SLAM中融合形状上下文和随机步进的图匹配数据关联算法

提出了一种在非确定环境下求解SLAM数据关联问题的图匹配算法. 算法建立了SLAM中数据关联的图论模型,对图模型节点提取了不依赖位置信息的形状上下文特征(shape context,SC),最后通过二次加权随机步进算法(reweighted random walks,RRW)得到图匹配问题的优化解. RRW&SC图匹配算法充分利用了路标间的拓扑结构关系以及路标间的形状结构,极大地扩展了数据关联时所依据的几何信息量. 仿真实验结果表明,与传统算法相比,该算法能有效处理SLAM中噪声干扰增加、机器人迷失、路标被动态遮挡等不确定程度高、歧义性大环境中的数据关联. 

Graph Matching Algorithm for the Data Association Problem of Simultaneous Localization and Mapping in Ambiguous and Dynamic Environments

Proposed a graph matching approach RRW&SC to tackle the data association problem inherited in the SLAM. In our framework, the graph theory was utilized to build a mathematical model for data association firstly. Then the shape context feature was extracted for each node. Reweighted random walks was lastly adopted as the optimization engine to obtain the optimal solution for the graph model. The topology structure of the landmarks and the shape of the landmarks was used by RRW&SC algorithm, thus the geometric information of the environment was greatly enhanced which facilitates the data association. Simulation results show that, compared with traditional algorithms, the proposed data association algorithm can effectively handle a variety of complicated scenarios which might occur in SLAM, including enlarged observation noise, robot being kidnapped, or dynamic occlusion.