An Tightly-Coupled VIO Algorithm with Backend Pose Graph Optimization for Outdoor Applications

Conventional SLAM systems often encounter drift and instability in outdoor environments due to lighting changes and weather interference. This paper presents a novel GPS-aided visual-inertial SLAM algorithm. It integrates GPS with visual and inertial sensors at the front end and uses pose graph optimization at the back end to achieve higher accuracy positioning and robustness. The proposed method uses an error-state Kalman filter (ESKF) to tightly couple GPS, IMU, and visual measurements, improving odometry and pose estimation accuracy. The back-end's sliding-window pose graph optimization greatly enhances computational efficiency and ensures high real-time performance. Compared to VINS-Fusion, the proposed system reduces positioning error by 56.3% on KITTI Sequence 07.