《自主机器人视觉信息处理与跟踪导航研究》由会员分享，可在线阅读，更多相关《自主机器人视觉信息处理与跟踪导航研究（83页珍藏版）》请在金锄头文库上搜索。
1、I 摘 要 自主移动机器人导航技术是智能机器人领域的一个重要研究方向， 其中视觉导航具有其它传感器导航方式所无法比拟的优点， 是自主移动机器人的关键技术和研究热点。本文从 Prof. Marr 视觉理论出发，对机器人低层视觉图像处理算法、视觉系统建模以及实时动态环境下的运动目标跟踪与导航进行了深入研究，并在 AS-R 自主机器人开发平台上进行了实验论证。 低层视觉图像处理是视觉导航的基础，针对环境图像存在噪声干扰的特点，文章分别从时域和频域上分析了图像滤波的方法。 在研究了经典边缘检测算子的基础上，介绍了基于小波变换边缘检测的一般方法，研究了一种自适应小波变换的图像边缘检测方法，获得了更为细致
2、准确的边缘。 在研究了自主机器人垂直异构双目视觉系统结构的基础上， 根据摄像机透视原理，对机器人视觉导航进行了建模，推导了世界坐标系、摄像机坐标系以及图像坐标系之间的映射关系。 分析了传统的摄像机标定技术和基于主动视觉的摄像机自标定技术。 本文设计了机器人运动目标跟踪导航的整体框架， 分别研究了基于自适应背景差分的运动目标跟踪导航与基于卡尔曼滤波与预测的跟踪导航。 前者采用帧间差分技术与动态背景更新模型识别运动目标， 后者采用不变矩理论与卡尔曼滤波相结合的方法进行运动目标检测与识别。卡尔曼滤波技术的引入，有效实现了目标特征点的位置预测及目标相对运动速度的获取。 相关算法研究在 AS-R 自主机
3、器人平台上进行了实验论证，从实时性出发，研究了基于颜色识别的双目导航系统，更趋于一般性的，研究了基于卡尔曼滤波的 AS-R 跟踪导航技术，从物体形状的角度，实现了运动目标的跟踪与导航。考虑到机器人跟踪与导航的实时性要求，本文选用了合适的预处理方法，基于图像块的动态窗口分割方法，减少了图像数据的计算量，降低了目标误匹配率，能较好地满足实时动态环境下的机器人跟踪与导航。实验表明 AS-R 机器人能实时、准确和稳定地识别并跟踪运动目标，实现定位与导航。 在路径规划部分，本文采用了一种新的遗传算法，小生境伪并行遗传算法用于机器人路径优化。实验结果表明该方法比 SGA 和并行遗传算法具备更快的收II 敛
4、性，能有效避免早熟现象的发生。文章最后从 AS-R 机器人模块化设计思想出发，探讨了一种基于分层递阶控制的多处理器机器人结构设计方案。 关键词：关键词：全自主移动机器人，运动目标跟踪，视觉导航，垂直异构双目视觉、卡尔曼滤波、小波边缘检测 III ABSTRACT Autonomous mobile robot navigation is one of the most important research branches of intelligent robot. Vision-based navigation is the pivotal technology and research f
5、ocus because of its overwhelming advantages than other type of sensors navigation. This thesis contains the following research based on professor Marrs vision theory. Firstly, robot low level image processing algorithms had been studied. Secondly, the vision system had been modeled. Thirdly, movemen
6、t detection and vision-based navigation had been studied under the condition of dynamically real-time background. Low level image processing is the base of navigation. The paper discusses time domain filtering and frequency domain filtering according to the noise disturbance in the surroundings imag
7、es. Then, classic edge detectors are introduced. Base on the analysis of wavelet edge detection algorithm, a type of self-adaptive wavelet edge detection is proposed to obtain much more clear and precise image edge. System structure of vertical iso-vision system with two CCD cameras is introduced in
8、 this paper. Mathematical modelling of robot vision navigation has been researched according to the camera perspective theory. The author discussed the mapping relation among the world coordinate, camera coordinate and image coordinate. Camera calibration is studied while the active vision-based cam
9、era self- calibration is introduced as well. The thesis totally introduces the whole structure of robot movement detection and navigation. Techniques such as self-adaptive background difference and Kalmann filtering and prediction are studied respectively. The former one applies difference algorithm
10、s between consecutive frames and dynamic background refreshing model to recognize moving objective while the later one combines invariant moment theory and Kalmann filtering together to realize movement detection and navigation. It is given to obtain both the position coordinate and relative velocit
11、y vector of the moving objective feature point effectively when Kalmann filtering algorithm is used. Corresponding algorithms are demonstrated by experiment using AS-R IV autonomous mobile robot. Robot navigation with two CCD cameras based on color segmentation is used to meet the real time requirem
12、ent. As for the purpose of universal application in movement detection and navigation, a Kalmann filtering-based technology is used in AS-R vision system to realize the moving objective recognition and robot navigation shaping from its figure. Considering the real time qualification of robot navigat
13、ion, reasonable image processing algorithms are chosen. Dynamic block-based image segmentation is used to reduce the calculation complexity. It is a betterment to satisfy the qualification of movement detection and navigation under the condition of dynamically real-time backgrounds. Experimental res
14、ults demonstrate that AS-R autonomous mobile robot vision system can recognize and trace the moving objective precisely and stably. It meets the real time qualification during the process of robot orientation and navigation using these algorithms above. In the part of path planning, a new genetic al
15、gorithm named niche pseudo parallel genetic algorithm is proposed to optimize the robot paths. Experimental result displays that NPPGA has better performance than SGA and DPGA as far as convergence is concerned. It avoids premature effectively. Last but not the least, autonomous robot structure base
16、d on multilevel hierarchical control is designed based on the AS-R module architecture. KEYWORDS: Autonomous mobile robot; Movement detection; Vision-based navigation; Vertical iso-vision; Kalmann filtering; Wavelet edge detection 硕士学位论文 1 第一章 绪论 1.1 引言引言 机器人视觉导航12以机器视觉的发展为前提。 机器视觉34是近几十年来发展的一门新兴技术，它是计算机科学、自动化技术和人工智能领域的重要分支。人类感知外部世界信息主要是通过视觉，统计表明，80%以上的外界信息是由视觉获取的。因此，对于智能机器来说，赋予机器以人类视觉功能对发展智能机器是及其重要的。 机器视觉主要借助各类成像系统，如单目视觉系统5、双目视觉系统6、立体视觉系统7等，以此来代替机器视觉器官作为图像采集与输入方式，并通过高速处理器进行图像
