A GPS-Aided Semi-Direct Visual Inertial Global Odometry

Motivated by the goal of realizing long-term drift free pose estimation, a global positioning framework integrating global positioning system (GPS) measurements and semi-direct visual inertial odometry is proposed in this paper. Photometric correction is integrated into the front-end to effectively deal with the visual degradation caused by drastic changes in ambient luminosity. In the global optimization module, the local motion estimation and the positioning information of GPS are fused in a pose graph optimization, which realizes the global drift free pose estimation. However, GPS measurements will deteriorate when it is blocked or interfered by the signal. Therefore, we use error-state Kalman filter (ESKF) to fuse noisy GPS and inertial measurement unit (IMU) measurements to obtain more accurate and smooth GPS measurements, which are then sent to global pose graph optimization module to fuse with local poses. We evaluate the performance of our algorithm on indoor and outdoor