Adaptive Attitude Estimation Using a Hybrid Model-Learning Approach

Attitude determination using the smartphone’s inertial sensors poses a major challenge due to sensor low-performance grade and variate nature of walking pedestrian. In this paper, data-driven techniques are employed address that challenge. To end, hybrid deep learning model based solution for attitude estimation is proposed. Here, classical equations applied form an adaptive complementary filter structure. Instead constant or weights, accelerometer weights in each axis determined by unique neural network. The performance proposed approach evaluated relative popular approaches experimental data.