Indoor Positioning in Large Shopping Mall with Context based Map Matching

This paper focus on large indoor environment and proposes an accurate indoor positioning system with context based map matching. The proposed system adopts the widely used smartphone as the positioning platform of the customer, and mainly depends on the motion sensors in a smartphone. The proposed system firstly provides the initial trajectory of the customer with Pedestrian Dead Reckoning (PDR) and recognize various human activities which are meaningful for the localization. In this paper, we divided the activities into two types: (1) transition activity between floors, such as taking the escalator and elevator; (2) moving on the floor, such as walking outside store, shopping in store and tuning. The hierarchical Long Short-Term Memory (LSTM) Network based activity model is developed to recognize those actives. Secondly, those location-aware activities, PDR trajectories and 2.5D indoor map are integrated in a Hidden Markov Model (HMM) to conduct an accurate indoor positioning. Because the 2.5D map include the position information of indoor facilities such as escalators and each store, these positions information are used as the assistant information for conducting Context based Map Matching. The proposed method has 2.21 meter of positioning error mean and can achieve "shop level" performance.