【作者】 张金玲；

【导师】 贾庆轩；

【作者基本信息】 北京邮电大学 ， 机械电子工程， 2009， 博士

【摘要】 空间机器人及其遥操作技术是目前主要航天大国研究的重点技术之一,它对提高国家在空间领域的竞争力和安全性具有重要的意义。空间机器人遥操作对信息反馈的实时性和准确性有着更为严格的要求,信息反馈的延迟或误差会影响操作人员的正确判断,导致控制失败,造成重大的经济损失。本文针对解决空间机器人遥操作中时延问题的预测显示控制方式,提出一种并行分布式的增强现实场景构建系统,构造空间舱内机器人遥操作的增强现实仿真场景。该场景构建系统通过构建高处理能力体系结构、设计快速高精度虚拟机器人注册算法以及多层次的虚实遮挡关系检测算法,实现虚拟机器人与空间舱内真实场景的全方位配准和可靠融合,并通过图像拼接技术和宽视场角显示系统对融合场景进行显示,为空间舱内机器人遥操作提供实时、准确、清晰的增强仿真信息。主要研究内容包括:第一,针对目前增强现实技术在处理性能上的不足,在对空间舱内机器人遥操作环境特点进行分析的基础上提出了一种面向空间舱内机器人遥操作预测仿真的并行分布式增强现实系统体系结构,并对并行结构所涉及的同步控制机制和通讯系统的实现方式等进行了深入研究。该并行分布式结构基于空间舱内环境固定可测的特点,通过多通道并行处理能够为操作人员提供实时、高分辨率、宽视场角、高沉浸感的增强场景预测立体显示。第二,基于直接线性变换(DLT)标定算法提出了一种快速精确的相机标定方法,并对其在增强现实虚实注册中的注册误差与Tsai两步标定算法进行了实验对比。该方法采用直接线性变换法进行摄像机标定和注册,不需要对拍摄图像进行畸变校正,可以提高系统的处理速度;同时,由于标定的测量体基本覆盖单个摄像机所对应的整个有效注册空间,因此即便采用不考虑畸变的DLT算法也可以获得较高的虚实对齐精度,能够为观察者提供可靠的相对位置关系。第三,针对目前虚实注册技术没有统一评价标准,对于不同类型的注册技术通常很难进行定量评估来判定技术优劣的问题,提出了一套虚实注册技术评价指标。此评价指标分为与注册质量相关和与系统需求相关两个方面,分别对注册技术的固定注册精度、注册空间大小、注册计算量、注册鲁棒性、硬件成本、前期工作量以及适用性等项目进行评测。这套评价指标可以用来对特定的虚实注册技术进行全面综合的评价。第四,提出了一种快速高精度多层次遮挡检测算法来实现增强现实系统中虚实投影间快速鲁棒的多层次遮挡检测。其中,基本双目立体匹配虚实遮挡算法通过比较虚实投影的相对位置来判定虚实空间点间的遮挡显示状态,不需要进行深度重构计算,可以提高遮挡检测效率;基于顶点显示状态在时间和空间上的连续性提出的状态拟合算法能够解决特征匹配失败顶点的状态检测问题;有效边点检测算法利用顶点邻接边上高可靠性边点代替低可靠性邻接顶点进行状态拟合运算能有效提高状态拟合算法的精度;最后,通过对检测出的显示状态进行整体连续性优化能够改善虚实融合图像的连续性效果。第五,本文采用基于图像绘制(IBR)的图像拼接技术来获得宽视场的增强仿真场景图像。针对目前的图像配准技术都是面向通用图像拼接,效率和精度都比较低,不能满足多路视频流之间实时精确的无缝拼接要求,提出了一种基于相机标定和相位相关相结合的图像配准拼接算法,利用相机标定信息和傅立叶变换的平移特性实现场景深度范围一定的多路视频流间实时高精度的无缝拼接。该配准算法利用精确的相机标定信息来实现图像间的旋转和缩放变换,对图像配准的准确度有很大改善。同时算法在预处理阶段就已经确定了图像的旋转缩放变换矩阵,实时处理阶段直接进行矩阵运算,避免了对数极坐标变换和反复的FFT正向、逆向操作,极大地提高了处理速度,使拼接算法具有非常好的实时性。最后,基于所设计的场景构建系统设计了简单的舱内机器人遥操作场景仿真实验原型。实验结果证明了系统设计的正确性和有效性。更多还原

