| 引用本文: | 苏全程,张金艺,李鹏,韩国川,何利康.融合双重闭环检测算法的ICP-SLAM修正策略[J].哈尔滨工业大学学报,2018,50(11):89.DOI:10.11918/j.issn.0367-6234.201708086 |
| SU Quancheng,ZHANG Jinyi,LI Peng,HAN Guochuan,HE Likang.Correcting ICP-SLAM strategy with double loop closure detection[J].Journal of Harbin Institute of Technology,2018,50(11):89.DOI:10.11918/j.issn.0367-6234.201708086 |
|
| 摘要: |
| 随着自动驾驶和人工智能的快速发展，SLAM作为其核心技术的研究日益深入.然而对SLAM日益深入的研究并没让其在生活和生产中被广泛使用.目前，阻碍SLAM发展的主要原因是其所涉及的算法和传感器都存在误差.这种误差随时间和空间累积直接导致创建的地图变形，甚至是创建地图失败，严重限制了SLAM的实际应用.本文针对这种累积误差导致地图变形的缺陷进行研究，使用激光雷达作为传感器，采用目前流行的ICP-SLAM算法，结合视觉SLAM中处理累积误差的闭环检测技术，提出融合双重闭环检测算法的ICP-SLAM修正策略对ICP-SLAM累积误差造成的地图变形进行修正.该策略首先对激光雷达采集的点云数据进行特征提取匹配，然后分析ICP对准产生的变换矩阵以确定闭环检测，最后采用均值分配方法修正地图变形.实验验证表明：该策略中的闭环检测准确率高于常规闭环检测方法，和深度神经网络的闭环检测相比有13.33%提升；同时该策略在闭环检测过程中可以计算出导致地图变形的累积误差；并且在旋转方向上变形度修正效果显著，可以提升54.61%.
|
| 关键词: 同步定位与制图 闭环检测 累积误差 特征匹配 激光雷达 |
| DOI：10.11918/j.issn.0367-6234.201708086 |
| 分类号:TP242 |
| 文献标识码:A |
| 基金项目:国家高技术研究发展计划(863计划)(2013AA03A1,3AA03A1122); 上海市教委重点学科资助项目基金(J50104) |
|
| Correcting ICP-SLAM strategy with double loop closure detection |
|
SU Quancheng1,ZHANG Jinyi1,2,LI Peng1,HAN Guochuan2,HE Likang2
|
|
(1. School of Communication and Information Engineering, Shanghai University, Shanghai 200072, China; 2.Microelectronic Research and Development Center, Shanghai University, Shanghai 200072, China)
|
| Abstract: |
| With the rapid development of automatic driving and artificial intelligence, SLAM has been paid more and more attention. However, more research on SLAM has not been widely used in life and production because of errors in the algorithms and sensors involved in SLAM. The errors accumulate over time and space that directly results the deformation of the map, which severely limits the practical application of SLAM. Based on the loop closure detection of visual SLAM and the advantages of lidar, correcting ICP-SLAM strategy with double loop closure detection is presented to deal with the accumulated errors. The strategy firstly extracts and matches the point cloud data collected by lidar, and then analyzes the transformation matrix produced by ICP to determine loop closure detection, and finally, uses the mean distribution method to correct the map deformation. Experimental verification shows that: the accuracy of the closed-loop detection is improved, which is about 13.33% higher than that of the depth neural network; this strategy can calculate the accumulated error that leading to the deformation of the map; it can effectively correct the map deformation caused by ICP-SLAM, and the deformation has a 54.61% improvement in the rotation direction.
|
| Key words: SLAM loop closure detection accumulated error feature matching lidar |
