Intelligent Extraction Method of Inertial Navigation Trajectory Behavior Features Considering Road Environment.

Inertial navigation systems play an important role in areas without satellite signals. However, the inertial navigation error changes over time. Therefore, it is essential to use external data sources to correct the trajectory data, and feature extraction is required for error correction. At present, the inertial navigation trajectory consists primarily of track points and segments, with relatively simple features, no time-series data, and no spatial attributes, such as elevation and speed. In this paper, a convolutional neural network (CNN) is established based on deep learning theory and the rich behavioral features of the inertial navigation trajectory. A feature classification model based on the CNN and gated recurrent unit (GRU) is proposed to extract features from the real-time inertial navigation trajectory. First, preprocessing of the vehicle data obtained by the inertial navigation system is performed to filter redundant and invalid data. Subsequently, the trajectory features are categorized according to the motion trend and time-series information. The trend feature vector CNN and the time-series feature vector are used as inputs to the CNN and GRU, respectively, and the trajectory model with the behavioral features is established. Finally, the long short-term memory (LSTM) model, which is prone to overfitting when many parameters are used, is improved, and the two feature vectors are input into the CNN model and GRU model for data fusion to ex-tract the behavioral features. A typical inertial navigation trajectory dataset of the Miyun area in Beijing in December 2015 is used to extract the behavioral features. The experimental results are compared with traditional feature ex-traction and neural network classification methods. The results show that the proposed method outperforms the other methods, with a feature extraction ac-curacy of 91.44%. The method shows excellent performance for extracting the behavioral features of the inertial navigation trajectory in a region with variable elevation and velocity.