【Abstract】 Now, space robotics and its tele-operation technology have become the most important research areas in many countries with advanced aerospace technology, and make great sense in enhancing country’s competition ability and security in space area. Because of the particularity and significance of its application areas, tele-operation of space robot has severe demand on the speed and accuracy of information’s feedback. The delay and error of information feedback can misguide operator’s judgment, and probably lead to control failure and great economy loss. Aiming at the time delay problem in space robot’s tele-operation, this research advanced a parallel and distributed augmented reality(AR) simulation scene construction system to construct augmented simulation scene for tele-operation of space robot in cabin real time and reliably. This scene construction system realized complete alignment and reliable fusion of virtual robot model and real scene of space cabin by constructing high performance architecture, using fast and precise calibration and registration method and designing multi-layer occlusion detecting algorithm. And at the same time, image mosaic technology and large field-of-view 3D display system are used too to realize large field-of-view and high immersion display of the fusion scene. Thus, real time, precise and immersive simulation information can be obtained for tele-operation of space robot in cabin. Main research contents include:Firstly, a parallel and distributed architecture of AR simulation scene construction is advanced base on that the robot tele-operation environment in space cabin is fixed and measurable. The related synchronization control mechanism and network communication techniques are researched. This kind of architecture can provide high performance process platform for reliable and immersive AR simulation scene construction.Secondly, full experiments and detailed analysis are made on registration error of AR for Direct Line Transformation(DLT) calibration algorithm and Tsai two steps calibration algorithm respectively, and a fast and accurate camera calibration method is put forward based on the experiment results. This calibration method realizes fast calibration and registration by using DLT algorithm which don’t need to do aberration correction, and realizes accurate registration by making the measurement volume occupy whole camera’s effective registration space.Thirdly, to resolve the problem that there isn’t any uniform evaluation standard of AR’s 3D registration technique and it’s difficult to make quantitative evaluation to compare the performance of different registration algorithms, a set of evaluation factors relative virtual-real registration are concluded and identified. These factors include stationary registration accuracy, registration space size, computation cost, algorithm’s robustness, hardware cost, preparation labor and system applicability etc. all these factors can be classified into to categories: one is concerning the registration quality, and the other is concerning the system’s requirements.Fourthly, a fast and accuracy multi-layer occlusion detecting algorithm is put forward to realize real time and robust multi-layer occlusion detection between virtual and real projections in augmented reality system. The basic occlusion detecting algorithm based on geometry stereo matching has low computation cost by avoiding deepness calculation. Based on display status’ spatio-temporal continuity, status combining algorithm is advanced to resolve occlusion detecting of points whose feature matching is unsuccessful. The effective edge points detecting algorithm is designed to increase the accuracy of status combining algorithm by using higher reliable edge points to do status combination. And finally, unitary continuity optimization is done on all points to improve the display quality of fusion image.Fifthly, large field-of-view AR simulation scene is obtained by image mosaic in Image-based Rendering technology. Aiming at the problem that current image alignment methods for normal images can not achieve speed and accuracy requirements for real-time and seamless mosaic among multiple video streams, an image mosaic solution based on camera calibration and phase correlation is put forward. Camera’s calibration information and FFT’s translation feature are used together to realize fast and seamless image mosaic among multiple video streams whose effective depth scale of scene is relatively small. This solution’s good accuracy comes from the precise calibration parameters, and the fast speed is obtained by avoiding repetitious FFT and IFFT computation.Finally, a simple experiment prototype of AR scene construction system for tele-operation of space robot in cabin is realized. The experiment results proved the correctness and validity of the system design.更多还原