一种室内动态场景下RGB-DSLAM的前端优化方法  

Front end optimization method of RGB-D SLAM in indoor dynamic scenes

作　　者：李泳 刘宏杰 周永录 余映[1] Li Yong;Liu Hongjie;Zhou Yonglu;Yu Ying(School of Information Science&Engineering,Yunnan University,Kunming 650500,China;Digital Media Technology Key Laboratory of Universities&Colleges in Yunnan Province,Kunming 650223,China)

机构地区：[1]云南大学信息学院,昆明650500 [2]云南省高校“数字媒体技术”重点实验室,昆明650223

出　　处：《计算机应用研究》2023年第4期991-995,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(62166048,61962060)。

摘　　要：针对动态物体会导致位姿估计误差过大的问题,提出一种适用于动态场景下的RGB-D SLAM算法。首先,将目标检测的结果根据物体是否具有自主移动能力划分为动态物体与潜在动态物体,并使用大津法在对应的深度图中完成检测框内区域的前、背景分割;然后,结合检测框的位置坐标与深度信息确定潜在动态物体与动态物体在三维空间内是否关联,初步筛选出具有一定运动概率的潜在动态物体;最后,通过运动一致性检测算法来最终确定是否将潜在动态物体作为动态物体处理。实验部分选用公开的TUM数据集,结果表明,该算法在保持实时性的同时具有较高的定位精度和鲁棒性。Aiming at the problem that dynamic objects will lead to excessive pose estimation error,this paper proposed a RGB-D SLAM algorithm for dynamic scenes.Firstly,the method divided the object detection results into dynamic objects and potential dynamic objects according to whether the object had the ability of autonomous movement,and used the Otsu method to complete the front and background segmentation of the area in the detection frame in the corresponding depth map.Then,combining with the position coordinates and depth information of the detection frame,it determined whether the potential dynamic object was related to the dynamic object in three-dimensional space,and screened out the potential dynamic object with a certain motion probability.Finally,it used the motion consistency detection algorithm to determine whether to treat potential dynamic objects as dynamic objects.The experiment selected the open TUM dataset.The results show that the proposed algorithm has high positioning accuracy and robustness while maintaining real-time.

关 键 词：同时定位与地图构建 动态场景 目标检测 图像分割 位姿估计 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]