Behavioral monitoring in smart-home environments for health-care applications

The contribution of our work is to design and experiment new methods to unobtrusively monitor the inhabitant of a smart-home, capturing the occurred fine-grained abnormal behaviors. The detection of those anomalies allows to support the clinical diagnosis of cognitive diseases; hence their models should be defined by experts in neuropsychology. It is important to note that the occurrences of those anomalies are not intended to provide an automatic subject's cognitive assessment. However, their frequencies and temporal trend can be used to derive behavioral changes. Differently from the other solutions, the occurred abnormal behaviors are inferred at a very fine granularity (e.g., “the subject is eating more cold meals than the usual”, “the subject retrieved a prescribed medicine from the repository but then forgot to take it“, . ..). Our specific contributions can be summarized as: a) the design of new supervised and unsupervised hybrid ADLs recognition algorithms which also deals with interleaved activities, b) the design of a fine-grained anomaly recognition framework capable to obtain a low number of false positives.