Statistical Metric-Theoretic Approach to Activity Recognition Based on Accelerometer Data

Abstract

Providing accurate information on people’s actions, activities, and behaviors is one of the key tasks in ubiquitous computing and it has a wide range of applications including healthcare, well being, smart homes, gaming, sports, etc. In the domain of Human Activity Recognition, the primary goal is to determine the action a user is performing based on data collected through some sensor modality. Common modalities adopted to this end include visual and Inertial Measurement Units (IMUs), with the latter taking precedence in recent times due to their unobtrusiveness, low cost and mobility. In this work we consider the accelerometer signals streamed through a wearable IMU unit and use this data to recognize the user’s activity. We develop a novel approach based on representing the coming signal as a probability distribution function and then use some distance metric to infer the dissimilarity between probability distributions corresponding to different accelerometer signals in order to infer the correct activity. Experiments are performed on 14 activities of daily living with results showing promising potential for this technique.