Preliminary Investigation on Recognizing Environment-dependent Human Actions using a Fusion of Wi-Fi Based Place Clustering and Motif Detection from Accelerometer Data

摘要

This paper presents a novel method to estimate a location label of a user using sensor data fromhis/her smart devices (smartwatch and smartphone) without using labeled training data collected in his/herenvironments. Specifically, we attempt to predict the user's location semantics, i.e., location classes such asrestroom and meeting room. We propose a novel matrix operation method that automatically finds acceler-ation data motifs specific to a certain location class, e.g., brushing teeth action observed at bathroom, basedon the idea of Gini coefficient. We evaluated the proposed method in 4 real environments.