【作者】 安婷婷；

【导师】 肖南峰；

【作者基本信息】 华南理工大学 ， 计算机应用技术， 2010， 硕士

【摘要】 随着社会老龄化程度的不断提高,以及现代人生活方式的转变,家政服务机器人正逐渐走入人们的生活,并且显示出巨大的市场需求。家政服务机器人要投入实际应用,自主导航能力是关键,也是实现其真正智能化和完全自主移动的关键技术。而家政服务机器人定位和环境地图的构建是实现导航的基础。定位过程依赖于精确的地图信息,而地图构建又反过来依赖于精确的位姿信息。二者既矛盾又相关,必须同时加以考虑。因此,本文的研究内容为家政服务机器人同时定位与地图构建(SLAM)。本文首先对基于距离传感器的室内环境下的同时定位与地图构建进行研究。针对传统的基于Rao-Blackwellized粒子滤波器的SLAM算法需要大量的采样粒子,而且频繁重采样操作可能导致粒子耗尽的问题,提出并实现了一种改进算法。改进的Rao-Blacekwellized粒子滤波SLAM算法在计算采样的提议分布时考虑了里程计信息和距离传感器信息,并且通过计算有效粒子数目适时进行重采样操作,通过加入随机粒子来维持多样性。该方法能减少粒子数目,同时保证算法的一致性。仿真实验证明,改进的算法提高了计算效率,创建的栅格地图具有更高的精度。其次,本文研究了基于视觉的家政服务机器人同时定位与地图构建(vSLAM)。vSLAM的关键点之一是视觉图像特征点的提取。本文创新性地将快速鲁棒特征(SURF)算法应用到家政服务机器人环境认知中,实现了基于单目视觉的SURF地图库图像匹配。机器人从成功匹配的图像中获得当前所处的环境信息。SURF算法比尺度不变特征变换(SIFT)算法的计算速度提高了3倍,因此具有更好的实时性。家政服务机器人获得了一定的环境认知后,同时结合距离传感器和里程计创建的室内环境地图,即可以生成包含语义信息的地图。有助于家政服务机器人完成路径规划、导航等后续任务。更多还原

【Abstract】 With the improvement of aging society and the change of modern lifestyle, home service robots graduately enter people’s lives, showing the huge market demand.In order to make home service robots practical application, the ability of autonomous navigation is the key issue, which is also the key technology to realize the real intelligence and mobility. Simultaneous Localization and Mapping (SLAM) of home service robot is the basis for navigation. On the one hand, localization relies on the accurate map information. On the other hand, map construction, in turn, depends on the precise pose information. They are contradict and related, thus must be considered simultaneously. Therefore, this dissertation researches on the SLAM problem of home service robots.This paper firstly studies the SLAM problem based on the range sensors. To due with the problem that the conventional Rao-Blackwellized particle filters based SLAM algorithm requires a large number of particles and that the frequent resampling might lead to the problem of particle impoverishment,an improved approach is proposed.It takes into account both the odometry and the observed information when computing the proposal distribution, resample according to the calculation of the effective sample number and adds some stochastic particles in order to maintain the diversity.Thus this novel method decreases the number of particles and is able to meet the requirement of consistence. Stimulation experimental results show that the proposed algorithm improves the computational performance as well as builds grid maps with higher accuracy.Secondly, this paper does researches on visual SLAM (vSLAM) problem for home service robots. One of the key issues in vSLAM is feature points extraction of visual images. In this paper, the Speeded Up Robust Features (SURF) algorithm is employed to solve the environment recognize problem of the home service robot, and the library based image matching using only monocular vision is realized. The robot learns about the environmental information from successfully image matching. The SURF algorithm is three times faster in calculating speed than the Scale Invariant Feature Transform (SIFT) algorithm, so it is better in real time performance. After recognizing the environment, the home service robot combines the visual information and the grid map constructed from range sensors and odometer, so as to generate an indoor environment map containing semantic information. This map is helpful for home service robots to achieve follow-up tasks, such as path planning, navigation, etc.更多还原