【作者】 丁卫；

【导师】 龚振邦； 谢少荣；

【作者基本信息】 上海大学 ， 机械电子工程， 2007， 博士

【摘要】 地面移动目标低空实时图像跟踪系统的研究是一个涉及多个学科具有重要意义的应用研究，主要是研究和开发一套基于超小型旋翼机的地面移动目标低空实时图像跟踪系统。该系统悬挂在超小型旋翼机上，从空中对地面目标实施监视与跟踪。跟踪系统能够锁定跟踪目标，排除干扰目标的影响，并且能够为飞行器的跟踪飞行提供控制参数。本论文围绕着地面移动目标低空实时图像跟踪系统所涉及的图像处理算法，系统硬件、软件及操作控制等多个方面进行研究。主要研究工作如下：1．分析地面移动目标低空实时图像跟踪系统的研究背景和意义，追踪国内外视觉监控方面的多个大型研究项目的进展。依据图像监控的基本框架，结合本系统应用的环境特征，分析得出构建基于超小型旋翼机的地面目标实时图像跟踪系统存在的几个问题，提出解决问题的房方案和技术路线。2．构建地面移动目标低空实时图像跟踪实验系统，并根据其优缺点给出了跟踪系统的硬件框架的最佳方案。跟踪系统采用并联控制回路，实现人工监视状态与自动跟踪状态无缝切换。此外，还搭建了一个跟踪系统的室内实验平台，模拟跟踪过程中飞行载体恶劣的运动状态。3．研究分析数字图像的阈值分割方法，根据跟踪系统实际应用环境的特征，采用基于直方图的自适应容忍度的多阈值分割方法对动态序列图像的运动区域进行分割，在直方图上，采用图像灰度值像素频率平均线限定图像上一些干扰子集的影响，并采用限制搜索范围的爬山算法来进一步确定每一帧图像上每个灰度子集的容忍度，确保图像分割的有效性和实时性，分割后可得到目标不完全分割的二值图像。4．针对分割得到的运动区域中常常包含有目标和干扰目标的问题，依据各个目标之间的相对聚集性，系统分别采用两种方法排除干扰目标。从数字图像处理的角度出发，采用数学形态学方法或者金字塔结构增加目标区域的连通性，然后采用数字图像处理的区域标识方法标识目标与干扰目标，通过判断靶框内标识区域的位置识别目标区域；从模式识别的角度出发，采用基于双重子窗口的动态聚类方法计算得到目标的形心位置，根据靶框内运动区域的面积设置一子窗口，通过子窗口聚类，最终搜索到目标的形心位置，从而排除了干扰目标的影响。动态聚类方法相对区域标识方法简单明了，如果要获得目标的形状，则需要进行后处理。5．针对目标的不完全分割对确定目标的形心位置可能存在影响的问题，系统根据形态学理论和计算几何的凸壳理论分别计算生成目标的凸壳，从而完整地分割目标。通过计算凸壳所围成的面积的形心位置，能够部分地消除了目标不完全分割的影响。其中形态学方法生成的是一个近似的凸壳，并且计算量大；计算几何方法生成的目标凸壳相对比较精确，且计算量较小。6．分析了跟踪系统云台的运动特性，实现了云台的跟踪运动与像平面上的目标跟踪的统一，讨论了云台运动的控制方式——速度控制和位置控制的特性及应用环境，并根据跟踪过程中云台相对于飞行器的姿态为飞行器的飞行控制提供必要的控制参数。7．从人机工程学的角度出发，充分考虑人的主观因素，设计良好的人机交互界面，将跟踪系统绝大多数的功能操作都集中在跟踪视野范围内的鼠标上，使得操作人员能够集中注意力，简单快捷地捕捉跟踪目标，合理的操作设置能够减轻操作人员的劳动强度。通过以上的研究工作，为构建一套实用化的图像跟踪系统提供了必要的理论基础和可行性验证。更多还原

【Abstract】 The study of the real time surveillance and tracking system of the object moving on the ground from low altitude is an important application research related to multi-subject. The primary purpose of study is to develop an applied surveillance and tracking system based on a small helicopter. The system, hanged under the helicopter, is used to watch and track the object moving on the ground. It can lock the tracked object steadily through eliminate the disturbance of other objects. And it also can provide the control parameter for the helicopter flying after the object. The thesis discussed some correlative key techniques about the surveillance and tracking system, such as the algorithm of image processing, hardware, software, control operation and etc. The main contents list as follows:1. Analyzed the background and significance of the research about the surveillance and tracking system, and traced some international research projects on surveillance system. According to the general frame and the applied conditions of the surveillance system, the reasonable approaches to construct the tracking system were presented.2. Chosen the best hardware architecture to construct an experiment system for surveillance and tracking. With parallel connection control loop, the system can switch smoothly between the manual surveillance and the automatic tracking. Besides, constructed an inside experiment platform, carried the surveillance and tracking system, to simulate the worst movement of the helicopter in the tracking process.3. Researched the threshold segmentation on digital image. The multi-threshold segmentation method with adaptive tolerance based on histogram is used to segment the motion region on the dynamic image. Some disturbance subsets are confined by the pixel frequency average of gray value. The tolerance of gray subset can be calculated by the hill-climbing search algorithm with confined scope. The image can be segmented into a binary image with an incomplete object, and the segmentation is validity and real time.4. Aim at the disturbance of other objects, there are two methods to remove these objects in the motion region. From the point of view of digital image processing, the connectivity of the object can be enhanced by pyramid data structure. Each region is labeled with a unique number by scan the image once. The object region can be identified based on its position in the subwindow. From the point of view of pattern recognition, the center of gravity of the tracked object can be located by dynamic cluster in the double sub-window. The method of dynamic cluster is simpler than the method of region label, but can not identify the object’s boundary unless post processing.5. Aim at the incomplete segmentation of the object, there two methods, based on the mathematical morphology and the computational geometry, to make the convex hull of the object to segment the object completely, and replace the center of gravity of the object pixels by the center of gravity of the region enveloped in hull. The convex hull made by morphology is approximate, and the computational cost is too high. The computational geometry can make a precise convex hull, and its cost is low.6. Analyzed the movement characters of the pan and tilt head. The movement of the pan and tilt head is correspond with the object tracking on the image plane. Two control modes of the movement of the pan and tilt head, speed control and position control, are discussed, and the control parameters of the aircraft’s flight control system can be calculated based on the pose of the pan and tilt head relate to aircraft during the tracking process.7. Considered man’s subjective factor, a friendly human computer interface was designed from the point of view of man-machine engineering. All the key operations are gathered on the mouse in the tracking view so that the operator can concentrate on the tracking view and hit the object easily. The reasonable setup of operation can decrease labor intensity.The thesis presented the theoretical foundation and validated the feasibility of the surveillance and tracking system.更多还原