Research on high-precision multi-sensor fusion localization technology for intelligent vehicles considering Earth’s rotation

Abstract This paper targets high-precision localization for intelligent vehicles by fusing binocular vision, an inertial measurement unit (IMU), and a global navigation satellite system (GNSS) within a factor-graph optimization framework. We adopt IMU pre-integration augmented with Earth-rotation compensation and provide consistent noise propagation and measurement update formulations to improve pre-integration accuracy. For feature tracking and stereo matching, Shi-Tomasi corners are combined with an INS-assisted pyramidal optical flow. The factor graph incorporates IMU, visual reprojection/stereo, and GNSS residuals, and is equipped with algorithmic strategies for obtaining optimal localization estimates. We evaluate on public datasets under identical settings and conduct real-time on-vehicle experiments. The results show consistent reductions in absolute and relative trajectory errors compared with non-assisted baselines and demonstrate robust real-time performance. Functional ablations isolate the contributions of Earth-rotation-aware pre-integration and the INS-assisted visual pipeline. Overall, the system provides a practical and effective integration of INS assistance and Earth-rotation-inclusive IMU modeling in stereo visual-inertial-GNSS localization.