FAME-ADL: A data-driven fuzzy approach for monitoring the ADLs of elderly people using Kinect depth maps

Document Type

Journal Article




School of Science


Originally published as: Pazhoumand-Dar, H. (2019). FAME-ADL: a data-driven fuzzy approach for monitoring the ADLs of elderly people using Kinect depth maps. Journal of Ambient Intelligence and Humanized Computing, 10(7), 2781–2803. Original publication available here


This paper presents a novel unsupervised fuzzy approach to monitoring the activities of daily living (ADLs) of elderly people living alone. Kinect sensors are employed in functional areas of the monitored environment to capture unlabelled training depth maps from the occupant’s ADLs. This training dataset is processed in order to extract a set of ADL attributes, including the occupant’s location and 3D posture. A novel data-driven technique is presented to define fuzzy sets over ADL attributes, and membership functions are learned based on the attribute data distribution. These fuzzy sets are used to transform the attributes’ values into fuzzy labels in order to linguistically describe the occupant’s behaviour patterns. A set of fuzzy rules is generated to model the frequency of behaviour patterns using a novel data-driven technique. Abnormal behaviours in subsequent data are detected as unusual behaviours that last for longer than a learned duration. The proposed monitoring approach was evaluated using a dataset collected from a real-life setting. The fuzzy rule set obtained from the output of the proposed membership function generation technique was found to improve monitoring of the elderly because it could accurately classify more testing scenarios of normal and abnormal behaviours when compared to the rule sets obtained using other techniques. Compared to the output of other parameterising techniques, this fuzzy rule set had fewer rules and greater tolerance of fine variations in activities. The experimental results also showed that the adopted fuzzy approach for monitoring ADLs outperformed existing camera-based approaches in terms of its classification accuracy for testing scenarios of normal and abnormal behaviours.