An Enhanced Vehicle Self-Positioning Method During GNSS Outages Using Factor Graph Optimization and CNN-LSTM-Attention

Integration of inertial navigation system (INS) and global navigation satellite systems (GNSS) is a promising approach to vehicle self-localization. However, such a scheme may have poor performance during GNSS outages, and the errors of INS also disperse rapidly, leading to a serious degradation of the positioning accuracy. To address the issue of GNSS signal failure, this article presents an enhanced vehicle self-positioning method based on factor graph optimization (FGO) and convolutional neural network (CNN)-long short-term memory (LSTM)-Attention. An odometer (ODO) sensor is introduced to provide additional velocity information in the direction of the vehicle’s movement, and the FGO is employed for achieving the fusion of GNSS/INS/ODO data. The position increments during GNSS outage are predicted by a CNN-LSTM-Attention model, which utilize a CNN to quickly extract features from the input signals, an LSTM network to model the time series, and an attention mechanism to dynamically focus on the most important parts of the input sequence, thereby improving the overall prediction accuracy and robustness of the model. The road experiments based on the real-world data collected from practical vehicle platform have demonstrated the effectiveness and advantage of our proposed method in enhancing the vehicle self-positioning performance during GNSS outages. For instance, in the experiments of GNSS interruption for 60 and 90 s, the RMSEs of CNN-LSTM-Attention + FGO(INS/ODO) for the position have the performance improvement of 34.62% and 46.58% over LSTM + FGO(INS/ODO), respectively.