基于改进点线特征的多源融合SLAM算法
Multi-source fusion SLAM algorithm based on improved point-Line features
林嘉洁 1徐胜 1苏成悦 2陈元电 1刘力斌 1施振华1
作者信息
- 1. 广东工业大学物理与光电工程学院,广州 510000
- 2. 广东工业大学物理与光电工程学院,广州 510000;广东工业大学先进制造学院,揭阳 522000
- 折叠
摘要
传统视觉SLAM算法在缺乏明显特征的场景或移动平台快速运动时,特征难以被检测和跟踪.提出了一种基于改进点线特征多源融合SLAM算法.前端通过相似短线段合并的策略改进EDLines线特征提取算法,提取长线段特征,后端融合了点、线特征和IMU数据,并采用非线性优化方法进一步估计相机位姿.实验结果表明,改进EDLines线特征提取算法相比传统LSD线特征提取算法,其特征提取速度提高4倍.在EuRoc数据集的测试中,本算法在不同场景下都具有很好的定位和建图效果,且具有较高的鲁棒性和实时性,对机器人实时的避障导航应用有较大的参考价值.
Abstract
Traditional visual SLAM algorithms are difficult to detect and track features in scenes lacking obvious features or when mobile platforms move rapidly.This article proposes a multi-source fusion SLAM algorithm based on improved point line fea-tures.The front-end improves the EDLines line feature extraction algorithm by combining similar short line segments,extracting long line segment features.The back-end integrates point,line features,and IMU data,and uses nonlinear optimization methods to further estimate the camera pose.The experimental results show that the improved EDLines line feature extraction algorithm has a feature extraction speed that is four times faster than the traditional LSD line feature extraction algorithm.In the testing of the EuRoc dataset,this algorithm has good positioning and image building effects in different scenarios,and has high robustness and real-time performance,which has great reference value for real-time obstacle avoidance navigation applications of robots.
关键词
SLAM算法/EDLines算法/IMU/点线特征/非线性优化Key words
SLAM algorithm/EDLines algorithm/IMU/point line features/nonlinear optimization引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据