Video analytics is software based on artificial intelligence. It detects and recognizes faces and other objects in a video stream and extracts various types of data from these streams. The global video analytics market is currently valued at $ 5.9 billion and is expected to reach $ 14.9 billion by 2026, according to MarketsandMarkets .
Video surveillance systems are no longer considered just a security tool. New use cases involve embedding video analytics technologies into a smart home, smart factory, smart seaport or smart city to create a comfortable and safe environment for everyone, including disabled people.
For instance, if someone stumbles and falls, and is then lying inert on the street, our systems recognize this and send out an alert. In another example city cameras collect traffic data and use this information to identify potential areas of danger.
For instance, you can identify a location where schoolchildren often cross the road in a certain location, and then introduce speed breakers or safety islands to reduce the potential for accidents. In this instance, facial recognition is not necessary, rather detecting silhouettes is sufficient.
Algorithms evolve into products
Early in the history of computer vision, developers were building their video analytics algorithms in academic environments using synthetic data to teach and essentially program algorithms. This was carried out in a vacuum, to understandably assess whether the technology worked. As in all areas of computing we’ve come a long way since then. Today, video analytics development is purely focused on practical use. For instance, companies create algorithms designed to solve real-life challenges and then release software as a ready-to-use product. A good example of this are self-service checkouts which allow customers to make payments based on facial recognition.
Today this productization is ubiquitous, ranging from biometrics to vehicle recognition, medical images analysis and more. Even academic researchers today use real data to teach their algorithms, with the focus on practical use.
All-in-one product concept
Early video analytics algorithms were developed and used in isolation. Recognition of faces, cars and responding to incidents were only available in separate applications. Today however, developers are combining several algorithms into an all-in-one package with interconnected analytics. This is for the benefit of users as this synergistic effect enables new use cases that were previously unattainable.
Today’s software suites typically include the recognition of faces, silhouettes, cars and other objects. Recognition of human actions is also coming soon. Users can manage all types of ‘objects' from a single user interface, and all of the object data is captured from the same cameras.
Among video analytics algorithms there are many that work fine but require large amounts of computing resources. This clearly doesn’t suit all users and as a result optimization is critical. For instance, hardware for large video analytics deployments requires significant investments.
As a result, developers are optimizing algorithms for fast operation on average hardware to help customers save costs. The less resource-demanding an algorithm is, the more customers can afford video analytics technology.
Requirements for ease-of-use of these new video analytics platforms are also growing. This has led to the development of systems that are almost plug and play. Users no longer need to spend time and resource configuring systems, they can simply unbox the product, press a few buttons, and the platform is up and running.
Predicting aggression today is a myth
While major metropolitan areas are generally becoming safer to live in, whether it be New York  or Moscow , we will remember 2021 for some striking incidents of gun violence, both on city streets and within the walls of educational institutions.
Dozens of people were killed and injured in an attack on a school in Kazan and a university in Perm. This led to a significant increase in interest in software that can detect guns and aggressive behavior, as well as software that can also predict aggression.
It’s already possible today to use video analytics to identify guns and dangerous actions, such as fights and people falling on the street. We expect this software to be operational within a real-world context, and with high levels of accuracy, as early as the next year.
However, artificial intelligence cannot yet predict aggression, at least by video analysis. Such predictive algorithms do exist; they are very much in early development. They are not usable in real life yet due to the large number of false positives.