Real-time High-quality 3D Content Generation Based On Double-sided DIBR

In recent years,3D display technology has been developing.As a new technology in the field of three-dimensional display,free stereoscopic display technology can enable the audience to see realistic three-dimensional images in front of the free stereoscopic display,obtain a real scene like visual experience,and can be viewed in a wide range,which has attracted extensive attention of experts and scholars and a large number of national investment,and has a very broad development prospect.Due to the large viewing Angle of free stereo display,the virtual viewpoint generated only by a single reference camera will generate many information holes that are difficult to fill.In the current research,how to generate 3D content with large viewing Angle and high resolution in real time has become a hot topic.In this thesis,based on the double-sided depth image rendering,the generation method of free stereo is studied.In order to solve the problems of voids and artifacts in virtual viewpoint rendering,the repair and optimization methods are proposed,and the algorithm is accelerated by GPU parallel processing.The main contents of this thesis and the innovative results obtained are as follows:(1)The reverse mapping process was improved to fill the small holes after the double-sided DIBR generated viewpoint picture.Before the reverse mapping,the depth map after the forward mapping is filtered repeatedly by means of multi-level means,and the filtering window is changed from large to small,so as to fill the small hole in the virtual viewpoint depth map.The padded depth map is re-projected to the reference image.Finally,the pixel value at the virtual viewpoint coordinate position corresponding to the hole is obtained by interpolation.(2)An image repair algorithm based on depth map is proposed.The larger holes generated during the DIBR process generally belong to the background region.Foreground pixels cause interference when repairing these holes,so this article use of depth image,separate the foreground and background,respectively,fill in background and outlook for the rest of the empty information,make the image quality has obvious promotion,image peak signal to noise ratio can reach more than 30 dB.(3)The improved algorithm is accelerated by GPU,and the real-time performance of the algorithm is evaluated and compared.The real-time performance of the algorithm is an important concern of this topic.In this thesis,the time complexity is reduced as much as possible and a more efficient algorithm program is adopted.The experimental results show that rendering an image of 1920×1080 takes less than 40ms,and the speed is increased by about 26 times compared with calculating with only CPU,basically reaching the level of real-time rendering.Finally,an interactive racing game is designed on the unity 3D rendering platform and runs on the free stereo display to realize the interaction between the operator and the game model.