Improvement of line feature extraction and matching based on visual-inertial SLAM with point and line features
For the frontend line feature extraction and matching process in current visual-inertial SLAM systems fusing point and line features,the traditional line segment detection algorithm produces a large number of short lines and fragmented line segments due to over-segmentation,increasing the uncertainty of pose estimation in the backend optimization.Meanwhile,the efficiency of line feature matching using descriptors is relatively low,limiting the system's real-time performance.To address these issues,we propose an improved method.First,an improved line segment detection algorithm is proposed based on the traditional line segment detection algorithm,adopting a short line suppression strategy and an adaptive threshold line merging strategy to enhance the quality of extracted line segments.Second,based on the principle of KLT optical flow point feature tracking,a line feature tracking algorithm using KLT optical flow is proposed to improve the efficiency of line feature matching.Finally,comparative validation experiments are conducted on publicly available datasets of different scenes.Our results demonstrate our improved line segment detection algorithm enhances the quality of line segment detection.Moreover,our line feature tracking algorithm using KLT optical flow significantly improves the speed of line segment matching.The improvements ensure positioning accuracy and enhance the real-time performance of the visual-inertial SLAM system.
国家科技期刊平台
NSTL
万方数据
