SPVIO: Pose-Only Visual Inertial Odometry With State Transformation and Visual-Deprived Correction
Xueyu Du, Lilian Zhang, Chengjun Ji, Xinbin Luo, Maosong Wang, Wenqi Wu, Jun Mao · IEEE Internet of Things Journal · 2025
Real-time and high-accuracy localization information plays an important role in the vehicle-based Internet of Things (IoT). Therefore, filter-based visual inertial odometry (VIO) has been widely noticed because its balance between accuracy and efficiency. However, typical methods such as MSCKF-based VIO suffer from insufficient accuracy due to inconsistencies and 3D feature estimation. To this end, we propose the State transformation and Pose-only VIO (SP-VIO) by rebuilding the state and measurement models, and considering further visual deprived conditions, namely the degradation environment where visual measurements completely invalidation. In detail, we first propose the double state transformation extended Kalman filter (DST-EKF) to replace the standard extended Kalman filter (Std-EKF) for improving the system’s consistency, and then adopt the Pose-only (PO) visual description, which utilizes the camera poses and 2D features to represent 3D features equivalently, thereby avoiding the accuracy and efficiency problems caused by 3D feature estimation. Meanwhile, we have completed the observability analysis of 2D feature-based VIO systems for the first time, and prove that SP-VIO achieves a more stable unobservable subspace, which can better avoid the inconsistency problem caused by spurious information. Furthermore, we propose an enhanced inertial-based correction method to optimize motion trajectories during visual interruption online. Monte-Carlo simulations and real-world experiments show that SP-VIO is more accurate and efficient than state-of-the-art (SOTA) VIO algorithms, and is more robust under visual deprived conditions. Meanwhile, the embedded system experiments have demonstrated that SP-VIO has ability for real-time edge computing and can be deployed in constrained IoT environments.