Impersonal Smartphone-based Activity Recognition Using the Accelerometer Sensory Data

Smartphone-based activity recognition focuses on identifying the current activities of a mobile user by employing the sensory data which are available on smartphones. A lightweight model and less inquiry users for true activities, are necessary for deploying the activity recognition on a mobile platform for identifying activities based on new sensory data in real time. In this paper, we propose a new smartphone-based activity recognition framework for evolving sensory data stream called ISAR. It stands for Impersonal Smartphone-based Activity Recognition. ISAR model is built using annotated sensory data from a panel of user as training data and are applied to the new users. Our new model is an offline and online phase. In offline phase, we propose a new method for finding the threshold value which used to distinguish between dormant activities and energetic activities. Only a set of the energetic activities are used to build a light-weight classifier model. In online phase, we introduce the recognition technique of unannotated streaming sensory data with different activities. The experimental results using real human activity recognition data have conducted and compared with STAR model in terms of the accuracy and time complexity. Our results indicates that ISAR model can perform dramatically better than STAR model. Moreover, ISAR can utilize better than STAR model in real situation, especially across different users and without inquiry users.