组合点线特征的视觉SLAM算法研究  

Research on visual SLAM algorithm for combining point and line features

作　　者：段元鹏 袁安富[1] 张建伟[1] 程畅 叶健峰 DUAN Yuan-peng;YUAN An-fu;ZHANG Jian-wei;CHENG Chang;YE Jian-feng(School of Automation,Nanjing University of Information Science&Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学自动化学院,南京210044

出　　处：《信息技术》2025年第1期133-140,146,共9页Information Technology

摘　　要：针对视觉SLAM算法在低纹理场景下容易跟踪丢失和鲁棒性差的问题,提出了点线特征融合的视觉SLAM算法。该算法在前端跟踪阶段同时提取点线特征,并通过三个约束条件合并短线段以克服LSD算法长线段分割问题。后端优化阶段利用滑动窗口方法优化点线特征的重投影误差来提高定位精度。闭环检测引入基于点线特征的DBoW模型,有效减小运动漂移。实验结果显示,改进后的线段检测算法最多能够合并50%的短线段,点线特征融合的视觉SLAM算法表现出高鲁棒性和定位精度,绝对位置误差保持在10cm以内,满足机器人实时定位和地图构建需求。Based on the problem that the visual SLAM algorithm is easy to track loss and poor robustness in low texture scenes,a visual SLAM algorithm based on point and line feature fusion is proposed.The algorithm extracts point and line features in the front-end tracking stage,and combines short line segments through three constraints to overcome the long line segment segmentation problem of the LSD algorithm.In the back-end optimization stage,the sliding window method is used to optimize the re-projection error of point-line features to improve the positioning accuracy.Closed-loop detection introduces a DBoW model based on point and line features to effectively reduce motion drift.The experiment results show that the im proved line segment detection algorithm can merge up to 50%of the short line segments.The visual SLAM with point-line feature fusion shows high robustness and positioning accuracy,and the absolute position error is kept within 10cm,which meets the real-time positioning and map construction requirements of the robot.

关 键 词：视觉SLAM 短线段合并 线特征提取 点线特征融合 滑动窗口优化 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]