Hybrid Multi-frequency Attitude Estimation Based on Vision/Inertial Integrated Measurement System

To ddetermine the accurate attitude in vision/inertial integrated measurement system, a hybrid measurement strategy termed as long short-term memory (LTSM) neural networks-based multi-frequency cubature Kalman filter (MFCKF) is proposed to achieve accurate attitude information and reduce angle errors. The process of the proposed LSTM-MFCKF contains two innovation steps: (1) firstly, inertial information which comprises of angular velocity and angle are considered as the inputs in the MFCKF loop, and visual attitudes are calculated out of the filter loop when the sampling frequency of MEMS-based gyroscope device match the vision sensor; (2) LSTM starts to train matched vision data in training mode and output predicted vision information to match inertial sensors in prediction mode; (3) the predicted vision results and high frequency inertial information are fused in next MFCKF estimation, and the vision results are stored as the observation values. The proposed integrated attitude determination system is tested on semi-physical simulation platform. Experiment results with superior precision and reliability show the feasibility and effectiveness of the proposed method for attitude measurement system.