Real-Time Anomaly Detection for Traveling Individuals
Tian-Shya Ma · 2009 · Proceedings of the 11th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '09) · doi:10.1145/1639642.1639712
Summary
This short paper presents a real-time anomaly detection system designed to help individuals with cognitive impairments travel safely and independently on routine journeys such as commuting via public transportation. The system works by comparing a traveller's current GPS trajectory against previously established normal route patterns to detect when they deviate from their expected path. Trajectories are modelled as a discrete-time series of axis-parallel constraints ("boxes") in 2D space, where each box summarises GPS data points collected at fixed intervals (e.g., every 10 seconds) using four values: minimum and maximum longitude and latitude. This box-based representation reduces computational complexity and storage requirements compared to raw coordinate tracking. The system compares incoming location boxes against norm trajectories using a similarity measure, with a threshold of 0.6 determining whether a deviation constitutes an anomaly. When an anomaly is detected, the PDA alerts the user through vibration and voice-over alarms, and an SMS notification is sent to a support team (referred to as a "shadow team"). The system was developed to support individuals with cognitive disabilities in community-based employment settings such as mail courier and parking patroller positions.
Key findings
The system was evaluated with 8 individuals with cognitive impairments (including TBI, intellectual and developmental disabilities, and schizophrenia) over 160 sessions across 2007-2008. Each participant completed 20 independent sessions — the first 10 trips established normal route patterns while the last 10 included arranged deviations. Without axis rotation, the system achieved 95.0% recall and 90.9% precision, with 4 false negatives and 6 false alarms. Applying a 45-degree axis rotation to the box model significantly improved performance, achieving 98.8% recall and 97.6% precision. The average elapsed time for sending location data, running anomaly detection, and issuing warnings ranged from 15.1 to 22.7 seconds, demonstrating real-time feasibility. Each travel session lasted 0.5 to 1.5 hours and included public transit use.
Relevance
This research addresses a practical barrier to independent community participation for people with cognitive impairments: the risk of getting lost during routine travel. Rather than providing turn-by-turn navigation directions, the system takes a less intrusive approach — it only intervenes when something goes wrong, allowing the person to travel normally without constant device interaction. This is an important design philosophy for assistive technology: supporting autonomy by monitoring rather than directing. The box-based trajectory comparison is computationally lightweight enough to run on a PDA in real time, making it practical for deployment. The involvement of people with cognitive impairments in real-world community travel over an extended period (rather than controlled lab settings) strengthens the ecological validity of the results. The system has clear applications for employment support, community integration programs, and caregiver peace of mind, though it raises important privacy and surveillance considerations that the paper does not address.
Tags: cognitive impairment · navigation · location awareness · anomaly detection · safety · independent living · mobile technology · ubiquitous computing