ADLnet: A 1d-CNN for Activity of Daily Living Recognition in Smart Homes

Abstract

Human activity recognition (HAR) systems enable continuous monitoring of human behaviours in several areas, including activity of daily living (ADL) detection in ambient intelligent environments. The extraction of relevant features is the most challenging part of sensor-based HAR. Feature extraction influences algorithm performance and reduces computation time and complexity. However, the majority of current HAR systems rely on handcrafted features that are incapable of handling complex activities, especially with the influx of multimodal and high-dimensional sensor data. Over the last few decades, Deep Learning has been considered to be one of the most powerful tools to handle huge amounts of data. Thus, we developed ADLnet, a One-Dimensional Convolutional Neural Network (1d-CNN) for recognizing ADLs in Smart Homes, as part of an ambient assisted living framework to provide assistance to elderly inhabitants. We propose an innovative method to scan and classify time-series sensor data such as the CASAS dataset, which has been used for the training/validation/testing process. Testing results show very high performance.