Particle Image Velocimetry Based On Light Field Imaging

The three-dimensional unsteady flow widely exists in many fields such as nature,energy and power,aerospace,environmental protection and medicine etc.The breakthroughs of many problems and key techniques in these fields are often closely related to the development of high-precision flow measurement technique.The measurement and visualization of velocity field of complex flow is helpful for the in-depth study of flow mechanism and the establishment of new hydrodynamic model,which provides theoretical basis for the optimal design of fluid system.Particle image velocimetry(PIV)has the advantages of non-intrusiveness and instantaneity,and is capable of achieving a full velocity field,so it has become one of the important means of the study of fluid flow.To counter the problems such as complicated system,high cost,difficult calibration and low efficient particle field reconstruction in the existing tomographic PIV(Tomo-PIV)system,a particle image three-dimensional(3D)flow field measurement technique based on the light field imaging of a single camera is proposed and investigated in this thesis,which provides a new technical means for the measurement of 3D flow field in a confined space.The forward ray tracing and the backward ray tracing techniques based on the geometric optics are studied.The light field imaging model of tracer particles in the flow field and the refocusing image model of light field image are established by the forward ray tracing method and the backward ray tracing method,respectively.The forward ray tracing technique is used for the calculation of light field imaging of tracer particle field.The imaging results of a single tracer particle at different depths are compared on the conventional camera,the plenoptic camera,the focused light field camera and the the commercial Raytrix R29 focused light field camera(Raytrix R29).Results showed that the conventional camera is not capable of providing the depth position of tracer particle according to the particle image.The light field image captured by the light field camera contains the particle depth information.The digital refocus technique based on the backward ray tracing technique is studied and used for the accurate calculation of the refocused image of light field image of tracer particle.These studies provide the theoretical basis for the calculation of the weight matrix and the tomographic reconstruction in light field Tomo-PIV,and the light field synthetic aperture PIV based on 3D U-Net neural network(light field SA-PIV).Aiming at the problems of low computational efficiency and large computational storage space in the calculation of weight matrix,and time-consuming in the tomographic reconstruction of light field tomo-PIV,a calculation method of the weight matrix based on the backward ray tracing technique combined with a Gaussian function(named as Gaussian function method)is proposed.The backward ray tracing technique is used to trace all pixels’line-of-sight of the light field image to the object space,and then the Gaussian function is used to calculate the contribution value of voxel to pixel.The summed line of sight(SLOS)is further employed for the pre-determination of non-zero voxels in the measured flow field,and then the calculation of weight matrix is optimized.On this basis,an expectation-maximization(EM)in combination with the SLOS(SLOS-EM)is proposed to improve the reconstruction efficiency of the 3D particle field.Numerical simulation results showed that the Gaussian function method has almost the same calculation accuracy of the weight matrix as Fahringer’s method.However,the Gaussian function method has higher computational efficiency than Fahringer’s method.The SLOS greatly improves the computational efficiency of the weight matrix.The SLOS-EM improves the reconstruction efficiency of 3D particle field.Meanwhile,the reconstruction quality of 3D particle field hardly decreased.The reconstruction quality of the 3D particle field in the flow field is closely related to the optical parameters of the light field camera(the inverse magnification of the main lens and the microlens(M_l and M_m),the focal length of the main lens and the microlens(f and f_m)and microlens pitch(P_m)).The effects of optical parameters of the light field camera on the length and the position error of reconstructed particle,and the size of the measured volume along the Z-axis depth direction are studied systematically.Results showed that M_l、f、f_m and P_m have little effect on the sampling of position and direction of the light field camera.M_m determines the sampling of position and direction of the light field camera.M_l、M_m、f_m and P_mdetermine the tomographic reconstruction quality.A small M_l,a small f_m,a large P_m and a large|M_m|is helpful for the improvement of the reconstruction quality and the extension of the size of the measurement volume along the Z axis.But f has no effect on the size and position of the measured volume.This provides a theoretical basis for the optimization of optical parameters of light field camera in light field PIV system.To further improve the reconstruction efficiency of 3D particle field and reduce the length of reconstructed particles in the Z-axis depth direction,by combining the digital refocusing with deep learning technique,a light field SA-PIV is proposed.A reconstruction model of 3D particle field based on the 3D U-Net neural network is established.The data set between the refocused image of light field image of particle field and the theoretical 3D particle field is established by using the light field digital refocusing technique.The data set is trained by the 3D U-Net neural network to determine the mapping model between the refocused image of the light field image of particle field and the theoretical 3D particle field.Then,the trained 3D U-Net neural network model is used to reconstruct the refocused image of light field image,and the 3D particle field in the measured volume is obtained.Numerical simulation results showed that the light field SA-PIV based on the 3D U-Net neural network only needs about 8 seconds to reconstruct the 3D particle field with 128×128×128 voxels.It takes about 220 seconds to complete the reconstruction of384×384×384 voxels.The 3D U-Net neural network model occupies almost no computer memory,greatly improves the reconstruction efficiency of particle field,and reduces the need for computer memory.At the same time,it can effectively reduce the length of reconstructed particles and improve the measurement accuracy of flow field.An experimental system of the light field PIV based on the Raytrix R29 is constructed.The optical parameters of Raytrix R29 are calibrated by using the magnification calibration method.The accuracy of optical parameters of Raytrix R29 calibrated by the magnification calibration method is then experimentally evaluated.Experiments are further carried out to test the performance of light field PIV on the laminar flow and submerged water jet flow devices,respectively.Light field digital refocusing PIV is used for initial measurement of the measured flow field to obtain the basic prior information of the measured flow field,which provides the basis for the measurement of light field Tomo-PIV and light field SA-PIV.The velocity fields measured by the light field digital refocus PIV,the light field Tomo-PIV and the light field SA-PIV are compared and analyzed.Experimental results illustrated that for the measurement of laminar flow,the one-dimensional velocity field curves along the Y axis measured by the light field Tomo-PIV and light field SA-PIV are closer to the theoretical distribution than those along the Z axis.For the measurement of submerged water jet,the two-dimensional velocity field of submerged water jet measured by the light field Tomo-PIV and light field SA-PIV in XOY plane is very close to the velocity field results measured by the light field digital refocusing PIV.The reconstruction time of a 3D particle field using the light field SA-PIV is about 1020 seconds,40 times faster than the light field Tomo-PIV.Theses results verified the feasibility of the light field Tomo-PIV and the light field SA-PIV for the measurement of 3D flow field.