基于视觉和距离传感器的SLAM和导航方法的探新

摘要

Simultaneous Localization and Mapping(SLAM) is widely used for generation of maps for autonomous robotic navigation etc.SLAM based on visual features is a kind of low cost solution. Besides, SLAM based on visual features have plenty information for matching and recognition.But appearance of actual world is always dynamic. Many kinds of vision-based SLAM method are not robust to influence of unstable objects such as walking humans in cafeterias or shopping malls.ICGM(Incremental Center of Gravity Matching) is method which extracts static visual features from images sequence based on feature points geometric structure. SLAMs based on ICGM can exclude bad influences of dynamic objects. This proposed method is using ICGM to extract static features and combines range finderinformation. Thus, the proposed method can achieve high accuracy robotic navigation in highly dynamic environment.We run our approach in a crowed cafeteria. The result shows that by using proposed method SLAM and navigation can be achievedfast enough for real-time processing. Comparing to previous, the proposed method's precision is higher. Proposed method is more suitable for actual needs.%SLAM(Simultaneous Localization and Mapping)被广泛应用于生成地图和机器人导航领域。基于视觉特征的SLAM是一种低成本的解决方案，且能够提取环境的丰富的特征点用于机器人导航，但是，现实世界的可视特征往往是动态的，很多基于视觉特征的SLAM方法都不能排除环境中移动物体的影响，比如：在餐厅或购物中心的行人等等。ICGM(Incremental Center of Gravity Matching)是一个通过图像队列之间特征点的几何结构互相匹配而得到稳定特征点的方法，基于ICGM的SLAM能有效地排除环境中移动物体的影响。本研究在前人研究的基础上提出了一种新的算法，利用ICGM的方法提取环境稳定的视觉特征，结合Kinect距离传感器，在高度动态环境中实现了较高精度的机器人自动导航。并通过实验检验了这种算法，实验结果显示，本方法的速度能够达到实时处理的要求，与此前研究的其它方法相比，这种方法能够达到更高的精度，所提出的SLAM和导航系统更加接近实用要求。