A Real-time Full-HD 2D-to-3D Conversion System Using Multicore Technology

The advancement of visual technology and the maturity of display process technique enable monitor industry to put forth multi-view stereo display systems, which allow consumers to watch 3D images with naked eyes. However, due to the insufficient 3D original videos, we are unable to achieve the end of enjoying 3D effect through stereo display. Therefore, this paper aims to provide a real-time algorithm for 2D-to-3D conversion, through which 2D video will be accompanied with a depth image so that one can obtain 3D images through auto stereoscopic display. The proposed algorithm focuses on the depth of background scenes and the active objects, while extracting depth information from the two for complete processing of depth generation. Since the production of depth information always requires heavy computation, we adapt VIA´s embedded systems with a 1.6GHz CPU, as well as nVIDIA CUDA (compute unified device architecture) technology which contains 256 core processor with the clock rate of each processor up to 950MHz. When accompanied with our proposed CUDA parallelization technique, the experimental results show an 83% of recognition rate for depth image, and a display rate of 32.52 frames per second in Full-HD format (1920×1088 pixels).