Autonomous Production of Personalized Videos for Heterogeneous Terminal Devices

多视点信息融合不仅能够实现准确的视频分析(如对象跟踪和场景理解)，而且还为内容呈现提供了高度的灵活性。以体育视频转播和智能视频监控为例，通过从语义故事组织、时间摘要和空间摄像规划三个不同的层次考虑内容混合，讨论了如何制作个性化视频以更好地呈现在各种终端设备上。在将跟踪对象划分为相互交互的组后，通过在交互组之间移动浏览目标来组织故事。然后，故事通过自适应快进和虚拟缩放来呈现，以更好地满足用户偏好和设备限制定义的条件。申请专利1项，发表期刊综述论文5篇，辅导论文1篇，国际会议论文11篇，参展2次。

Multi-view information fusion not only enables accurate video analysis (e.g. object tracking and scene understanding), but also provides high flexibility to content presentation. Given sport video broadcasting and intelligent video surveillance as two use cases, we discussed the production of personalized videos for better presentation on various terminal devices, by considering content remixing from three different levels, i.e. semantic story organization, temporal summarization and spatial camerawork planning. After dividing tracked objects into mutually interactive groups, we organize the story by shifting the browsing target between interactive groups. The story is then rendered with adaptive fast-forwarding and virtual zooming to better satisfy the conditions defined by users preferences and device limitations. We have applied for 1 patent, published 5 reviewed journal papers, 1 tutorial paper, and 11 reviewed international conference papers, and attended two exhibitions.