Real-Time 3D Indoor Positioning with Human Activity Transition Recognition on Mobile Devices

This work develops real-time three-dimensional indoor positioning techniques for mobile platforms such as smartphones by exploiting human activity recognition (HAR). Taking into account practical constraints on mobile platforms including limited computing power and noisy built-in sensors, we develop a deep neural network (DNN)-based approach to perform HAR by focusing on the pedestrian activity transitions, rather than recognizing each individual activity class. Furthermore, we propose to first extract key location information from indoor floor-plans before combining the information with improved Pedestrian Dead Reckoning (PDR) technology to achieve real-time estimation and correction of pedestrian locations. Extensive computer simulation and field experiments confirm that our proposed system can achieve three-dimensional indoor positioning of impressive positioning accuracy.