Falls in homes of the elderly, in residential care facilities and in hospitals commonly occur in close proximity to the bed. Most approaches for recognizing falls use cameras, which challenge privacy, or sensor devices attached to the bed or the body to recognize bedside events and bedside falls. We use data collected from a ceiling mounted 80 × 60 thermal array combined with an ultrasonic sensor device. This approach makes it possible to monitor activity while preserving privacy in a non-intrusive manner. We evaluate three different approaches towards recognizing location and posture of an individual. Bedside events are recognized using a 10-second floating image rule/filter-based approach, recognizing bedside falls with 98.62% accuracy. Bed-entry and exit events are recognized with 98.66% and 96.73% accuracy, respectively.
Keywords: artificial intelligence; bedside event detection; classification; fall detection; thermal array; ultrasonic sensor.