Ambient Assisted Living (AAL)

Project objectives

The project aims to develop an Ambient Assisted Living (AAL) platform based on video recognition, audio and multisensor integration technologies to monitor the daily activities of elderly people. The platform detects sudden dangerous events (such as a fall) and carries out a medium and long-term analysis of the behavior of the person being observed, to detect behavioral changes that could be linked to neuro-degenerative disorders.

The platform comprises a 2D and 3D vision subsystem, which interfaces with a set of RGB cameras together with a set of infrared structured light-based range sensors to obtain a 3D image of the scene. It analyzes the video streams with artificial vision techniques to extract information on the person's movement, position and posture, determining the position and pose of the observed subject. The recognition subsystem then integrates with data from other heterogeneous data sources such as (eg audio, smoke, etc.) to detect the occurrence of significant events.

The information produced is sent to the short-term activity recognition subsystem. This component recognizes complex gestures and actions. Sequences of gestures and actions are grouped into significant short-term activities, using a dictionary (or better, an ontology) that includes both normal behaviors and abnormal conditions that need to be reported.

The description of the short-term activities is used by the reasoning subsystem on the long-term activities to build models of the user's actions and activities, to analyze their evolution over time, to infer patterns of behavior and long-term trends , and to detect behavioral changes. An a priori knowledge of complex behaviors is exploited, expressed in terms of ontologies and high-level rules, using statistical Pattern Recognition and Machine Learning techniques.

Finally, a user interface subsystem aims to provide the person taking care of the observed person with information on his short- and long-term activities, which can be used to monitor the onset or progression of neuro-degenerative disorders, but also timely notifications of short-term alarm conditions that may require immediate action.

While the platform is applicable to a broad spectrum of clinically interesting behaviors, it is our interest to validate it on behaviors that are relevant to the two most common neurodegenerative pathologies: dementia and Parkinson's disease. To this end, the laboratory works on a set of themes for the elaboration of videos and images that adapt to the scenario described, with particular reference to the recognition of activities of daily living or common life (activities of daily living or ADL) and the how to learn the training algorithms when the reference context, the person to be monitored and/or the specific activities of interest change.

Start and end date

01/01/2013 - In progress

Project Manager

Prof. Giulio Iannello

Ing. Paolo Soda

Coordinating institution of the project

Università Campus Bio-Medico di Roma

Other Institutions involved

Eindhoven university of technology

Funding source(s).

Internally funded by the participating institutions