WhereWear: Calibration-free Wearable Device Identification through Ambient Sensing

With the growth of wearable devices, numerous health and smart building applications are enabled. As a result, many people wear multiple devices for different applications, such as fitness tracking. Being able to match devices' physical identity (e.g., smartwatch on Bob's left forearm) to their virtual identity (e.g., IP 192.168.0.22) is often a requirement for these applications, especially when different people could use the same device --e.g., home, gym equipment-- or be worn in different places. Context-aware sensing, utilizing the insight that co-located sensors detect the same events, has been used to establish such association ubiquitously. However, challenges arise with the growing number of on-body devices, and existing approaches fail to distinguish where the devices are on the body. In this paper, we present WhereWear, a human pose estimation based calibration-free wearable device identification mechanism through ambient sensing of vision (camera in the environment) and motion (IMU in wearable devices). Our system utilizes the key fact that the orientation change of the devices is related to the orientation change of the body part where the device is on, and matches each device's physical and virtual identities at the bodypart--any part of the body between two joints, e.g. forearm, thigh, etc.-- resolution. We evaluate the proposed mechanism on a publicly available dataset, where multiple human subjects perform activities with 13 devices on their body. Our system demonstrates up to 64% device identification accuracy on average and up to 2.7X improvement over existing baselines with five simultaneous users. With only one user, it achieves up to 92.2% accuracy.