Uncertainty Analysis And Three Dimensional Flame Reconstruction For Microlens Array Based Plenoptic Imaging

As an emerging imaging technology that can simultaneously detect multidimensional photothermal radiation signals,plenoptic imaging(PI)has the advantages of non-invasive,high spatial and temporal resolution,and visualization.In recent years,this technology has played an important role in many complex scene measurements such as high temperature flame monitoring,and three-dimensional(3D)flow velocity measurement.At present,plenoptic cameras are the most widely used plenoptic measurement sensors,which can flexibly sample and record four-dimensional(4D)spatio-angular information of light rays by using a built-in microlens array(MLA)in front of an image sensor.The high quality PI data is the foundation of the above measurement applications.However,the unavoidable errors of the MLA in actual systems would cause the PI uncertainty,thus significantly reducing the measurement accuracy and reconstruction efficiency of the target.Besides,the reconstruction strategy of the detected target is another important factor that affects the reconstruction performance of the PI measurement system.Especially for the temperature reconstruction of participatory media,its accuracy and complexity depend on the selection of the camera structure and reconstruction scheme.Therefore,it is urgent to study the PI characteristics and degradation mechanism caused by MLA errors,provide error calibration models and correction methods of the distorted light fields,and develop a fast and efficient strategy of the medium temperature reconstruction,which has important theoretical significance and application value for the development of complex target measurement technologies based on radiation PI.To realize these goals,this dissertation focuses on the uncertainty analysis and 3D flame reconstruction for MLA based PI.Firstly,the MLA error model and the PI model are built,and the evaluation scheme of the PI uncertainty is proposed.The PI uncertainty and influence mechanism of different MLA errors are analyzed.Then a local light-field correction method based on feature point extraction and a rotation error estimation and correction method are proposed to compensate the distorted light-field images caused by the MLA errors.On this basis,the PI characteristics of the flame apparent radiation in the cases of MLA errors are evaluated and corrected,and a flame reconstruction strategy based on the light-field refocused stack is developed,which can realize the rapid reconstruction of the 3D structure and temperature field of the gas flame with high resolution.The main contents include the following five aspects:The PI principle and the advantages of optical path structure of unfocused plenoptic camera are introduced systematically.An unfocused PI system model is established,and a super-resolution digital refocusing algorithm based on sub-aperture image is developed.The sources and features of MLA errors in PI system are summarized,and a basic error parameter characterization model is constructed.Two quality evaluation indices for the local subimage of the light-field image is proposed,and an evaluation scheme for PI uncertainty is further established.Consequently,a complete analysis framework form the PI process simulation to the light-field data post-processing and to the quantitative evaluation of light-field image distortion is formed.By using the PI model combined with MLA error model and the uncertainty evaluation scheme,the PI characteristics of each basic error parameter are investigated,including the subimage distortion features and intensity and structure distortion distributions of the light-field image.The degradation mechanism of the corresponding error images is also analyzed.Moreover,the light-field distortion distributions under the coupling of various form errors are discussed.According to the PI characteristics and degradation mechanism of MLA error,the study of the correction of distorted light fields is carried out.For the image distortions caused by different errors,a local correction method of distorted light fields based on feature point extraction,a subpixel-level optimization correction method and an estimation and correction method of MLA rotation error are proposed.Through the simulation experiments of light field correction,the applicable range and rectification accuracy of these methods under different error conditions are determined,and their effectiveness and feasibility are verified in the real scene light fields.Considering the PI uncertainty in the flame measurement due to the system errors,a flame radiation PI model is established.The PI characteristics of the flame apparent radiation under error conditions and the uncertainty transfer process of continuous layers along the radiation transfer direction are analyzed.The effects of the flame radiative properties(extinction coefficient,scattering albedo,and scattering phase function)on the PI uncertainty are discussed.Then the proposed correction methods are used to rectify and evaluate the distorted light field images of the flame with various radiative properties.For the unfocused PI system,the reconstruction strategy of the flame temperature field based on the light-field refocusing stack is proposed,in which Richardson-Lucy iterative algorithm is adopt to solve the image sequence of the flame refocus stack and the 3D temperature distribution is estimated by the relation between the image pixel intensity and radiation intensity.Through the simulation reconstruction on the 3D structure and temperature field of the laminar axisymmetric flame,the reasonable iteration number and the reliability of the restored image data are discussed,and the feasibility of the reconstruction method is also verified.