基于点线特征融合的实时视惯SLAM算法  

Real-time visual-inertial SLAM algorithm based on point-line feature fusion

作　　者：王磊[1] 陈帅坤 齐俊艳[2] 袁瑞甫[3] Wang Lei;Chen Shuaikun;Qi Junyan;Yuan Ruifu(School of Computer Science&Technology,Henan Polytechnic University,Jiaozuo Henan 454003,China;School of Software,Henan Polytechnic University,Jiaozuo Henan 454003,China;School of Energy Science&Engineering,Henan Polytechnic University,Jiaozuo Henan 454003,China)

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454003 [2]河南理工大学软件学院,河南焦作454003 [3]河南理工大学能源科学与工程学院,河南焦作454003

出　　处：《计算机应用研究》2024年第10期3008-3014,共7页Application Research of Computers

基　　金：河南省科技创新团队资助项目(22IRTSTHN005)。

摘　　要：为了在光照不足和低纹理场景中实现移动机器人的高精度实时定位和建图,提出了一种基于视觉点线特征以及IMU特征融合的实时SLAM算法。首先通过跳跃路由策略和自适应阈值策略改进了EDlines算法,提高了线特征提取的质量,从而提高了特征跟踪的有效性。然后将视觉惯性特征紧耦合建立约束,通过滑动窗口和边缘化模型进行非线性优化,实现了高精度高实时性的状态估计。实验证明,所提算法在线特征提取的有效性方面优于传统的线段提取算法,同时SLAM系统的定位精度和鲁棒性均得到有效提升。This paper introduced a real-time SLAM algorithm designed for mobile robots operating in environments characte-rized by low illumination and low texture.The algorithm leveraged the fusion of visual point-line features and IMU characteristics.It enhanced the EDlines algorithm by incorporating a leapfrog routing strategy and an adaptive threshold strategy,thereby improving the quality of line feature extraction and consequently enhancing feature tracking effectiveness.It established tight coupling constraints between visual and inertial features and employed a sliding window,along with a marginalization model,for nonlinear optimization.This enabled high-precision and real-time state estimation.Experimental results show that the proposed algorithm is superior effectiveness in online feature extraction when compared to traditional line segment extraction methods.Concurrently,the SLAM system achieves enhanced localization accuracy and robustness.

关 键 词：视觉同步定位与建图 特征提取 视觉惯性紧耦合 滑动窗口 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]