导航

ACTA AERONAUTICAET ASTRONAUTICA SINICA ›› 2022, Vol. 43 ›› Issue (3): 325117.doi: 10.7527/S1000-6893.2021.25117

Previous Articles     Next Articles

Stereo visual-inertial SLAM algorithm based on merge of point and line features

ZHAO Liangyu1, JIN Rui1, ZHU Yeqing1, GAO Fengjie2   

  1. 1. School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081, China;
    2. Hiwing Aviation General Equipment Co., Ltd., Beijing 100070, China
  • Received:2020-12-16 Revised:2020-12-31 Online:2022-03-15 Published:2021-03-01
  • Supported by:
    National Key R&D Program of China (2017YFC0806700); National Natural Science Foundation of China (12072027, 11532002); Open Research Project of The Beijing Key Laboratory of High Dynamic Navigation Technology under grant (HDN2021101)

Abstract: In indoor weakly textured environment, it is difficult for the SLAM algorithm based on point features to track sufficient effective point features, which leads to low accuracy and robustness, and even causes the system to fail completely. For this problem, a stereo visual SLAM algorithm is proposed based on point and line features and the Inertial Measurement Unit (IMU). The data association accuracy is improved by using the complementation of point and line features, and meanwhile the IMU data is incorporated to provide prior and scale information for the visual localization algorithm. More accurate visual pose is estimated by minimizing multiple residuals function. The environment point and line feature map, dense map and navigation map are then constructed. To overcome the disadvantages of traditional line feature extraction algorithms, which are easy to cause detection of a large number of short and similar line segment features and over-segmentation of line segments in complex scenes. The strategies of length suppression, near line merging and short line chaining are introduced, and an improved FLD algorithm is proposed to reduce the mismatch rate of the line features, and the running speed of the algorithm proposed is more than twice of that of the LSD algorithm. By comparing the simulation results obtained from multiple groups of public datasets and real-world weak texture scenes, it can be seen that the proposed algorithm can obtain richer environment maps with great positioning accuracy and good robustness.

Key words: simultaneous localization and mapping, weakly textured environment, point and line feature, stereo visual-inertial system, chain short line segment

CLC Number: