Video-based activity level recognition for assisted living using motion features

Activities of daily living of the elderly is often monitored using passive sensor networks. With the reduction of camera prices, there is a growing interest of video-based approaches to provide a smart, safe and independent living environment for the elderly. In this paper, activity level in context of tracking the movement pattern of an individual as a metric to monitor the daily living of the elderly is explored. Activity levels can be an effective indicator that would denote the amount of busyness of an individual by modelling motion features over time. The novel framework uses two different variants of the motion features captured from two camera angles and classifies them into different activity levels using neural networks. A new dataset for assisted living research called the Sheffield Activities of Daily Living (SADL) dataset is used where each activity is simulated by 6 subjects and is captured under two different illumination conditions within a simulated assisted living environment. The experiments show that the overall detection rate using a single camera setup and a dual camera setup is above 80%.