【摘要】 为了提高基于点特征的SLAM算法的定位精度和鲁棒性，提出一种点线特征融合的快速单目视觉惯性SLAM算法。使用ELSED算法快速提取高质量的线特征，在非关键帧追踪时，基于连续帧之间微小运动的假设，实现连续图片间的快速线段匹配，且无需线段描述子。在插入新关键帧时，提取线段描述子来完成关键帧之间的线特征匹配，创造新的地图线。最后在公开数据集上进行实验，结果表明：ELSED算法在提取高质量线段的同时，耗时仅为LSD算法的13%;与传统利用线段描述子的匹配算法相比，此算法的时间效率提升了83%,并减少线段误匹配，提高系统定位精度，系统的平均跟踪帧率为33帧/s,保证了系统的实时性。更多还原

【Abstract】 To improve the localization accuracy and robustness of point features-based SLAM algorithm, a monocular visual inertial SLAM algorithm with fast point-line feature fusion was proposed.In the algorithm, high-quality line features were extracted fastly with ELSED(enhanced line segment drawing) algorithm; for non-keyframe tracking, fast line segment matching between consecutive images was implemented based on the assumption of tiny movements between consecutive frames with no line segment descriptors sub required.Line segment descriptors sub was extracted to complete the matching of line features between keyframes when new keyframes were inserted, which created new map lines.Finally, experiments were performed on a public dataset.The results demonstrate that the ELSED takes only 13% of the LSD algorithm while extracting high-quality line segments; compared with the traditional matching algorithm utilizing line segment descriptors, the proposed algorithm improves the time efficiency by 83% and reduces line segment mismatching to enhance the system positioning accuracy; the average tracking frame rate of the system is 33 frame per second, which ensures the real-time performance of the system.更多还原