A real-time image-based rendering and compression system with Kinect depth camera

This paper presents a novel Kinect-assisted real-time image-based rendering and compression system for high definition (HD) multiview video sequences. It is constructed from a Kinect depth camera and multiple stereo cameras and is intended for applications involving indoor scenes with static cameras. The information from the Kinect depth camera and background modeling can be utilized to segment the scene into foreground objects and the background in real-time. The segmentation and depth information are then encoded together with the texture information for real-time view synthesis at the decoder. The compression algorithm is developed around the AVS framework because of its simplicity and good performance. Novel features such as multiview coding and virtual view synthesis are incorporated, which are also applicable to other video coding algorithms. To achieve real-time operation, extensive GPU acceleration and optimization were performed. Experimental results were presented to illustrate the design and usefulness of the proposed system.