Detecting anomalous activities by fusion of accelerometer and passive infrared sensor

In ambient intelligence environment, cameras serve as interface or monitoring devices for user activities. However, the use of cameras is often associated with privacy concerns and resource limitations. In order to alleviate these problems, anomaly detection by fusion of an accelerometer and a passive infrared (PIR) sensor can be employed to trigger the corresponding camera for further analysis only as the need arises. This paper discusses the combination of these off-the-shelf sensors to detect specific anomalous activities. In particular, an example of detecting an irregular user in the work environment is presented. We describe how to extract expressive features from both modalities and combine them to train a classifier on highly imbalanced datasets. The experimental results over real-life data show the effectiveness of our approach in detecting anomalous activities and therefore potentially reducing the use of cameras.