 1), extracting two diagnostic window data used for three-dimensional cross-correlation calculation in each data block through a self-adaptive window selection technology, and storing the two diagnostic window data in a GPU memory.
In this embodiment, the diagnostic window size is 32 × 32, the window overlap rate is 50%, and the total number of diagnostic windows in three dimensions of the measurement region is 32 × 64 × 32; the diagnostic window data a with the voxel (particle) coordinate center (17voxel ) at the first time point is extracted, and the diagnostic window data b with the voxel coordinate center (17voxel,17voxel,17voxel) at the second time point is extracted and copied to the GPU memory as shown in fig. 7.
42) Adaptively designing the number of concurrent processes according to the size of a GPU memory of the computing equipment, wherein each process can complete the selection of the data of the diagnosis window in the step 41), and independently process one three-dimensional cross-correlation computing module, and the maximum concurrent process simultaneously processes all the three-dimensional cross-correlation computations of the diagnosis windows.
In the embodiment, the number of concurrent processes is set to be 4 according to the size of a GPU memory of the computing equipment; 4 processes respectively extract diagnosis window data a1, a2, a3 and a4 at the first moment; and the diagnostic window data b1, b2, b3, b4 at the second time, as shown in fig. 8, are copied into the GPU memory.
43) And for each process in the step 42), according to the two diagnosis window data extracted in the step 41), performing three-dimensional Fourier transform on the data by using a CUFFT fast Fourier transform module based on the GPU respectively to obtain frequency domain transform results of the two diagnosis windows.
44) Performing GPU-based CUFFT convolution calculation on the frequency domain transformation results of the two diagnosis windows obtained by calculation in the step 43) to obtain energy spectrums of particle fields of the two diagnosis windows; then, a GPU-based CUFFT inverse fast fourier transform module performs an inverse fourier transform to obtain a cross-correlation function of two diagnostic window particle fields of 4 processes, as shown in fig. 9.
45) According to the cross-correlation function of the two diagnostic windows obtained in the step 44), the peak value of the cross-correlation function and the three-dimensional array index where the peak value is located are obtained, and the average speed of the tracer particles in the diagnostic windows is obtained through calculation.
In this embodiment, the peak coordinates, the peak values and the average velocities of the three dimensions obtained from the cross-correlation function distribution shown in the four processes of fig. 9 are shown in table 2.
Table 2 the peak coordinates, peak values and average velocities of three dimensions of the cross-correlation function calculated by the four processes
45) Step 42) is repeated, step 43), step 44) until the average velocity calculation for all diagnostic windows is completed.
In this embodiment, the diagnostic window size is 32 × 32, the window overlap rate is 50%, and the total number of diagnostic windows in three dimensions of the measurement region is 32 × 64 × 32; a total of 32 x 64 x 32-65536 cross-correlation cycles are performed.
46) The average velocity of the trace particles in all the diagnostic windows obtained in step 45) is analyzed to replace the erroneous velocity vectors therein.
5) And (4) outputting and storing the three-dimensional particle field and three-dimensional velocity field results obtained by calculation in the step 2), the step 3) and the step 4), and finishing the calculation.
The three-dimensional velocity field results for this example are shown in fig. 10.
The performance improvement of this embodiment using a CPU and different GPUs is shown in table 3.
TABLE 3 analysis of properties
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (10)
1. A three-dimensional particle field and velocity field reconstruction method based on GPU acceleration is characterized by comprising the following steps:
acquiring two groups of multi-camera image data with a certain time interval and a weight matrix search curve of each camera;
adopting a GPU (graphics processing unit), carrying out three-dimensional particle field chromatographic reconstruction based on the multi-camera image data and the weight matrix search curve to obtain a reconstructed three-dimensional particle field, and in the chromatographic reconstruction process, using the thread number equal to the number of the voxels to calculate the light intensity reconstruction of all the voxels in parallel;
and partitioning the reconstructed three-dimensional particle field, extracting two diagnosis window data of each data block, performing three-dimensional cross-correlation calculation by adopting a GPU concurrent process to obtain the average speed of all windows, and reconstructing to obtain a three-dimensional speed field.
2. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 1, wherein the weight matrix lookup curve is obtained by:
calibrating multiple cameras to obtain a camera calibration matrix;
mapping the coordinates of the experimental image to a space three-dimensional real coordinate system based on the camera calibration matrix;
according to the space three-dimensional real coordinate obtained by mapping, the sight line of each image pixel point in the space three-dimensional real coordinate system is approximate to a cylinder, each voxel in the three-dimensional real coordinate system is approximate to a sphere, the contribution value of each voxel corresponding to the image pixel point is approximate to the ratio of the volume of the intersection of the cylinder and the voxel to the volume of the whole voxel, and the contribution value of each voxel corresponding to each pixel, namely a weight matrix searching curve, is fitted.
3. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 2, wherein the camera calibration matrix is obtained based on control constants and calibration data, the control constants comprising particle field tomographic reconstruction parameters and velocity field reconstruction parameters.
4. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 3, wherein the particle field tomography reconstruction parameters include image data number, image data type, image data size, image data storage path, number of cameras, measurement region size, and number of voxels per millimeter,
the velocity field reconstruction parameters include a diagnostic window size and a window overlap ratio.
5. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 1, wherein in the three-dimensional particle field tomographic reconstruction, spatial distribution of the three-dimensional particle field is obtained by using a GPU-parallel-based MART algorithm tomographic reconstruction.
6. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 5, wherein the MART algorithm has the following calculation formula:
where k is the number of iterations, μ is the relaxation parameter, E (x)j,yj,zj) Is the voxel gray value, I (x)i,yj) Is a camera image pixel value, Wi,jA contribution weight matrix for each voxel j to the gray value of each pixel i, the contribution weight matrix being obtained from the multi-camera image data and a weight matrix look-up curve, NiIs the total number of voxels.
7. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 4, wherein the reconstructed three-dimensional particle field is partitioned based on the diagnostic window size and window overlap ratio, and two diagnostic window data of each data block are extracted by an adaptive window selection technique.
8. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 1, wherein the three-dimensional cross-correlation calculation specifically comprises:
performing three-dimensional Fourier transform on the two diagnosis window data based on CUFFT fast Fourier transform of the GPU respectively to obtain frequency domain transform results of the two diagnosis windows;
performing CUFFT convolution calculation based on a GPU to obtain energy spectrums of particle fields of two diagnosis windows;
performing inverse Fourier transform based on CUFFT inverse fast Fourier transform of the GPU to obtain a cross-correlation function of particle fields of two diagnosis windows;
and solving the size of the peak value of the cross-correlation function and the index of the three-dimensional array where the peak value is located, and calculating to obtain the average velocity of the tracer particles in the diagnosis window.
9. The GPU-acceleration-based three-dimensional particle field and velocity field reconstruction method of claim 1, wherein after obtaining the average velocity of all windows, searching whether an error velocity vector exists therein, and if so, performing replacement processing.
10. A GPU acceleration based reconstruction method of a three dimensional particle field and velocity field as defined in claim 1, wherein the time interval is 0.001 s.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111003737.5A CN113744358B (en) | 2021-08-30 | 2021-08-30 | Three-dimensional particle field and speed field reconstruction method based on GPU acceleration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111003737.5A CN113744358B (en) | 2021-08-30 | 2021-08-30 | Three-dimensional particle field and speed field reconstruction method based on GPU acceleration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113744358A true CN113744358A (en) | 2021-12-03 |
| CN113744358B CN113744358B (en) | 2024-02-20 |
Family
ID=78733812
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111003737.5A Active CN113744358B (en) | 2021-08-30 | 2021-08-30 | Three-dimensional particle field and speed field reconstruction method based on GPU acceleration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113744358B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114972614A (en) * | 2022-05-09 | 2022-08-30 | 北京理工大学 | Reverse mapping cross-interface tomography method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111899344A (en) * | 2020-06-30 | 2020-11-06 | 南京理工大学 | Flame emission tomography reconstruction device and method based on camera array |
| WO2021114306A1 (en) * | 2019-12-14 | 2021-06-17 | 深圳先进技术研究院 | High-precision pet reconstruction method and apparatus based on gpu acceleration |
-
2021
- 2021-08-30 CN CN202111003737.5A patent/CN113744358B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021114306A1 (en) * | 2019-12-14 | 2021-06-17 | 深圳先进技术研究院 | High-precision pet reconstruction method and apparatus based on gpu acceleration |
| CN111899344A (en) * | 2020-06-30 | 2020-11-06 | 南京理工大学 | Flame emission tomography reconstruction device and method based on camera array |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114972614A (en) * | 2022-05-09 | 2022-08-30 | 北京理工大学 | Reverse mapping cross-interface tomography method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113744358B (en) | 2024-02-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ray et al. | Ray casting architectures for volume visualization | |
| Krone et al. | Fast Visualization of Gaussian Density Surfaces for Molecular Dynamics and Particle System Trajectories. | |
| CN110187143A (en) | A kind of chromatography PIV reconstructing method and device based on deep neural network | |
| Faridani et al. | Introduction to local tomography | |
| Liu et al. | Surface area estimation of digitized 3D objects using quasi-Monte Carlo methods | |
| Zeng et al. | GPU-accelerated MART and concurrent cross-correlation for tomographic PIV | |
| Xie et al. | Artifact removal using GAN network for limited-angle CT reconstruction | |
| Shen et al. | 3D shape reconstruction from images in the frequency domain | |
| Thomas et al. | Influence of geometric parameters and image preprocessing on tomo-PIV results | |
| CN113744358B (en) | Three-dimensional particle field and speed field reconstruction method based on GPU acceleration | |
| Zhu et al. | Photon-driven neural reconstruction for path guiding | |
| Sano et al. | Parallel processing of the shear-warp factorization with the binary-swap method on a distributed-memory multiprocessor system | |
| Mousazadeh et al. | GPU implementation of a deformable 3D image registration algorithm | |
| Mujahid et al. | GPU-accelerated multivariate empirical mode decomposition for massive neural data processing | |
| Russig et al. | GPU-based raycasting of hermite spline tubes | |
| Schick et al. | Real-time GPU-based voxel carving with systematic occlusion handling | |
| He et al. | GPU-based active contour segmentation using gradient vector flow | |
| EP3493163B1 (en) | Compressive sensing of light transport matrix | |
| Xing et al. | FPGA-accelerated real-time volume rendering for 3D medical image | |
| Hochstetter et al. | Adaptive sampling for on-the-fly ray casting of particle-based fluids. | |
| Feng et al. | Cicero: Addressing Algorithmic and Architectural Bottlenecks in Neural Rendering by Radiance Warping and Memory Optimizations | |
| Vyatkin et al. | Multi-level ray casting of function-based surfaces | |
| Alakkari et al. | Principal component analysis techniques for visualization of volumetric data | |
| Bajaj et al. | Compression-based ray casting of very large volume data in distributed environments | |
| Jian-lin et al. | Matrix approach for processing of iterative reconstruction on cone beam CT |
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 |