Video analytics are undergoing a fundamental change in the market as machine learning enhances their accuracy while expanding their capabilities. But what are those expanded capabilities and how are they impacting the operation of security and video systems? We asked this week’s Expert Panel Roundtable: What new video analytics are having an impact in the market and how?
Video analytics are improving the efficiency and accuracy of facility and enterprise security with the integration of advanced technologies. Technologies like AI and machine learning can help pinpoint potential threat locations and identify patterns not only within building systems, but in the buildings themselves. For example, as a standalone solution video analytics provide real-time insights into occupancy patterns – but when equipped with AI and machine learning, data collection is automated and analysed faster and more efficiently, offering recommended actions based on the situation. If video analytics identify an unauthorised visitor, AI and machine learning can notify security personnel who can then take appropriate action, whether it be escorting the occupant out or issuing a lockdown.
Re-identification and classification are two analytics that are gaining traction. Classification is the most important of the deep learning analytics, as from here all other deep learning analytics are built. This is because classification determines the type of object present: human vs. car vs. bus, for example. This is the first step in identifying and then tracking objects of interest. Re-identification is a deep learning analytic that allows the camera to recognise an unidentified person or object as a whole. For example, with classification, the camera can determine the difference between a bus and a car or a person and a horse. This is because, when the camera spots a person, it will ‘learn’ multiple data points including: height, size, and gait, the colour of clothes, the sex, face and ethnicity. Re-identification works by matching these multiple data points, but it does not need them all.
Although basic video analytics have been around for many years, today’s organisations aim to take this technology to the next level and use data and insights to help solve business problems and improve processes. For example, financial institutions desire the ability to leverage surveillance cameras for more than just capturing footage — they want to achieve tasks such as enhancing workforce optimisation and customer service. A challenge exists, however, in properly capturing, analysing, and processing the overwhelming amount of information that advanced systems provide. The key to successfully utilising video analytics is leveraging intelligent technology that can simplify and automate a large amount of data to determine what is critical and make it actionable. This then empowers operators to make informed and swift decisions based on reliable and detailed information.
Facial recognition is the most widely discussed analytic available today. Even so, the discussions are not always positive. Depending on where you are in the world, it is differently perceived as good or bad. The east seems more open to the technology, recognising that it streamlines and further authenticates the access control process. Individuals in the west are less likely to jump on board with the technology, fearing that their facial data will be wrongfully used. There is a big debate going on between the two sides, but one thing is certain: facial recognition is not going anywhere.
Video analytics has come a long way since being introduced more than 10 years ago. While initial results were mixed due to the technology not being mature enough, video intelligence capabilities have significantly evolved and are now a critical component of intelligent solutions that drive informed response and business results. For instance, video analytics can identify when an item is left behind in an airport and then alert the appropriate personnel to the situation. Alternatively, in protest situations, pattern-recognition algorithms detect changes in crowd activity and signal when behaviour shifts toward hostility. With the addition of facial recognition, we can automatically detect, track, and alert on persons of interest. Moreover, in an active incident, we can instantly alert operators, activate cameras and other controls in the vicinity and dispatch appropriate resources to mitigate threats. These examples are part of a drive to increase overall situational awareness to increase safety.