| Autostereoscopic display provides an immersive 3D perception and a compelling sense of stereoscopic vision without the need of glasses or specific devices for viewers.As a hotspot research area in three-dimensional display technology,autostereoscopic display can be widely applied in virtual reality,telemedicine,entertainments and simulations.Dual-view 3D video files are used as a most popular method to store and exchange stereoscopic multimedia materials at present.This dissertation focuses on the tiled-projection autostereoscopic display technology of dualview 3D video files and the compression technology of 3D models,the main contributions are summarized as follows:(1)A real-time algorithm for the tiled-projection autostereoscopic display of dual-view 3D video files is presented.Based on the strategy of sort-last parallel rendering,the dual-view images are screen captured and compressed into JPEG files which will be distributed to all rendering clients.Then,a GPU-based single-pass rendering technique is utilized to generate the two single-view slanted stripe sub-images to enhance the autostereoscopic vision.Experimental results show that,the proposed algorithm can provide the glasses-free stereoscopic experience at 24 frames per second.Furthermore,it is verified that the autostereoscopic display reaches the optimum for young adult when the slanted degree of stripes is 10~°.(2)A novel end-to-end tiled-projection autostereoscopic display algorithm including dual-view video streaming acquistion,compression,transmission,and visulization is proposed.On the server side,the dual-view images are transcoded into H.264/MPEG-2 data by the Intel QuickSync Video engine,and distributed to all rendering clients through the real-time transport protocol.On the client side,the decoded images are rendered by interleaving,and displaying in a tiled-projection manner.Tests validate that the proposed method can process real-time 3D video captured by dual-cameras,not merely to 3D video files,affording us more capability and scalability.The self-calibrated method based on a slope voting policy designed to eliminate the dual-viewpoint perspective distortion so that it outweighs other approaches not only in convenience by in visual quality.(3)A progressive-interpolation-based biorthogonal Loop subdivision wavelet construction is presented to generate a new control mesh whose limit surface progressively interpolates all vertices in the original triangle mesh with arbitrary topology.The difference of subdivision surfaces between successive levels is slight and most of the wavelet coefficients fluctuate to zero in magnitude which makes it feasible to achieve high compression performance by utilizing the zero-tree encoding algorithm.The proposed wavelet transform achieves a higher compression ratio and better simplification quality,and decreases the Bits/vertex of 3D meshes about 14.05%,while improving the PSNR of the reconstruction model about 1.40%.(4)A progressive-interpolation-based biorthogonal Catmull-Clark subdivision wavelet construction is presented to generate a new control mesh whose limit surface progressively interpolates all vertices in the original quadrilateral mesh with arbitrary topology.It can achieve an exponential convergence rate and slight differences of subdivision surfaces between successive levels,which are suitable to data compression.Furthermore,two shape control parameters are introduced to custom its local orthogonalizing coefficients targeting on the vertex valences of quadrilateral patches,the novel scheme can greatly strengthen the control capability of the model's sharp features,the compression performance,and the shape preserving of the model at the lowest resolution level. |
PDF Full Text Request