基于视觉传感网络的机器人环境感知与导航控制

CASIA OpenIR > 毕业生 > 博士学位论文

	基于视觉传感网络的机器人环境感知与导航控制
	袁文博
	2016-05-25
学位类型	工学博士
中文摘要	视觉传感网络辅助的机器人系统，可以更好地实现对环境的感知，进而提升机器人任务执行的质量，已成为移动机器人领域的研究热点，具有重要的研究价值和广阔的应用前景。本文针对基于视觉传感网络的机器人环境感知与导航控制开展研究，论文的主要内容如下：首先，介绍了视觉传感网络辅助的机器人系统的研究背景与研究意义，对基于视觉的目标识别与跟踪、移动机器人导航进行了综述，阐述了视觉传感网络及其辅助的机器人系统的研究现状，并对论文内容和结构做了介绍。其次，基于视觉传感网络开展了目标识别的研究。针对环境中的静态目标，设计了一种基于线段本体特征和线段间相对关系的推理模型，以对线段生成模块输出的线段集进行推理，确定出符合指定目标描述的线段集进而完成对目标的识别。此外，考虑场景切换等因素引起的动态目标再识别问题，在目标区域分割的基础上，设计了特征描述子对其中的关键区域进行颜色和形体信息的提取，通过判断目标间的相似度实现再识别。第三，提出了一种大尺度场景下多目标跟踪的方法。通过对粒子滤波方法进行改进，设计时空动态模型和表现模型以实现高质量的多目标跟踪。同时，为应对传统的马尔可夫链蒙特卡洛(MCMC)数据关联的跟踪方法大都需要事先已知目标最大速度的问题，提出一种改进MCMC数据关联的多目标跟踪方法，利用解空间压缩和提议分布优化提高了解空间的搜索效率。第四，基于视觉传感网络，开展了机器人导航研究。面向构建拓扑图的环境，提出了一种包含主节点及其附属隐节点的拓扑地图构建方法，以此为基础实现机器人的导航；面向无拓扑图的环境，设计了一种基于多模式RANSAC的轨迹划分算法用于获取当前运动模式的轨迹点，利用姿态估计模型以估计机器人的姿态，进而实现对机器人的导航指引。第五，提出一种视觉传感网络框架下基于局部关键信息的机器人导航控制方法。机器人借助激光传感器对环境中的关键特征进行提取，并改进ICP算法以更好地对关键特征进行配准，进而实现监控网络盲区或需求更高精度等情形下的机器人位姿估计，在此基础上，结合基于局部感知环境分区评价的机器人运动控制方法，最终实现机器人的导航控制。最后，对本文工作进行了总结，并指出了需要进一步开展的研究工作。
英文摘要	Robotic system aided by visual sensor network has become a research hot in mobile robot domain because it can achieve better environmental perception and higher task execution quality. It is significant with a broad applications. This thesis has conducted the research on environmental perception and navigation control of robot based on visual sensor network. The main contents are as follows: Firstly, the research background and significance of the robotic system aided by visual sensor network are given. The survey of vision-based object recognition and tracking as well as mobile robot navigation is presented. The research development of visual sensor network and the robotic system under its framework is then described. The contents and structure of this thesis are also introduced. Secondly, the object recognition has been conducted on the basis of visual sensor network. For static objects in the environment, an inference model based on ontology characteristics of line segments as well as their relative relationship has been designed. This model is adopted to analyze the line segment set which is generated through the line segment generating module, and thus extract the effective line segments which are matched with given objects. And then, the static objects are recognized. Moerver, considering the moving object re-identification caused by scene switching, after region segmentation for objects, feature descriptors are designed to extract color and shape information of key regions. By evaluating the similarity among the objects, the re-identification is achieved. Thirdly, the methodology of multiple objects tracking at large-scale scene is proposed. Spatio-temporal dynamic model and appearance model have been presented to improve the particle filter method for better tracking. In addition, traditional Markov chain Monte Carlo data association methods require a priori knowledge about the maximum speeds of targets. In order to solve this problem, an improved approach is proposed by compressing solution space and optimizing proposal distribution, which enhances the searching efficiency of solution space. Fourthly, the robot navigation is conducted based on visual sensor network. On the one hand, a topological mapping method for topology environment is presented, and the map contains some host nodes and accessory hidden nodes. This map provides the basis of robot navigation. On the other hand, for environments without topological nodes, a multi-mode RANSAC algorithm for trajectory division is designed to acquire the trajectory points of current motion mode. On this basis, an attitude estimation model is utilized to estimate the robot attitude for the guidance to the robot. Fifthly, a robot navigation control method based on local key information under the framework of visual sensor network is given. Considering the cases where there exists monitoring dead zones for visual sensor network or a higher precision of robot pose is required, the laser sensor is used to extract the key features of the environment, and an improved ICP algorithm is introduced for better registration of key features. The robot pose estimation is then accomplished. Furthmore, a motion control method based on sub-regions evalution in local sensing environment is also presented. Finally, the naviation control of robot is implemented. Finally, the conclusions are given and future research is addressed.
关键词	视觉传感网络目标识别多目标跟踪移动机器人导航控制
文献类型	学位论文
条目标识符	http://ir.ia.ac.cn/handle/173211/11727
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	袁文博. 基于视觉传感网络的机器人环境感知与导航控制[D]. 北京. 中国科学院研究生院,2016.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
基于视觉传感网络的机器人环境感知与导航控（4930KB）	学位论文		限制开放	CC BY-NC-SA