面向对象和手术器械光学定位的三维医疗视频编码与视点合成技术研究

The remote surgery guidance and teaching are the effective and important means for the solution of the uneven distribution of medical resources and the improvement of the medical service ability of medical staff in the undeveloped areas. As the development of the techniques of tri-play network, three dimensional and the future medical robot, it is possible to carry out a full range of medical guidance and instruction at any time and place for the medical staff. Considering the problems of short shooting distance, limited operative field and occlusion of surgical instruments, which cause the true depth information of the tissue or organ is hard to capture and the coding efficiency is low, we propose the 3D medical video coding and virtual view synthesis techniques that based on the object-oriented and optical positioning of surgical instruments. This project is going to research the edge extraction and depth generation of medical video based on the multi-scale singularity detection and stereo matching with variable windows; to research the high efficiency of 3D medical video coding with the block level adaptive boundary dependent transformation kernel coding and the DMM model with the help of real edge; to research the 3D positioning of optical surgical instruments by using the marker matching and apparatus rotation method; to research the hole filling method with optical positioning and linear auto regressive super resolution model with distance probability. Our research provides the technical supports for the remote 3D surgical video displaying and consultation, remote surgical or clinical teaching, surgical robotics development and application.

远端手术指导与示教是解决我国医疗资源分布不均、提高欠发达地区医疗人员业务水平的有效手段。随着三网融合、三维技术的发展及未来医疗机器人的研发，医疗人员在任何时间和地点进行全方位医疗指导和操作成为可能。面对医疗视频拍摄距离短、术野有限、手术器械多、出血干扰等导致的手术器械与组织器官真实位置难以捕获、压缩效率不高等问题，本项目研究面向对象和光学手术器械定位的三维医疗视频编码和视点合成技术。通过多尺度奇异性检测和可伸缩窗立体匹配，研究医疗视频中对象的边缘提取和深度生成；通过块级自适应边界依赖的变换核编码及基于真实边缘的DMM模式，研究三维医疗视频的高效编码；通过标记物匹配和旋转器械法，研究光学手术器械空间定位；通过光学定位的空洞填充和结合距离概率的线性自回归超分辨率模型的建立，研究虚拟医疗视点的合成。本项目的开展，为我国在未来远程手术指导与示教、远程医疗教学、医疗机器人的研发与应用提供了技术储备。

远端手术指导与示教是解决我国医疗资源分布不均、提高欠发达地区医疗人员业务水平的有效手段。随着互联网行业的不断发展，医疗人员通过电视、电脑、手机等媒介进行远端手术指导成为可能。面对医疗视频拍摄距离短、术野有限、手术器械多、出血干扰等导致的手术器械与组织器官真实位置难以捕获、手术视频编解码速度要求高等问题，本项目开展了面向对象和光学手术器械定位的三维医疗视频编码和视点合成技术研究。通过近红外光学定位和可伸缩窗立体匹配，研究了医疗视频中多种对象的边缘获取和深度生成；通过视频纹理分析和物体边界角点特征检测，研究了基于特征点检测的快速编码单元尺寸预选帧内快速编码算法；通过形状识别和运动矢量跟踪，研究了带光学标记物的多手术器械准确识别和实时追踪算法；通过手术场景左右视图不同对象的真实光学定位点和视觉注意度模型，研究了光学定位技术辅助下的手术场景全景拼接算法；通过可伸缩窗立体匹配和结合距离概率的线性自回归超分辨率模型的建立，研究了手术视频虚拟视点的合成。本项目面向医学、计算机科学与生物医学工程学交叉融合研究热点，面向国家远程医疗、手术导航和手术机器人研发需求，开展了相关研究。为手术场景的快速编码与三维呈现、为术中机体组织及器械的准确定位提供了技术储备，为我国在远程手术指导与示教、医疗机器人的研发与应用提供了新思路。