一种移动设备视频全景并行生成技术

Parallel generation for video panorama in mobile environment

通过移动摄像机合成视频全景图,由于视频中存在时间冗余以及现有合成技术对大角度旋转存在失配,提出一种鲁棒性较好的视频全景并行合成方法。该方法首先对选取的关键帧采用并行线程提取可重复性特征,并利用特征估计透视运动模型,从而对相邻关键帧进行全局配准;然后对重叠区域进行双线性融合,以减少视差对重叠场景的影响,同时并行优化提高了视频全景的合成速度。不同条件下移动合成实验表明,在大角度移动旋转以及视差等情况下仍能较好地合成视图。

Video panorama was generated by moving camera. Because of the temporal redundancy in video and misregistration in the large angle rotation, many existing algorithms perform poorly. To solve these challenging situation, this paper proposed an parallel synthesis approach to generating panorama which used texture feature to select key-frames. This method first extracted repeatable features from the selected key-frames using parallel threads, and then estimated the perspective motion model to align the adjacent key-frames by features. Employed bilinear blending to reduce the parallax effect in the overlapped region. Meanwhile, parallel threads programming improved the synthesis of speed. The experimental results show that the proposed method can effectively synthesize panorama with misregistration and parallax.