A Consistent Navigation System Using CNN-LSTM Assisted by IMU Recomputed Method

The integrated Global Navigation Satellite System (GNSS)/Inertial Navigation System (INS) system has been widely used in vehicular positioning and navigation. However, the complex unstructured environments would lead to positioning degradation due to the GNSS outage. This paper proposed a Convolutional Neural Network-Long Short Term Memory (CNN-LSTM) assisted 21-dimensional GNSS/INS integrated navigation system using a Recomputed Method based on the Bias and Scale factor (BS-RM) error of the Inertial Measurement Unit (IMU). When GNSS is available, the obtained accurate GNSS/INS integrated navigation information is used as the input of the proposed BS-RM model to calculate the precise theoretical bias and scale factor error, which are trained as the target values of CNN-LSTM. When GNSS is unavailable, the trained CNN-LSTM is utilized to predict the accurate bias and scale factor. The consistent system positioning could be obtained with the suppressed INS error divergence by applying the IMU dead reckoning. To verify the performance of the proposed BS-RM model, three GNSS signal failure segments at different periods were randomly selected to evaluate the system. In addition, two GNSS failure segments were selected in the straight and curved roads respectively to further evaluate the performance of the CNN-LSTM assisted 21-dimensional GNSS/INS navigation system based on BS-RM. Compared with the traditional model of predicting position and velocity, the horizontal Distance Root Mean Square Error (DRMS) of BS-RM in the straight and curve tracks is improved by 83.85% and 88.06%, respectively, which confirms the improvement and consistent accuracy capability of the proposed method.