利用先验点图模型的SLAM后端优化算法  

A Back-End Optimization Algorithm of SLAM Based on Graph Model with Prior Points

作　　者：吕瑞[1] 陈龙[2] 翁雪[3] 何金城[3] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430072 [2]中山大学移动信息工程学院,广东珠海519000 [3]武汉大学计算机学院,湖北武汉430072

出　　处：《武汉大学学报（信息科学版）》2014年第6期745-749,共5页Geomatics and Information Science of Wuhan University

基　　金：中央高校基本科研业务费专项资金资助项目(2013460003161001);空间信息智能感知与服务深圳市重点实验室开放基金资助项目(2013460004299001)~~

摘　　要：目前基于因子图的后端优化算法具有优越性。在因子图中,节点代表姿态,节点之间的边代表里程信息和封闭循环约束。由于因子图并未描述每个节点精度的差异,导致整体定位精度仍有提高的空间。针对这个问题,提出了一种基于先验点图模型的后端优化算法,依据前端提供节点精度的差异,在因子图中引入高精度点,然后采用改进的Levenberg算法进行全局优化,从而实现在结合原有概率约束的基础上,利用少量高精度点牵引其他点向真实值靠近,完成更为精准的自身定位。并在公开数据集上进行了实验,结果证明,本文提出的算法增强了前后端的关联,提高了定位精度。Simultaneous localization and mapping （SLAM） is a hot issue in the field of robotics, the problem consists of two parts, front-end perception and back-end optimization. At present, the backend optimization algorithm based on factor graph works well. In the factor graph, nodes represent poses, the edges between nodes represents the range information and closed loop constraint. Since the accuracy differences of nodes are not described in factor graph, the global positioning accuracy can be improved no further. To solve this problem, we propose a back-end optimization algorithm based on graph model with prior points which are introduced from the front-end that updates the graph model by fixing high-precision poses during optimization. The back-end can thereby use these fixed high-precision poses to drag low-precision poses closer to the ground truth and increase overall accuracy. We demonstrate the approach and present results on public datasets. The experimental results show that the maps acquired with our method show increased global precision.

关 键 词：图优化 机器人 同时定位与地图构建 后端优化算法 

分 类 号：P237.3[天文地球—摄影测量与遥感]