A Robust Factor Graph Optimization Method for Navigation in Land Vehicles Based on Dynamic Kernel Principal Component Analysis

Integrated navigation systems provide precise navigation solutions for advanced driver assistance systems in land vehicles. Factor graph optimization promotes the information fusion of such systems and achieves efficient real-time scalability; however, it cannot detect and process measurement outliers in real time, which limits the robustness of the system. To solve this problem, a novel robust factor graph optimization method based on the dynamic kernel principal component analysis is proposed. This method uses a time-lagged matrix to extract nonlinear features and Markov characteristics from normal navigation incremental data. A performance-monitoring system is constructed for real-time detection and evaluation of faults. Fusion weight coefficients are reasonably adjusted based on the degree of measurement outliers, avoiding the wastage of sensor information caused by binary decisions. Finally, an actual land vehicle trial is conducted in an urban street canyon. Experimental results illustrate that, compared with existing state-of-the-art methods, the proposed method can accurately detect and absorb mild outliers to ensure improved robustness and positioning accuracy.