【作者】 李新德；

【导师】 黄心汉；

【作者基本信息】 华中科技大学 ， 控制理论与控制工程， 2007， 博士

【摘要】 信息融合技术是上个世纪70年代末发展而来的一门新学科,它最早是从军事上的C3I(Command, Control, Communication and Intelligence)和IW(Information Warfare)系统上发展而来的,但随着科学技术的飞速发展,特别是近十几年,信息融合技术由军事应用向民用迅速转化,其应用领域不断扩大。随着人类对未知领域认识的不断深入,对识别精度,准确度,融合控制的鲁棒性和实时性,融合模型建立的合理性,以及融合管理的高效性的要求不断提高。但由于传感设备的物理局限性,系统运行的不确定性,环境的未知动态干扰等导致的不完善信息融合问题日益成为信息融合技术研究的一大挑战。信息融合理论方法的研究是解决多源不完善信息融合问题的有效途径,尽管近20年中提出很多信息融合的理论和方法,有助于不完善信息融合特定问题的解决,但实践证明现有的理论和方法都有一定的局限性,使其应用受到很大的限制。因此迫切需要提出一种广适方法来解决不完善信息的定量和定性融合问题。本文以DSmT(Dezert-Smarandache Theory)和中智理论为框架,以移动机器人在未知环境自主创建地图为背景,对多源不完善信息的广义融合展开了深入的研究。本文从信息的不精确定量融合的角度研究出发,将模糊理论、中智理论与信度赋值技术进行关联,进一步扩展了信度赋值技术的组合规则,提高了广义融合机融合算子组合证据源的能力,扩大了不完善信息的广义融合范围。在广义框架下,深入地研究信息的不确定定性融合问题,在改善定性运算算子的基础上,提出细化定性信度语义标签的思想,给出相应的EQB(Enrichment of Qualitative Belief)运算算子,定性合取、DST(Dempster-Shafer Theory)、DSmT组合规则,以及冲突分配规则,仿真计算表明:新提出的定性融合方法融合范围广,融合精度高,能够大大提高广义融合机融合算子组合证据源和冲突分配器分配冲突的能力。从信度赋值技术的角度出发,在广义幂集空间下,提出衡量两个证据源之间接近程度的证据支持贴近度的思想,因此有助于选择基本一致证据源,减少信息融合计算的复杂度,提高了融合精度和准确性。在广义幂集空间下,提出了由ESMS(Evidence Supporting Measurement of Similarity)信息过滤器、融合算子、冲突分配器构建广义融合机的方法,克服了错误、虚假和不一致等信息对融合机的干扰影响,充分发挥了融合机优势,减少融合计算的复杂度。进一步提高了不完善信息融合的准确性和精度,扩展了融合算子的融合空间和范围,提高了冲突分配的灵活性。针对Sonar传感器获取信息的不确定,在快速Hough变换自定位的基础上,应用广义融合方法对结构化环境进行了在线栅格地图创建,并同其它信息融合方法(DSmT,DSmT耦合PCR5(Proportional Conflict Redistribution Rule No.5),概率论、模糊理论、DST和灰色系统理论)进行了比较,比较结果说明了新方法具有融合精度高、计算效率高、算法稳定性好,适用范围广等优点,为智能机器人的导航、路径规划、SLAM(Simultaneous Localization And Mapping)等研究提供了新的思路。最后,本文基于VC++ 6.0和OpenGL设计开发了机器人智能融合感知系统,并且以Pioneer II移动机器人作为实验平台,验证了系统的可行性。该软件系统具有动态在线显示和在线控制的功能,以及面向对象模块化的设计风格,成为机器人感知和信息融合仿真与实验的平台。更多还原

【Abstract】 Information fusion technology is a new subject, which was developed from the martial application at the end of 1970s. With the rapid development of science and technology, especially since ten years ago, it has also got extensive applications in unmilitary fields. Then higher requirements are needed in precision and correctness of recognition, robustness of real-time fusion control and high effecicies of fusion management. However, due to the physical limitation of sensors, uncertainty of system and dynamic disturbance, imperfect information fusion is becoming a challenging problem.The researches on theories and methods are very useful to solve the problem of imperfect information fusion. Though many theories and methods have been proposed since 20 years ago, they can just deal with the special case in imperfect information fusion. Therefore, an extensive method needs to be proposed urgently to solve the quantitative and qualitative fusion problems. Here we take Dezert-Smarandache Theory (DSmT) and Neutrosophic theory regards as the framework, carry out studying deeply the generalized fusion problem of imperfect information from multi-source on the background of map building of mobile robot in unknown environment.From the point of view of quantitative fusion of imprecise information,we associate Fuzzy theory and Neutrosophic theory with belief assignment technology, and extend the combinational rules of belief assignment technology. So we improve the ability of fusion operator in combining evidence sources, and expand the fusion scope.Within the generalized framework, we study deeply on the qualitative fusion of uncertain information. We propose the idea of enrichment of qualitative belief (EQB) linguistic label on the basis of improving the qualitative operators, and also propose the corresponding EQB operator, EQB combinational rule (i.e. conjunctive rule, DSmT, etc), and EQB conflict redistribution rule. Some examples show that the method of qualitative combination has the advantage of extensive fusion scope, high fusion precision. The new method also improves greatly the ability in combining evidence sources and in redistributing conflict mass.Considering the belief assignment technology, we propose the idea of ESMS within the generalized power-set space, which can weigh the similarity between two evidence sources. The ESMS functions are useful to select consistent evidential sources to combine, reduce the complexity of computation, and improve the precision and correctness of fusion.We propose an approach to construct generalized fusion machine within the generalized power-set space, which consists of fusion operator, conflict redistribution, and information filter. Generalized fusion machine not only can avoid from the influence of mistaken, illusive and inconsistent information on fusion machine, exert the advantage of fusion machine adequately, and reduce the complexity of computation, and improve the effect of fusion greatly, but also improve the precision and correctness of fusion, expand the fusion scope of operator, and improve the flexibility of conflict redistribution.Aiming at the uncertainty of acquiring information with sonar sensors, we apply generalized fusion machine to grid map building online of mobile robot in structured environment with the help of self-localization based on fast-Hough transform. At the same time we also compare the generalized fusion machine with other theories (i.e. Probability theory, Dempster–Shafer Theory (DST), Fuzzy theory, Gray theory,DSmT,DSmT coupling with proportional conflict redistribution No.5 (PCR5)) in building map. The results of comparison show the advantage of high fusion precision, high computation efficiency, good stability and extensive applicability etc. The new tool also proposes a new research approach for navigation, path planning, SLAM (Simultaneous Localization And Mapping) and so on.At last, we design and develop an intelligent perception system for mobile robot based on Visual C++ 6.0 and OpenGL, and take Pioneer II mobile robot regards as the experimental platform, where the soft system is testified to be very valid. In addition, the system has the functions of dynamic display and control on-line and also has the design style of object-oriented. It becomes the platform of simulation and experiment for robot perception and information fusion.更多还原