| With the development of science and technology and the progress of the times,people began to pursue video information with better quality and better experience,and 3D light field display as a stereoscopic display technology has naturally attracted widespread attention.In recent years,the 3D light field display technology has become more and more mature,and can support more and richer display content.The traditional 3D light field display uses virtual model modeling content for display,and has achieved good results.It can achieve large 3D display with large size,large viewing angle,high resolution and high dynamics.However,the 3D light field display of the real scene greatly complements the display content and makes the display content more flexible and changeable.Therefore,this thesis studies the high-definition transmission method for real-time 3D light field display,designs and builds a transmission system,and successfully verifies that the system works normally under 3-channel 8K video streams and correctly displays 3D light field;at the same time,researches and proposes a 3D light field Three noise models in the process of field information collection and transmission,and noise removal through deep neural network to optimize the final 3D light field display effect.The research works and results of this thesis are summarized as follows:1.Research and propose a high-definition video transmission method for 3D light field.The acquisition,transmission and display of 3D light field video information are studied,and the workflow is determined.Aiming at the multi-viewpoint images at the acquisition end,a multiviewpoint image calibration method based on reprojection is proposed to simulate off-axis shooting,correct camera distortion to meet the display requirements,and write it into a visual desktop application.2.Based on the workflow of the transmission system,three noise models are proposed for the noise in the camera,between cameras and in the transmission process.Use a residual network based on skip connections to remove the influence of noise in the camera;use a fully connected network to remove the influence of noise between cameras;use an autoencoder network to remove the influence of noise during transmission;after these processes,the three-dimensional display image is clearer.A high-bit-rate-low-bit-rate video frame dataset for training was also produced,enabling the autoencoder network to process noise during transmission more effectively.3.Build and verify the real-time 3D light field high-definition video transmission system.A system based on a single video stream is built to verify the feasibility of the transmission method,and the system is improved so that it can process multiple 8K ultra-high-definition video streams.The synchronization mechanism of multi-channel video streams is designed to enable the system to process multiple video streams synchronously,and a multi-channel-based 3D light field player is written,and the multi-view encoding method based on multi-channel video is accelerated by CUDA to realize real-time processing.After the above measures are processed,the scene collected in real time is correctly displayed on the three-dimensional light field display.The above researches and achievements of this thesis have enriched the implementation technology of high-definition video stream transmission and multi-viewpoint image denoising for 3D light field display,and provided a good reference for subsequent research.Further development and application of these technologies will contribute to 3D light field In-depth research and development in the fields of display,video transmission,and image denoising play an important role in promoting. |
PDF Full Text Request