The evolution of AI is transforming the nature of video analytics and its role in the surveillance sector.
Advancements in Artificial Intelligence (AI) have come a long way since the classic object detection algorithms of the early 2000s. No longer do users have to be bombarded with false positives and a cluttered arrangement of detection algorithms. The new era of video analytics has AI at its core.
Built on reliable datasets, the AI of today allows for targeted detection that supports its users in detecting safety and security threats without overwhelming them. Delivering structure in detection, AI goes beyond object detection and opens the door to event detection. For example, whilst historical video analytics may have detected the presence of new objects in a scene, AI-driven analytics can link the object to a pattern of activity.
Traditionally viewed as an amenity, the employment of video analytics is becoming ever more ubiquitous in video surveillance. Advancements in algorithms have opened detection capabilities into new markets and verticals. To make the most of these advancements, security integrators and professionals need to differentiate between the AI-driven video analytics of today and the classical video analytics of yesterday. They must also recognize the complexities in detection, its limitations and how best to apply it to the environment.
Understanding the limitations
Whilst it can seem compelling to detect a wide range of objects and activities, in the active environment this will do nothing but accelerate alarm rates and overwhelm users. Achieving the sweet spot between detection rates, and the user’s capacity to review and respond is key to the successful integration of video analytics into the wider system.
Whilst facial recognition has been the sweetheart of video analytics over the last few years, not all facial recognition is the same. Detecting a face in a crowd is very different from verifying a face against an existing catalogue of faces in a structured environment. Whilst we would expect to see low false positive rates with one-to-one matching in a structured environment; detecting a face in a crowd using standard surveillance cameras is a very different beast. Although some may promise high rates of accuracy, with big claims must come conclusive evidence.
Ultimately it is down to the experts to be honest and direct about the limitations and variations in accuracy their technology has to offer. Whilst it can be compelling to exaggerate capabilities as a route to market differentiation, usability should be at the core of all commercial video analytics products. Although detecting a single face in a crowd of thousands might be desirable, is it achievable. When it comes to video analytics, variables matter, beards occlude a third of the face, ageing effects contracts and symmetry, and finally data out is only as good as data in.
The future of video surveillance
As video analytics proceeds to become more and more mainstream, innovation in applications in the surveillance sector will continue to evolve. Whilst the detection of high-level threats such as gun crime will continue to have its place in the market, the detection of lower-level criminal activity will open.
Beyond that of market growth, innovation will continue to accelerate, and the autonomy of the machines will grow. Capabilities such as semi-automated response procedures and data fusion will become more and more reliable and interoperability between sensors will develop. Users will continue to hand over cognition-heavy tasks to computers and a greater level of situational awareness will be achieved.
Underpinning all of this is innovation and market adoption, allowing new technologies to claim their stake in the market. It is key for governments and industries to continue to invest in and embrace new technologies. Identifying new ways to collect, verify and utilise data whilst maintaining high levels of standards and reliability.
About the author:
Eleanor Wright, COO TelWAI
Holding a BA in Marketing and a MSc in Business Management, Eleanor Wright has over ten years of experience working in the security sector. Formally based out the UK’s largest robotics laboratory, Eleanor has project-managed multiple innovations in computer vision and machine learning.
You can follow TelWAI on LinkedIn and Twitter or visit thier website to know more: https://www.telwai.com/
About TelWAI:
Historically, surveillance has been built on complicated and expensive networks of cables and low autonomy. TelWAI believes the future of surveillance is built on wireless mobile networks that can be rapidly deployed at low-cost.
Founded in 2022, TelWAI brings together expertise in machine vision and video surveillance. Established with a goal to make surveillance agile and targeted, TelWAI is built on the back of innovation. Their cameras are not limited by location or image processing, they can be deployed and redeployed as safety and security risks develop.
