Human tracking using wearable sensors in the pocket

Human tracking with wearable sensors such as Inertial Measurement Unit (IMU) is of great significance for ubiquitous computing and ambient applications. This paper proposes a novel Dead Reckoning based tracking algorithm using IMUs placed in the pocket. The contribution of our approach lies in three-folds: (1) Precise steps are detected according to people's repetitive moving patterns. (2) In each step, heading direction is estimated by the principle frequency of filtered acceleration. (3) Rather than inferring the heading direction of each step independently, we compute a vector in the IMU coordinates which can be transformed to the world coordinates to represent the heading direction, by solving an optimization problem with all historical tracking data considered. The proposed tracking algorithm is tested on a public dataset and outperforms five state-of-the-arts. We also apply it to real scenarios where our IMU tracking algorithm successfully assists visual tracking to overcome the challenging visual occlusion problems.