It is common knowledge that CCTV systems have become so cheap that surveillance cameras are installed almost everywhere they can be and where are needed, but also that someone must analyze what those cameras record and make evaluations about whether human presence is needed at a particular spot - and that someone must be paid.
But if the presented research is anything to go by, in the not-so-distant future these people will no longer be needed.
The researchers started by analyzing the core features of human "visual intelligence", and then simulating this faculty through computer vision algorithms complemented with higher-level tools of analysis involving, in particular, knowledge representation and reasoning.
In short, the system is able to identify relevant actions and spot the most anomalous ones, and to alert humans when that happens.
The researchers say that there are many improvements that need to be made in order to make the system good enough for actual use, but they are working on it.
For more details about the results of the research and future plans, download their paper.