Using algorithms to interpret surveillance camera footage: the thin end of the wedge?
I was struck by a piece in Ars Technica about London Underground’s decision to use algorithms to examine footage from network of security cameras in real time to tackle fare dodging, violence,terrorism or other issues.
We’ve known for some time that algorithms can be used for this purpose: companies like Google use the billions of videos tagged on YouTube to train its algorithms. But real-time uses of this type have been less common until now, although it looks as though they will become more widespread in the future.
Transport for London, which runs the British capital’s metro system, has an extensive network of surveillance cameras, and their images will now be analyzed every tenth of a second. To monitor all of them systematically would be a major commitment of resources, which means that in most cases their use will be limited to serving as evidence once an incident has already taken place. Generally, it is security staff or passengers who report situations, often leading to delays to deal with issues.
If cameras are capable of detecting such situations and generating an automated alert, this could serve to generate faster responses and reduce the possible impact on the safety of public transport. That said, there are other issues, such as false…