We don't always realise it, but every day, dozens of cameras can capture our actions. Globally there are 770m surveillance cameras in use, according to a recent analysis of the world's most surveilled cities. And by the end of 2021, over 1 billion surveillance cameras will be installed globally.
So, the video surveillance camera is a tool that is already widely deployed but we are currently experiencing a pivotal period of technological transformation in the security market. The democratization of Artificial Intelligence and the availability of metadata from image analysis are reshuffling the cards in terms of video surveillance. These developments are as stimulating technologically as they are ethically: on both these fronts, Europe is ahead of the game.
The evolution of CCTV cameras
From military use to the democratization of domestic use, to the connected video surveillance we know today, the market is not new and has already experienced several small revolutions.
Initially, it was only a question of positioning a camera to record a defined space. One of the first major changes was the arrival of the IP camera - or network camera - which, connected via an Ethernet network, went from a simple sensor to a fully-fledged computer tool. It was at this time, in the 2000’s when the video protection market was booming, that the first artificial intelligence algorithms were developed and deployed on the systems, making it possible to analyse video flows and giving birth to the image analysis industry via security cameras. Thanks to the ability to analyse volumes of moving pixels in the image, these systems can, for example, detect an abnormal movement in the video, understand that an individual is present and report the information. The only downside is the accuracy of this analysis is conditioned by the environment in which the system is located. For example, heavy rain or strong winds can make detection difficult by raising false alarms or even missing an intrusion altogether. Another obstacle to these technologies is the cost of the infrastructure: in order to analyse videos and provide a relevant result, it is necessary to deploy computer servers with a large computing capacity. And these are not the only complaints that have been noted over the years. Security cameras are often criticised because they raise another operational problem: all these collected videos have to be processed. This requires a substantial human resource, especially in urban security centers.
But this does not take into account the recent evolution of video surveillance systems, which, boosted by advances in artificial intelligence, are now increasingly efficient and autonomous.
Deep Learning and metadata: the leap forward in video surveillance
It's a reality that the more precise the algorithms, the more proactive the cameras equipped with image analysis become in feeding information back to the user of a video surveillance system. The new generation of Artificial Intelligence today allows us to go further and make more precise pre-analyses of situations. In the case of a city, for example, the new algorithms will be able to draw the operators' attention to crowds or abnormal crowd movements. How is this possible? Quite simply by using a new method for designing these image analysis algorithms called "Deep Learning" that allows large volumes of data to be used to "train" Artificial Intelligence. When we "train" Artificial Intelligence systems with all this data, they are able to recognise different shapes: a man, a woman, a cat, but also to recognise colours, a man wearing a red T-shirt and a moustache, a child on a scooter, a particular face. All this is made possible by Deep Learning algorithms, capable of processing an exponential amount of data. Added to this are advances in terms of computing power and the miniaturisation of computer processors, which are now integrated directly into security cameras, making the cost of deploying these technologies much more accessible to end users.
Today's video surveillance systems have therefore become experts in pattern detection and recognition. Based on what they have detected, they create a meta-database that is self-powered and allows for an ever finer analysis of the images. This is the basis of biometric and facial recognition tools. Scary for the protection of personal data? Not really, because the regulatory framework in Europe strictly protects individual liberties (GDRP). From a security perspective, the aim of metadata is not to identify citizens (which is not allowed in Europe anyway), but to optimise the use of video protection, and make it proactive and predictive. Connected in-car cameras to avoid accidents, smart cities optimizing flows and limiting incidents, cameras that detect the micro-expressions of an individual having a stroke... the perspectives are unlimited and some are already a reality. From their conception, today's security systems are designed to go beyond a protective logic. By 2024, it is estimated that 30% of cameras sold on the market will be capable of embedding deep learning. Artificial intelligence-based image analysis is booming. It is now only a matter of time: the time needed to deploy infrastructures and to create the AI applications to positively change our society for the better.