P-VIDO: A Tightly-Coupled Vehicle Dynamics and Plane Priors Extension to Visual-Inertial State Estimator

Traditional monocular visual-inertial odometry (VIO) system cannot be applied to vehicle platforms due to unobservable degenerate motions, which result in poor initialization and low positioning accuracy. We propose the P-VIDO, a fast monocular visual-inertial odometry system with tightly coupled coplanar constraints and vehicle dynamics constraints. Plane landmarks are constructed using 3-point RANSAC methods and introduced into the optimization problem to provide coplanar constraints for point features. This significantly improves the accuracy of the odometry system in a structural environment such as an indoor parking lot. The experiments show that the system is executable in different vehicle motions with higher precision, as far as existing VIO methods are concerned. Besides, the system proposed in the paper can achieve 55fps averagely with a single thread.