Real-time 3D video synthesis from binocular stereo camera

In this paper, an efficient method to synthesize 3D video in real-time from binocular stereo camera is presented. Our stereo camera consists of two pre-calibrated cameras with the optical axis perpendicular to the camera displacement. While much work has been done on stereo processing recently, most of them are focused on high-quality depth map generation, which are usually time-consuming and unsuitable for on-line processing. In our method, we apply a dynamic programming based stereo matching algorithm to compute the dense depth map on the fly. A fast depth image based rendering algorithm is subsequently developed to render several virtual views. Compared to other similar work, our method can synthesize 3D video of good performance in real-time on commodity PCs. Besides, benefiting from the well organized system architecture, our scheme is flexible for various 3D displays. Experimental results show that our method enables effective synthesis of 3D video in real-time.