London, Montreal subways trial surveillance for public nuisance, suicide prevention

Public safety initiatives are going underground, as subway systems in the UK and Canada experiment with computer vision camera systems, but no facial recognition, for crime and suicide prevention.

Weapons, ‘anti-social behavior’ among alert triggers for Tube pilot

In September 2023, the London Underground concluded a proof of concept at Willesden Green Tube station for an AI-assisted “Smart Station” to provide video analytics and real-time data insights on customer behavior. A final report on the pilot lays out design principles for further iterations of the system and defines use cases and triggers, which range from counting customer entries and exits, to real-time alerts triggered by patrons evading fares, leaning over the tracks, vaping, sitting on benches for too long, or unfolding their e-scooters.

Although the partly redacted document specifies that no facial recognition is performed by the Smart Station platform, the pilot led to the testing of additional requirements for the fare evasion use case, including the unblurring of facial images to identify repeat offenders.

A report on the program from Wired says municipal transit operator Transport for London (TfL), tested 11 algorithms, which issued more than 44,000 alerts, 19,000 of which were delivered to staff in real time. It also catalogs the expected list of objections and concerns from privacy and consumer advocates, who worry in particular about the accuracy of a system that claims to capture nuanced behavioral biometrics that are prone to misinterpretation.

Warning signs tell AI when a Metro patron is likely to commit suicide

A somewhat more altruistic program is testing computer vision and behavioral analytics tools for preventing suicides in Montreal’s Metro system. The CBC reports that the Société de transport de Montréal (STM) and researchers from the Center for Suicide Intervention (CRISE) are working on artificial intelligence to scan CCTV footage for signs of people in distress.

Machine learning capable of recognizing warning signs can send real-time alerts to operators, who can take steps to mitigate harm. According to Brian Mishara, director of CRISE and a professor of psychology at Université du Québec à Montréal (UQAM), the algorithm can presently identify one out of four people who will attempt suicide, based on their behavioral signals.

Mishara says the system is a cheaper alternative to physical barriers or screens, which are still on STM’s wish list, but cost in the millions. The organization says it hopes to implement the AI system within two years.

Article Topics

AI  |  behavioral analysis  |  computer vision  |  London  |  Montreal  |  transportation  |  video analytics