An Acoustically-Analytic Approach to Behavioral Patterns for Monitoring Living Activities

Risk prevention and alarm are crucial for home care of aged people who live alone. In this paper, an acousticallyanalytic approach is proposed to extract behavioral patterns for modeling and monitoring living activities. An experimental environment was established for collecting sound tracking data of behaviors from daily activities. Each sound tracking data was transcribed to the corresponding sound event sequences by caregivers and psychologists. A living activities mining algorithm is applied to extract meaningful sequences of sound event as behavioral patterns. Then K-means algorithm is adopted to classify events into several fuzzy clusters relating to living activities for modeling behaviors in daily activities. An experimental database consists of 150 sound tracking data was established by collecting three guided behaviors included rolled out of bed to drink water, go to the toilet and watch the fish jar from 5 subjects in two weeks. 476 meaningful sequences of sound events were explored and clustered into 32 quasi-activities. These quasi-activities were further concluded into 5 kinds of behavioral patterns for modeling living activities. The preliminary result shows the potential for modeling and proactively detecting abnormal behaviors or changes of the aged.