GPS-Aided Stereo Inertial Navigation Localization Algorithm for Outdoor Scenarios

Abstract

To address the issues of visual-inertial simultaneous localization and mapping (VISLAM) methods in outdoor environments, such as inevitable drift accumulation during trajectory exploration, loop closure errors due to similar local visual appearances from different perspectives, and degradation of structural features under varying lighting and weather conditions, a tightly coupled GPS-aided visual-inertial SLAM algorithm (TGA-EFG) is proposed in this article. The algorithm enhances the accuracy of the front-end odometry and prior poses by employing the error-state Kalman filter (ESKF) method to fuse GPS measurements, inertial measurement unit (IMU) measurements, and visual measurements for initial value calibration. And then, a sliding window mechanism is introduced the factor graph optimization to overcome the computational redundancy issues of traditional one. In such way, it avoids high-dimensional matrix inversion, thereby improves the overall processing efficiency of the system. The Experimental results launched in both public KITTI dataset and the real environment demonstrate that, comparing our algorithm with mainstream SLAM algorithms, our algorithm improves localization accuracy by 51.63% and processing speed by 12.55%.