基于特征点云配准的SLAM重建深度优化方法  

SLAM reconstruction depth optimization method based on feature point cloud registration

作　　者：曹学伟 袁杰[1] 梁荣光 CAO Xue-wei;YUAN Jie;LIANG Rong-guang(College of Electrical Engineering,Xinjiang University,Urumqi 830017,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830017

出　　处：《计算机工程与设计》2025年第3期657-664,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62263031);新疆维吾尔自治区自然基金项目(2022D01C53);新疆大学研究生案例库建设基金项目(XJDX2023YALK26)。

摘　　要：针对视觉SLAM特征法中深度估计准确性和可靠性偏低的问题,提出一种基于特征点云配准的SLAM重建深度优化方法。利用SHOT特征描述子对三维点云提取特征,使用特征点云配准替代传统视觉特征法进行相机位姿估计。通过设定节点误差函数,优化视觉特征法建立的点云及其二义性点,建立带有纹理效果的稠密点云模型。在TUM和ICL-NUIM数据集上进行仿真实验,其结果表明,该方法相对于传统SLAM方法的相机位姿轨迹精度提升了10%。采用Kinect v2型RGB-D相机验证了该方法的有效性,实现了室内场景具有一定纹理效果的模型建立。Aiming at the problems of low accuracy and reliability of visual SLAM depth estimation by feature methods,a SLAM reconstruction depth optimization method based on feature point cloud registration was proposed.SHOT feature descriptors were used to extract features from three-dimensional point clouds and the feature point cloud registration was used to replace the traditional visual feature method for camera pose estimation.By setting the node error function,the point cloud established using the visual feature method and the ambiguous points in the feature point cloud were optimized to establish a dense point cloud model with texture effects.Simulation experiments on the TUM and ICL-NUIM datasets show that this method improves the camera pose trajectory accuracy by 10% compared to the traditional SLAM method.The effectiveness of this method is verified by using Kinect v2 RGB-D camera,and the establishment of a model with certain texture effects for indoor scenes is achieved.

关 键 词：点云配准 相机位姿估计 深度优化 稠密建图 同时定位与地图构建 点云地图 点云特征 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]