PLMCVIO: Point-Line based Multi-Camera Visual Inertial Odometry

To achieve robust and accurate state estimation for autonomous mobile systems, we propose a novel point-line based multi-camera visual-inertial odometry, called PLMCVIO. To improve the robustness of state estimation, line features are added into the state vector of multi-camera visual inertial odometry (MCVIO). In particular, Plucker coordinates and minimal parameterized orthonormal representation are adopted to model line features, which ensures system precision and computation efficiency. Further, a point-line based initialization method is developed to obtain accurate and reliable initial pose. Finally, a general hyper-graph optimization method is proposed for accurate and efficient state estimation. Extensive experimental results on public dataset and in real-world scene demonstrate the accuracy and robustness of the proposed PLMCVIO.