Reseach And Optimization Of Three-Dimensional Reconstruction Light-field Algorithm Based On Multi-projector Type Light-field Display System

The 3D light-field display technology based on multi-projector type light-filed display system is considered as one of the 3D technologies with great development potential because it can provide large area and large view of the naked eye 3D display effect.However,based on the current research status,it is still a great challenge to reconstruct dynamic light-field information in real time by using this system.To solve the above problems,this paper focuses on the multi-projector-type light-field display system to explore the principle of light-field reconstruction,optimization of lightfield reconstruction algorithm under sparse capture,optimization of lightfield reconstruction algorithm under free camera capture,and optimization of light-field reconstruction algorithm rate:1)Based on the basic theory of light-field,the principle of light-field reconstruction based on projector array is explored,and the light-field display and sampling system on this basis is built.The principle and effect of 3D display technology are analyzed systematically,and the advantages of light-field reconstruction system based on projector array are expounded.Combining the theory of light-field and the principle of human stereoscopic vision,the principle of the reconstruction system was deduced,and the light-field display system was built.According to the principle of light-field reconstruction,the capture requirements are determined,and the arrangement and position information of the capture system are designed.2)In order to solve the problem that the real capture camera cannot be arranged densely due to the influence of volume,this paper proposes a light-field reconstruction algorithm under sparse capture condition.In sparse capture,due to the loss of capture information,the relationship between display pixels and captured pixels cannot be established.In this paper,based on the Lambert body characteristics of the object,the target pixel matching algorithm is proposed.If the light emitted from the same object point is the same at all angles,the pixel values obtained by the adjacent camera sampling the object point are very similar.By comparing the most similar pixels of adjacent cameras,the algorithm obtains the object point imaging information under different depth information,synthesizes each display pixel with the imaging information,and accurately recovers the light field information under different depths in the original scene under the condition of sparse light field sampling.Finally,a virtual camera is used to simulate sparse sampling of real scenes,and the effectiveness of the algorithm is verified by experimental comparison.3)The light-field reconstruction algorithm of free camera is proposed.The real free camera is affected by the installation and assembly errors.Compared with the virtual camera,there are displacement and rotation deviations.At this time,it is impossible to obtain accurate camera lightcenter coordinates,main point coordinates and focal length information,so it is difficult to determine the actual imaging information of a point in the real world.In this paper,the camera calibration is used to obtain the internal participation and external parameters.Through the coordinate system transformation,the li ght-center,imaging surface and stagnation point pixel coordinates of each camera can be obtained.For the information of each object point with known depth information,the sampled ray can be determined by its connection with the light-center of the camera.The intersection point of the sampled ray and the sampled plane is the actual imaging information.Combining with sparse light-field coding algorithm,the transformation from two-dimensional parallax image shot by free camera to projection image array which can restore light-field information of real scene is completed.4)Rate optimization of light-field reconstruction algorithm is studied.In order to meet the requirement of real-time dynamic reconstruction of 3D light-field,CUDA computing platform is used to convert the process of successive calculation of each display column image in the traditional light-field reconstruction algorithm into parallel computation.During the design of the light-field reconstruction algorithm,the calculation of each display column pixel is independent of each other,which meets the conditions of parallel calculation.Through the design,the corresponding relationship between thread ID and display column pixel number is established.Each thread runs independently and synchronously,and the value of each display column pixel is calculated respectively.For a single frame projection image array with a horizontal resolution of 720*36,the synthesis time can be reduced to 19ms by using this acceleration algorithm.