FLM PL-VIO: A Robust Monocular Point-Line Visual-Inertial Odometry Based on Fast Line Matching
Shuyue Lin, Xuetao Zhang, Yisha Liu, Hanzhang Wang, Xuebo Zhang, Yan Zhuang
Abstract
This article proposes a monocular point-line visual-inertial odometry (VIO) with line filtering and fast line matching (FLM PL-VIO), improving the localization accuracy and robustness. Different from the existing point-line VIO frameworks, we construct a new line filter to make the line features uniformly distributed, providing better spatial geometric constraints. In addition, we propose a fast line matching method based on the multisegment tracking and extended verification, which can solve the line matching failure caused by the fragmentation of line detection and object occlusion. As a result, it guarantees the accuracy of line matching while significantly improving the matching speed. The state estimator is constructed by jointly optimizing visual residuals from the point and line features, inertial residuals, and prior information of marginalization. Our system is evaluated on the challenging datasets and the real-world flight experiment, demonstrating superior performance in the localization accuracy, the time of line detection and matching compared with the state-of-the-art localization algorithms.