HINNet + HeadSLAM: robust inertial navigation with machine learning for long-term stable tracking

In recent years, human position tracking with wearable sensors has been rapidly developed and shown great potential for applications within healthcare, smart homes, sports, and emergency services. Unlike tracking researches with sensors on the foot, human positioning studies with head-mounted sensors are fewer and still remain problems that have not been solved. We have proposed two studies solve part of the problems separately: HINNet is able to track people with free head rotations; HeadSLAM allows long-term tracking with stable errors. In this paper, to allow free head rotations meanwhile support long-term tracking, HINNet is combined with HeadSLAM and tested. The result shows that the combination could effectively distinguish head rotations and keep a low and stable position error in long-term tracking, with an absolute trajectory error (ATE) of 2.69m and relative trajectory error (RTE) of 3.52m.