Abnormal Activity Detection for the Elderly People using ConvLSTM Autoencoder

Abstract

Abstract Independent living for elderly people is often viewed as an impossible task, due to the great many perils and difficulties. With advancements of Ambient Intelligence, this scenario is no longer out of reach and smart homes offer a computationally inexpensive solution to this problem. In this paper we address these difficulties and propose a novel method for Anomaly Detection in elderly’s daily routine and behavior. Our proposed model is a ConvLSTM Autoencoder for processing spatiotemporal data, given the fact that these type of behavior and anomalies are sparse and rare, it is safe to presume anomalies are harder to recreate in Autoencoder and have a much higher reproduction error compared to normal behavior. We utilized two datasets from the WSU CASAS smart home project to validate our proposed method by comparing it to the other state-of-the-art approaches.