Pose Invariant Activity Classification for Multi-Floor Indoor Localization

Smartphone based indoor localization caught massive interest of the localization community during recent years. GraphSLAM (Simultaneous Localization and Mapping) is one of the most effective approach to optimize pedestrian trajectory given a set of visited landmarks during movement. One can use visual cues to detect a landmark. Otherwise, NFC tags or QR codes can be installed where the pedestrian detects landmarks from a suitable smartphone application by reading tags and codes. Alternatively, human activity can be classified to detect organic landmarks such as visits to stairs and elevators while in movement. We provide a novel human activity classification framework which is invariant to the pose of the smartphone. Pose invariant feature allows robust observation no matter how a user puts the phone in the pocket. In addition, activity classification obtained by an SVM (Support Vector Machine) is transmitted to an HMM (Hidden Markov Mordel) which improves the activity inference based on temporal smoothness. Furthermore, the HMM integrates inferences with activity and floor effectively because they provide complementary information. Our experiments show that the proposed framework detects landmarks accurately and enables multi-floor indoor localization.