Ambient and Wearable Sensor Fusion for Abnormal Behaviour Detection in Activities of Daily Living

Abstract

Abnormal behaviour in the performance of Activities of Daily Living (ADLs) can be an indicator of a progressive health problem or the occurrence of a hazardous incident. This paper presents an initial fusion approach of data collected from ambient (contact and thermal) and wearable (accelerometer) sensors in a smart environment to improve the recognition of the main steps of ADLs. An accurate recognition of these steps can support detecting abnormal behaviour in the form of deviations from the expected steps. The smart environment used is a smart kitchen and the ADLs considered are (i) prepare and drink tea, and (ii) prepare and drink coffee. These ADLs are deemed to have many occurrences during a typical day of a (elderly) person. The fusion approach presented considers the extraction of the most relevant features of the data collected from the two types of sensors (ambient and wearable) and the subsequent data analysis to recognise the main steps involved in the ADLs. Results show that this initial approach slightly improves the recognition of the main steps involved in the ADLs compared to the results obtained with just using data from the wearable sensors.

Fluid Intake Monitoring Systems for the Elderly: A Review of the Literature

“…Garcia-Constantino et al fused data from a wearable accelerometer, contact sensors placed around a kitchen (on kettles, cups, doors, cupboards, containers) and thermal sensors placed on the ceiling [ 165 ]. This study involved 30 participants all performing the same action of entering the kitchen, preparing a hot drink, consuming the drink and leaving.…”

“…This study involved 30 participants all performing the same action of entering the kitchen, preparing a hot drink, consuming the drink and leaving. No other activities were recorded (no Null class) [ 165 ]. Drinking actions were detected with 95% accuracy, but the overall system accuracy for 4 classes was only 73.51% since entering and exiting were often poorly classified [ 165 ].…”

“…No other activities were recorded (no Null class) [ 165 ]. Drinking actions were detected with 95% accuracy, but the overall system accuracy for 4 classes was only 73.51% since entering and exiting were often poorly classified [ 165 ]. The HydraDoctor by Du et al uses a commercial wrist accelerometer to first detect drink events and then triggers smart eye glasses (Google Glasses) to record a video [ 166 ].…”

DOLARS, a Distributed On-Line Activity Recognition System by Means of Heterogeneous Sensors in Real-Life Deployments—A Case Study in the Smart Lab of The University of Almería

“…However, the features and window size for evaluating binary sensors [ 19 ] or wearable sensors [ 29 ] differ. For that purpose, it is necessary to provide ad hoc processing for each type of sensor data [ 30 ]. Automatic extraction of features, such as Convolutional Neural Networks, present similar or slightly lower results in daily activity datasets [ 31 ].…”

Fusion of Unobtrusive Sensing Solutions for Home-Based Activity Recognition and Classification Using Data Mining Models and Methods

“…Many SSs have been deployed over the years for the purposes of activity recognition [17][18][19]. These have included the use of wearable or non-wearable solutions or the fusion of both.…”