Accelerometer-based smartphone step detection using machine learning technique

Under the limitation of GPS measurement in indoor environment, one of alternatives for smartphone navigation is using embedded inertial sensors. In order to perform pedestrian dead reckoning with accelerometers and gyroscopes, step should be preferentially detected. In addition, there are a variety of smartphone placements such as handheld, texting, and trouser pocket. In this paper, step detection methods for various placements of smartphone are proposed using machine learning technique and attitude computation of the device. The experimental results show that the proposed method is able to detect steps robustly under handheld, texting, trouser back and front conditions.