Algorithm Research And Implementation Of Video 2D-to-3D Based On FPGA

With the development of video display technology,3D display technology is widely recognized by people because of its realistic 3D effect, and develop rapidly in field of television entertainment, medical care, education, and so on. There are two main forms for 3D content to obtain:Multi-camera acquisition and 2D to 3D conversion. The target image collected by multi-camera is limited because of the great difficulty and much cost, and large bandwidth it needed for the transmission. But on the contrary, the 3D conversion technology needs less bandwidth in the process of transfer because only reference image and its depth image is needed. And it is possible to convert and display in real time for HD video with the development of hardware and continuous improvement of conversion algorithm.This thesis designs a real-time 2D to 3D conversion system based on DE4. The system collect HD video by D5M camera, then the video will be processed by Altera Stratix IV and stored by DDR2 SDRAM data storage module, and at last it will be displayed through DVI interface in real-time. The algorithm used in each step are verified by Matlab, then constructed by Verilog language, at last all modules are finally debugging and implementation after downloaded to FPGA development board.The thesis first introduces 3D display technology and basic principle of 2D to 3D conversion. And several core algorithm in 3D conversion are researched:the depth information extraction, DIBR virtual perspective drawing and hole filling algorithm. In terms of depth information extraction, the depth extraction algorithm based on relative height depth and VIBE algorithm are introduced. Then the depth extraction algorithm based on relative height depth is improved. The thesis introduces a DIBR technology without camera calibration parameters at step of virtual perspective drawing, before that the cause of matching errors and its solution will be given first. However, there are holes on target image after drawing. Then the thesis first give the reason for the formation of holes, then several popular hole-filling algorithm are given. At last, this thesis gives a hole-filling algorithm based on gradient transition between foreground and background, which improved the visual effect of 3D image.