Testbed development for a novel approach towards high accuracy indoor localization with smartphones

Due to its deep penetration in people's daily life, smartphone has been proposed as a practical platform for indoor localization. Yet one major challenge is how to handle the non-negligible sensor errors that can become problematic when accumulated over time. To this end, a series of approaches such as fingerprint and pedestrian dead reckoning have been proposed, which, however, either need WiFi infrastructure, pre-installed beacons or can only support certain movement patterns or scenarios. In this paper, we take one step further towards tackle this challenge by carefully developing a testbed that can enable deep investigation on the smartphone-based indoor localization problem and the potential for promising practical solution design. In particular, our testbed only accesses the raw inertial measurement unit and orientation data from the smartphone, making it infrastructure-free and require no pre-installation, and providing an in-depth view of sensor errors and their impacts on the localization accuracy. Our testbed also provides built-in functionalities for localization and supports real-time data processing and visualization, which can be extremely valuable for solution development and practical usefulness. We have conducted extensive experiments to evaluate our testbed, and obtained interesting observations that not only validate the effectiveness of our testbed design, but also opens a future direction to develop more advanced mechanisms such as deep learning based approaches to better compensate sensor errors and achieve high accuracy in practice.