Heading estimation indoors using a smartphone in the pocket

The problem of heading estimation is central for indoor positioning system using smartphone inertial sensors. Instead of assuming that the smartphone is held in hand as many previous works, this paper aims to develop a heading estimation approach in a more common situation that the smartphone is held in a pocket. The proposed approach exploits the acceleration patterns, whose first principal component in the horizontal plane is always parallel to the user heading during pedestrian walking. Since the device coordinate systems vary with the leg movements during walking, we project all the acceleration signals into the initial device coordinate system, and then apply principal component analysis in the projected horizontal plane at the initial device coordinate system for walking direction extraction. To validate our approach, a total of 10 straight traces with different device orientations in the pocket are carried. Experimental results show that our approach has a mean absolute estimation error of 11.3 degrees with 9.5 degrees standard deviation.