Accurate and reliable GNSS/INS/vision positioning system using consumer-grade sensors in urban complex environments

Abstract Continuous and accurate positioning is crucial for emerging applications such as assisted driving and mobile robots. The integration of the global navigation satellite system (GNSS), inertial navigation system (INS), and vision sensors has drawn extensive interest due to their complementary capabilities. However, consumer-grade sensors often suffer from high noise levels and significant systematic errors, making it challenging to satisfy positioning requirements. To solve this issue, we propose a GNSS/INS/Vision positioning system suitable for consumer-grade sensors, which can achieve drift-free, accurate, and reliable positioning in large-scale urban complex environments. In the proposed system, a hybrid short-term/long-term visual update method is adopted to maximize the utilization of visual common view relationships and minimize the rapid accumulation of INS errors. Moreover, with the assistance of GNSS, INS initialization can be implemented in the global frame, facilitating the uniform fusion of visual and GNSS observations without complex frame transformation. Additionally, the position error drift can be effectively corrected through the drift-free position provided by GNSS. The proposed integration system is evaluated on public and private datasets. Real-world experiments demonstrate that the positioning accuracy is 0.88 m, and the 95th percentile of errors is 0.95 m. Compared with state-of-the-art methods, the position accuracy of the proposed system on the public datasets is improved by more than 60%.