Attitude Adaptive Estimation with Smartphone Classification for Pedestrian Navigation

Accurate attitude for wearable devices and smartphones is needed for many applications. The major challenge is to cope with the acceleration resulting from the user or smartphone dynamics. To that end, a two-stage adaptive complementary filter for attitude estimation is proposed. Upon identifying the smartphone location on the user using a deep learning approach, the accelerometers weights in each axis are adjusted according to an optimized gain map. To evaluate the benefits of the proposed approach it is compared to commonly used algorithms both in simulation and experiments.