Safety on the streets with new video intelligence
Atos Vice President, Head of Machine Intelligence
With thousands — if not millions — of people on the streets of a city every day,
the primary concern of that city’s public authorities will always be to keep those people safe. It should come as no surprise, therefore, that video surveillance has become a critical tool for the police and other agencies. At the same time, there are public concerns about the role of surveillance in modern society and how citizen privacy will be protected.
The reality is that while there may be thousands of cameras positioned in public spaces around a city, the pictures they capture are monitored by only a handful of operators. Keeping watch on all that visual information in real time and making best use of it during an incident can be a huge challenge. Now, digital capabilities can make a dramatic difference, giving operators crucial assistance to focus their activities on events that require their immediate attention.
Incorporating powerful machine learning and data analytics, a solution called video intelligence means that operators can now monitor the movements of an individual from camera to camera, and even predict where they will go next. This can be done in two ways: by helping operators quickly locate and follow a particular individual; or by enabling them to choose to follow an individual who is behaving suspiciously.
Lower crime, higher public confidence
In cases when a person needs to be found very quickly, such as a crime suspect or missing person, they can be pinpointed either by what clothes they are wearing or by using their photograph. In cases where a person seems to be behaving unusually, the operator can click that person on the screen, then follow them as they move around. In all cases, the operator gets a unified view of the individual’s movements; from there, they can decide to stop following, continue following, or send an appropriate person to an exact location to intervene.
With video intelligence already live in several cities, the benefits include lower crime rates — because people can be located and apprehended much faster, plus higher levels of public confidence — because the authorities’ responses can be swift and precise.
What does all this mean for citizen confidentiality and the use of personal data? In fact, video intelligence is fully compliant with data protection regulations, with no implications for the individual’s right to privacy. There is no facial recognition and all data and visual information is anonymous. There is no intention or ability to identify individuals, and data management rules are fully compliant with security policies for the use of video and for the city as whole.
Let’s look at an example. The city authorities have received an alert about a lost child. Based on the parents’ description of what the child is wearing, an operator opens up their video intelligence app, enters the clothing details, quickly finds the child in a crowd and follows her from camera to camera while a police officer is rapidly dispatched to the scene to ensure the child’s safety. There is no matching of any information about the child to any database, no correlation of any other information about her, and no long-term storage of information about her.
In addition to machine learning and analytics, what makes this kind of dynamic monitoring possible is real-time computing at the edge (that is, close to the camera itself). This means that the analytics can be generated rapidly, and cameras can communicate with each other to follow individuals across public spaces. With technologies advancing constantly, in the near future it will become possible to combine visual data with sound for an even richer real-time understanding of events.
Perhaps what is most interesting about this kind of technology is how it augments human intelligence rather than replacing it. Operators are still a critical part of the process: they just have the huge advantage of AI to help them focus more efficiently and make the timely, evidence-based decisions.