大尺度时空视频数据表达及巡航无人机视频高效编码技术

64 Mbps is needed in order to transmit 4K HD videos. However, the transmission capability of UAV is 4 Mbps and decrease dramatically with the increase of transmission distance. UAV cruising business is different from regular video surveillance business. The cruising areas of UAV are fixed and the tracks of UAV are different for every single cruise. Landforms of large scale change slowly, while its illumination is complex. Unlike classic video encoding methods, which mainly concentrates on short-term local spatio-temporal redundancy, our project takes the video data repeatedly shot in the cruising area as the object of study and is based on the transform from local dynamic coordinates to global static coordinates. First, we study the generation mechanism of global redundancy according to the slowly changing property of landform and establish the illumination-dependent multi-modal representation model of long-term background. Then, targeting at the problem of foreground object removal for UAV videos, we introduce the nearest reference frame to develop the background restoration method based on the illumination compensation for reference frames. Finally, to solve the problem of synchronization of dictionaries for UAV and receiver, we study the multi-dictionary entanglement properties and introduce entanglement mechanism to realize zero flow synchronization update. Our work is expected to double the coding efficiency compared to the latest HEVC. Also, our project can promote the interdiscipline of related research fields such as video information, space information and knowledge representation.

4k高清图传码率高达64Mbps，无人机视频传输能力仅为4Mbps且性能随传输距离增大急剧下降。巡航无人机图传业务不同于常规视频监控业务，具有巡航区域固定但单次巡航轨迹各异，大面积地貌变化缓慢但光照环境复杂的特点。与经典视频编码方法主要面向短时局部时空冗余的研究思路不同，本项目以巡航区域内多次采集的视频数据为研究对象，以无人机视频从局部运动坐标系到全局基准坐标系的变换为基础，首先针对地貌变化缓慢的特点研究背景数据全局冗余产生的机理，构建具有光照依赖的长程背景多模态表达模型；其次针对无人机视频中前景移除问题，引入视频邻近参考帧实现基于参考帧光照补偿的背景修复方法。最后针对机-地背景字典更新同步问题，研究多字典纠缠特性的产生条件，引入纠缠机制实现多字典零流量同步更新方法。预期成果与最新的HEVC方法相比效率提高一倍以上，可探索、推动视频信息、空间信息和知识表达等领域技术的交叉与融合。

4k高清图传码率高达64Mbps，无人机视频传输能力仅为4Mbps且性能随传输距离增大急剧 下降。巡航无人机图传业务不同于常规视频监控业务，具有巡航区域固定但单次巡航轨迹各异 ，大面积地貌变化缓慢但光照环境复杂的特点。与经典视频编码方法主要面向短时局部时空冗余的研究思路不同，本项目以巡航区域内多次采集的视频数据为研究对象，以无人机视频从局部运动坐标系到全局基准坐标系的变换为基础，首先针对地貌变化缓慢的特点研究背景数据全局冗余产生的机理，构建具有光照依赖的长程背景多模态表达模型；其次针对无人机视频中前景移除问题，引入视频邻近参考帧实现基于参考帧光照补偿的背景修复方法。最后针对机-地背景字典更新同步问题，研究多字典纠缠特性的产生条件，引入纠缠机制实现多字典零流量同步更新方法。预期成果与最新的HEVC方法相比效率提高一倍以上，可探索、推动视频信息、空间信息和知识表达等领域技术的交叉与融合。