节点文献
融合全局描述子和半直接法的双目SLAM
Binocular SLAMcombiningglobaldescriptorsandsemi-directmethod
【摘要】 针对半直接法SLAM方案的回环闭合问题,提出融合全局描述子和半直接法的双目SLAM方案。全局描述子计算模型结合主成分分析法(PCA)和K近邻方法构建关键帧的全局描述子管理模块;在位姿跟踪模块中使用最小化光度误差的半直接法;结合关键帧全局描述子和半直接法完成闭环检测与全局位姿优化。在公开数据集KITTI和Euroc上进行实验,与ORB-SLAM2、Open VSLAM进行对比,对比结果表明,该方法在大部分有回环的序列中有更小的绝对轨迹误差,在保证有相似定位精度的同时,提高了算法的跟踪帧率。
【Abstract】 Toaddresstheloopclosureproblemofthesemi-directmethodSLAMscheme,abinocular SLAMschemethatintegratedglobaldescriptorsandsemi-directmethodswasproposed.Theglobaldescriptorcomputingmodelwascombinedwith principalcomponentanalysis(PCA)and K-nearestneighbormethodtoconstructglobaldescriptormanagementmodulesforkeyframes.Inthepositionattitudetrackingmodule,asemi-directmethodtominimizethephotometricerrorwasused.Fortheloop closuremodule,theglobaldescriptionofthekey-framewascombinedwiththesemi-directmatchingmethodtocompletethe closed-loopdetectionandglobaloptimization.Theaccumulatederrorofthesystem wasthendecreased.Experimentswere carriedoutontheopendataset Kittiand Euro c,andtheproposedmethodwascomparedto ORB-SLAM2 and Open VSLAM.The resultsshowthatthealgorithmhaslesserroroftheabsolutetrajectoryinmostclosedloopsequences,andimprovestherateof frametrackingwhilemeetingthesimilarpositioningaccuracy.
【Key words】 globaldescriptor; semi-directmethod; loopdetection; binocularposetracking; knowledgedistillation;
- 【文献出处】 计算机工程与设计 ,Computer Engineering and Design , 编辑部邮箱 ,2022年03期
- 【分类号】TP391.41
- 【下载频次】185