Towards 6DoF live video streaming system for immersive media

Based on the three rotational degrees (video in three dimensions, on the X, Y and Z axes) of freedom provided by VR, the viewer is free to control the viewing point and has six degrees of freedom (6DoF). When watching a sports game, the audience is no longer limited by the position of the camera, and can freely choose the viewing angle and position just like watching in the real world, which can greatly improve the immersion of viewing. However, the major barrier that prevents 6DoF video live from being industrialized lies in the extremely high computational complexity, of which multi-view depth estimation and Depth Image Based Rendering (DIBR) is difficult to realize. And existing devices do not have hardware interfaces that support multi-views coding technology. Therefore, we need new technologies for depth estimation and virtual view synthesis, and we need to use existing hardware coding/decoding interfaces to reduce power consumption. In this paper, we provide a 6DoF live video system, which includes multi-view depth estimation technique based on unsupervised learning, virtual viewpoint real-time rendering technology and 6DoF video coding. Experimental results demonstrate that our proposed acceleration method can speed up the original depth estimation algorithm by more than 34x, and can speed up the original DIBR algorithm by more than 168x. With our 6DoF video coding method, experimental results show that the bit rate achieves an average of 70%, 64%, 33%, 60% and 66% bitrate saving for AVC, HEVC, AV1, AVS3, VVC codec standard respectively.