Deep learning-aided Kalman filter for low-cost attitude estimation via angular accelerometer and gyroscope fusion

The estimation of attitude plays a crucial role in various fields. In this paper, a novel method utilizing angular accelerometers and low-cost gyroscopes to enhance attitude estimation accuracy is proposed. Two distinct sensor integration schemes are designed according to the characteristics of two sensors, and the corresponding system state model and observation model are constructed respectively. By fusing angular accelerometer data with gyroscope measurements through Kalman filtering, the proposed methodology effectively reduces angular velocity measurement inaccuracies. To further improve the accuracy of attitude estimation, a deep learning method is designed to automatically obtain the noise covariance parameters of the Kalman filter for sensor fusion. This method eliminates the need for time-consuming data analysis and manual tuning, achieving a reduction in estimation error of over 17%. The experimental results demonstrate that the proposed method enhances attitude estimation accuracy in low-cost inertial measurement systems, providing a cost-effective solution for high-precision attitude estimation in resource-constrained applications.