The Key Technology Of Computaional Three Dimensional Display With Dense Viewpoints

Human's eyes are very sophisticated photosensitive devices. They can not only get high dynamic color information, but also feel the spatial information. Binocular vision is an important way for us to understand the three-dimensional world. However, traditional methods of capturing and displaying, such as digital camera and LCD monitor, can only record and transmit two-dimensional images. In this way, eyes only see a plane at different angles, different from the experience of the three-dimensional world. Thus, understanding and perception of the content are reduced. At present, more and more fields, such as particle simulation, architectural design, life science and so on, have a strong demand for the capturing and displaying of realistic 3D information.In recent years, multi-view captruing technology and 3D display technology have an impressived development. However, due to the lack of viewpoints, the amount of information is not enough, which leads to the fatigue of human eye observation, and the poor quality of three-dimensional display. In order to solve these problems, the technology of dense views interpolation and computational light field display is researched in this paper.The main contents and innovations of this paper are as follows:(1) A method of viewpoint rendering based on image matchingAt present, binocular capture method is widely used. However, it will result jumps between viewpoints, flase perspective relationship and low vision experience. Addressing these problems, this paper proposes a method of viewpoint interpolation based on sparse feature points. The feature points of the sparse matching are triangulated and form the structure of the left and right images. The experiment results show that the viewpoint quality is good and the robustness is high.In addition, this paper also presents a method of viewpoint interpolation based on dense disparity grid. The dense matching depth results are organized into a regular grid structure, and the virtual viewpoint can be generated by rendering. In the implementation, we will generate different views, and display them on the three-dimensional display in realtime. The experiment results show that this method is very fast and can be processed in real time. The speed can be 40fps, which is very suitable for multi-view three-dimensional display.(2) A method of 3D reconstruction based on depth capturing with horizontal railSince most of the indoor modeling methods using Kinect only deal with the three-dimensional data within the depth range. Such content out of depth range will result in black and empty background. In order to carry out a better 3D display, this paper presents a system of 3D reconstruction and 3D display based on a horizontal capture rail with Kinect. Color and depth information is collected from different angles by the Kinect. The surface of the object is reconstructed with three-dimensional points, and the texture of the background is calculated by the method of triangulation.After the model is obtained, virtual parallax viewpoints are resampled and displayed on an autostereoscopic display. Moreover, zero plane can be adjusted to change the viewing experience.(3) A fast view synthesis method based on texture arrayDue to the large amount of computation cost in the synthesis of parallax images, it is difficult to carry out the operation on CPU in real-time. According to the characteristic of viewpoints synthesizing, this paper puts forward a fast view synthesis method based on texture array with D3D10. The experiment results show that this method can generate 3D display images in real time. Since the number of views is increased, the changes of time cost is not obvious due to the excellent characteristics of the texture array.(4) An improved halftoning method for autostereoscopic display based on float grid-division multiplexingWhen making the three-dimensional display of printed materials, the relationship between lenticular-lens array and printer should betaken into consideration. When the array is not matched with the printing dot, there will be a serious dislocation which results in the display quality deterioration. In this paper, we propose an improved halftoning method for autostereoscopic display based on float grid-division multiplexing, which can express pixels with partial dots. This method can be used to align the lens array with printer dots, and can reuse more views. Experiments and analysis show that, we can eliminate the impact of dislocation and improve the quality of the image. In addition, under the request of vis ion quality, we can add more views to enhance the smoothness between views.(5) Performance improvement of compressive light field display with the viewing-position-dependent weight distributionThere are several limitations of the compressed light field, such as poor image quality and limited field of view (FOV). This paper proposes a method of compressive light field with the viewing-position-dependent weight distribution. By determining the position of the viewer, a correlation weight distribution function can be established. And when relevant views are given high weights in the optimization, the displaying performance at the viewing-position can be noticeably improved. Simulation and experimental results demonstrate the effectiveness ofthe proposed method.The PSNR of narrow FOV can be effectively increased with 7dB. Wide FOV can be extended to 70° x 60°. Four viewers can be supported in wide FOV.