未知环境中基于视觉的增量式拓扑建模及导航

Vision-based incremental topological mapping and navigation in unknown environments

针对未知环境提出并实现了一个新的基于摄像机视觉的增量式拓扑建模及导航系统。该系统包括自然路标提取、建模与定位、地图及路标库管理、规划4个主要部分。采用局部显著图像区域构造自然路标,并利用隐马尔科夫模型（HMM）建模当前场景中所获得路标间的关系,构造拓扑顶点。为提高定位精度,设计了最大后验概率（MAP）的学习策略。设计了基于竞争学习的路标管理机制,采用简单的最短路径算法在拓扑图上进行路径规划。该系统支持在线增量式建模,利用局部图像特征及其关系表示环境,定位算法简洁有效,辅以路标管理使之能够适应大规模环境导航任务。实验结果表明该系统路标提取稳定、位置识别率高、定位精确,能够保证机器人在未知环境中的安全导航。

This paper presents and implements a new incremental topological mapping and navigation system based on vision for unknown environments. The system consists of 4 major parts of natural landmark extraction, mapping and localization, map and landmark library management, planning and navigation. Salient local image features are adopted to create natural landmarks whose relationships are modeled by HMM to construct the vertex of topological map. To improve the accuracy of localization, a MAP learning scheme is designed. A landmark management mechanism is presented based on competitive learning. The shortest path algorithm is adopted for planning on the created topological map. The system has some features： mapping incrementally, representing environment using local image features and their relationship, compact localization algorithm, effective landmark management by which the system has the ability of working in large scale environments. Experiments have shown that the system can extract landmarks stably, achieve high accuracy rate of place recognition and guarantee safety navigation in unknown environments for mobile robot.