High-Precision GNSS/INS Tight Integration With Finer Signal Classification for Urban Areas

The global navigation satellite system (GNSS) real-time kinematic precise point positioning (PPP-RTK) technique can be significantly impaired by severe signal obstruction conditions in urban areas. The fisheye camera can detect structures present along the satellite-receiver line-of-sight (LOS) path, and all satellites blocked by structures are generally discarded or down-weighted, resulting in a limited number of satellites and/or poor satellite geometry. However, the seemingly blocked satellites might imply distinct signal reception situations, such as non-line-of-sight (NLOS) with only reflected path (i.e., blocked by the main body of structures) and diffraction with a slightly bending LOS path (i.e., close to the boundaries of structures). GNSS diffracted signals have significantly smaller ranging errors compared to NLOS signals, which has not been adequately addressed in open literatures leveraging fisheye cameras for satellite visibility detection. In this paper, to reasonably exploit such diffracted signals propagating close to structure boundaries, we develop a fisheye camera-aided PPP-RTK/INS (inertial navigation system) tight integration system. With those seemingly (i.e., fully or partially) blocked satellites identified by the fisheye camera, a finer classification approach is introduced to further identify the diffracted signals which could be beneficial for improving the positioning performance under GNSS-challenging scenarios. Then, a dedicated weighting model to such urban GNSS measurements, which combines conventional weighting model and robust estimation with the finer classification results, is established. Experimental results from both unmanned and manned vehicle-based kinematic tests in typical complex urban environments demonstrate that, compared to conventional fisheye camera-aided approaches that totally discard or down-weight signals, the proposed method improves 3D positioning accuracy by more than 22% for PPP-RTK and more than 47% for PPP-RTK/INS. Notably, only the proposed method achieves decimeter-level positioning precision in the vehicle-based test, while others remain at the meter level.