【作者】 孙明竹；

【导师】 赵新；

【作者基本信息】 南开大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 微操作机器人是机器人技术向微细操作领域的延伸,它使微操作从繁复的手工操作中解放出来。随着微操作技术的发展和微观研究的深入,多目标操作和批量目标操作已逐渐成为微操作的研究重点。针对批量目标微操作存在的问题,本文将复杂作业的概念引入微操作领域,在原有面向生物工程的微操作机器人基础上,针对微操作复杂作业所需要的视觉反馈与定位问题进行了深入的研究。本文工作主要包括以下几个方面:显微镜狭小的视野范围给微操作复杂作业造成了诸多不便,是制约大范围微操作的主要因素,为此,本文提出了一种基于图像拼接的显微视野拓展方法。该方法结合基于区域和基于特征两类图像拼接方法,在保证拼接精度的前提下,可在线完成视野拓展。同时,本文借助不同放大倍数的显微镜物镜,建立分层递阶的体系结构,最终在线构建了具有厘米级范围、亚微米分辨率的全局视野,为微操作复杂作业提供了广阔的视觉反馈。在分析显微成像特征和显微成像模型的基础上,提出了一种适用于模糊显微图像的半盲图像复原方法。该方法将成像系统的点扩散函数建模为二维高斯函数,并将Canny边缘检测引入图像复原领域,最终实现了模糊显微图像的复原,为微操作复杂作业提供了清晰的视觉反馈。在现有目标定位方法基础上,提出了批量目标分层全局定位方法。该方法基于分层策略,将全局视野分块处理,有效缩短了目标定位的总体时间;同时,分层信息也可用于简化属于NP完全问题的目标排序问题,由此完成了微操作复杂作业的准备工作。针对批量生物目标微操作中特有的目标重定位问题,还提出了一种目标组重定位方法。该方法基于邻近目标构型,可在批量目标多次显微观察中找到同一目标,为进一步批量生物目标微操作的成功率与致死率的定量统计打下了基础。显微镜视野狭小表现在光轴方向就是景深小:只能在很小范围内成清晰的像,而离焦面只能得到模糊的显微图像。若以离焦显微图像作为视觉反馈,手工方式将无法完成微操作。但这种离焦显微图像本质上是一种“有规律”的模糊,只要理解人眼无法读懂之“规律”,便可解决微操作工具在离焦状态下的定位问题,微操作机器人系统在这种状态下的操作也成为了可能。本文在分析显微镜物镜成像原理的基础上,提出了一种基于离焦光学传递函数的深度方向定位方法。该方法在显微成像模型中加入了透镜参数与衍射效应的影响,具有较高的定位精度和较好的线性度。该方法实现了微操作工具的在线定位,可应用于高精度微操作复杂作业。最后,结合当前基因工程领域的前沿问题,设计并实现了针对批量生物目标的显微图像平面视野拓展与定位实验和基于离焦显微图像反馈的深度方向定位实验,由此验证了文中方法的有效性。更多还原

【Abstract】 Micro-manipulation system, which relieves operators from heavy and complicated manual operation, is the extension of robot technology to micro-manipulation field. With the development of micro-manipulation technology and research, multi-target operation and batch-target operation become the focus in micro-manipulation. In order to solve the problems in batch-target operation, this paper introduces the concept of complex task into micro-manipulation field, and does in-depth studies in visual feedback and location in complex task. The research contents are as follows.The narrow field of view (FOV) of microscope causes various difficulties in complex task in micro-manipulation. In other word, it is the major factor which limits huge-range micro-manipulation. A method to extend microscopic FOV based on image stitching is put forward in this paper. This method combining stitching methods based on image region and image features, can accomplish extending microscopic FOV online with high stitching precision. At the same time, hierarchical structure is utilized to establish the global FOV image with centimeter range and sub-micron resolution. The method provides wide visual feedback for complex task in micro-manipulation.Based on the analysis of microscopic imaging features and microscopic imaging model, a semi-blind image restoration approach is presented to restore blurred microscopic images in this paper. The point spread function of the imaging system is described as Gaussian function, and Canny edge detection is introduced into image restoration field. Results prove that this approach is effective and provides clear visual feedback for complex task in micro-manipulation.Based on the target location methods offered, a method on batch-target hierarchical global location is addressed in this paper. This method divides microscopic global FOV image into a series of sub-images based on hierarchical strategy, and shortens the whole time for target location. Moreover, hierarchical information can also be utilized to simplify the sort problem of the targets, which is one of NP-complete problems. This method completes the preparation for complex task in micro-manipulation. On the other hand, In order to solve re-location problem for batch biologic targets, a relocation method is put forward. Using configuration of a group of adjacent targets, this method can find the same target in each observation, and lays the foundation of quantitative statistics in micro-manipulation of batch biologic targets.In the direction of optical axis, microscope suffers from limited depth-of-field: clear image can only be obtained in small range around focusing plane, while in defocused plane, just blurred microscopic images will be obtained. Micro-manipulation can not be done manually with defocused visual feedback. However, the defocused microscopic image is a kind of regular blurring. If this rule is understood, the location problem of micro-manipulation tools can be solved, which makes defocused micro-manipulation possible. Depending on the analysis of microscopic imaging principle, a location approach in depth direction based on defocused optical transfer function is put forward in this paper. This approach has high location precision and good linearity, since lens parameters and diffraction effect have been taken account. This approach achieves on-line location of micro-manipulation tools, and can be applied in complex task in micro-manipulation with high precision.Finally, as for the modern problems in gene engineering, experiments, including plane FOV extension and location of batch biologic targets, and location in depth direction based on defocused microscopic images, are designed and accomplished in this paper. The experiment results demonstrate that all the methods in this paper are effective.更多还原