POPL-VIO: A Novel Pose-Only Measurement Model for Point-Line-Based Visual Inertial Odometry
Zhaolong Yang, Shuwen Chen, Hai Zhang · 2024
The classical filter-based Visual Inertial Odometry (VIO) system, MSCKF, requires triangulation to obtain feature landmarks and constructs measurement equations on the historical co-visible frames. This approach introduces linearization errors from feature landmarks and delayed updates for the current frame state. Building upon the recent pose-only measurement model for point features, we propose a pose-only measurement model for line features and develop a point-line-based VIO system, denoted as POPL-VIO. This system eliminates both point and line feature landmarks in the measurement equations, thereby avoiding linearization errors from feature landmarks and providing immediate updates for the current frame state. Additionally, we propose a unified base-frame selection strategy specifically designed for the pose-only measurement model of both point and line features. Experiments on the EuRoC dataset show that POPL-VIO improves localization accuracy by 32% over MSCKF and outperforms both the point-based system, OpenVINS, and the state-of-the-art point-line-based system, EPLF-VINS.