Improved Multi-Sensor Fusion Positioning System Based on GNSS/LiDAR/Vision/IMU With Semi-Tight Coupling and Graph Optimization in GNSS Challenging Environments

With the development of autonomous driving, precise positioning capabilities are becoming increasingly important. GNSS (Global Navigation Satellite System) is normally utilized for vehicle positioning, but is susceptible to factors such as urban canyons, especially in increasingly urbanized scenario nowadays. The interpretation of relative positioning information by means of multi-source sensors such as LiDAR (Light Detection And Ranging) or camera, has also been widely investigated, but there are deficiencies in the precision and reliability of the sensors due to their operating principles. For GNSS challenging environments, we give a semi-tightly coupled sensor fusion system based on factor graph optimization. The system is tightly coupled with the raw observations from the LiDAR, camera, and inertial measurement unit. Based on this, the back end is optimized by means of the factor graphs, which ultimately give optimized positional information of the vehicle, and the GNSS sensor is decoded to reduce the cumulative error, which is regarded as the loosely coupled component of graph optimization. We selected three representative datasets from publicly available datasets in different categories and conducted experiments respectively. The results indicate that this semi-tightly coupled system outperforms both the GNSS strategy and the typical SLAM (Simultaneous Localization And Mapping) strategies in terms of precision and possesses very high reliability, exhibiting the best performance.