Verification of Daily Activities of Older Adults: A Simple, Non-Intrusive, Low-Cost Approach
Loïc Caroux, Charles Consel, Lucile Dupuy, Hélène Sauzéon · 2014 · Proceedings of the 16th International ACM SIGACCESS Conference on Computers & Accessibility (ASSETS) · doi:10.1145/2661334.2661360
Summary
This paper presents a sensor-based approach to verifying whether older adults perform their daily activities at home, with the goal of supporting aging in place and informing caregiver decisions. The key conceptual distinction is between activity verification and activity inference: rather than using complex machine learning to recognise arbitrary activities from sensor data, the system verifies whether specific known routines have been performed. This is possible because older adults increasingly organise their daily activities into strict routines as they age — a phenomenon called age-related routinisation — meaning their activities become predictable enough to verify rather than infer. The system uses three types of low-cost, non-intrusive, wireless sensors: motion detectors (detecting presence in a room), contact sensors (detecting opening of doors, drawers, and cabinets), and smart switches (detecting appliance use such as a coffee maker). Sensors are placed at strategic locations in the home based on a prior routine-sketching process where an ergonomics-trained researcher interviews the older adult about how they perform each activity. The researchers developed activity-specific verification formulas grounded in three criteria: spatial context (which room), temporal context (expected time of day and minimum duration), and environment interactions (specific markers like opening the wardrobe or using the coffee maker). The study monitored four community-dwelling older adults (mean age 83.5, all female, living alone) in their own homes over five weekdays, tracking three activities: getting dressed (a basic ADL), taking a shower (a basic ADL), and preparing breakfast (an instrumental ADL).
Key findings
The verification formulas achieved high accuracy compared to a naive human observer manually analysing the same sensor logs. Using Signal Detection Theory, the meal preparation formula achieved perfect sensitivity (A'=1.00) and no response bias (B''D=0.00), meaning it matched the human observer exactly. The shower formula was highly sensitive (A'=0.94) but conservative (B''D=1.00), meaning it occasionally missed showers but never produced false positives. The getting dressed formula was also highly sensitive (A'=0.93) with a slight conservative bias (B''D=0.39). Longitudinal analysis over the five weekdays showed that breakfast preparation and showering were reliably detected as once-daily events, while getting dressed was more variable — sometimes undetected, sometimes detected multiple times per day — likely because opening a wardrobe is not unique to dressing. The approach deliberately excluded cameras and body-worn sensors, as older adults in prior interviews massively refused cameras and found wearable sensors intrusive. RFID tags were also excluded because they require tagging objects and cannot adapt as new items enter the home. The researchers validated that their participants did follow strict routines through standardised assessments including the MMSE (cognitive screening), time-based IADL test, and a routinisation scale. The system is limited to single-occupant homes, as multiple occupants introduce attribution errors.
Relevance
This research addresses a critical need in aging and accessibility: enabling older adults to remain living independently at home while providing caregivers and health professionals with reliable information about functional status. Activities of daily living (ADLs) are the standard clinical measure of an older person's independence, and their decline signals the need for increased support. The verification-based approach is notable for its simplicity and respect for privacy — using only three types of inexpensive sensors, no cameras, no wearables, and no complex machine learning. This makes the system practical for real-world deployment in ways that more technically sophisticated approaches often are not. For accessibility practitioners and technology designers working in aging-in-place contexts, the paper offers important design principles: monitoring should be non-intrusive and privacy-respecting; older adults' established routines can be leveraged rather than fought; and simple verification may be more reliable and deployable than complex activity recognition. The routine-sketching methodology, where the system is personalised through structured interviews about individual habits, provides a model for person-centred technology design. The limitations acknowledged — single occupancy, coarse activity granularity, small sample — point to areas where further research is needed but do not diminish the practical value of the approach.
Tags: aging · activities of daily living · activity recognition · smart home · sensors · independent living · pervasive computing · older adults