Computer vision: digital understanding of the physical world
From face recognition to fire prevention, autonomous cars to medical diagnosis, the promise of video analytics has enticed technology innovators for years. Video analytics, the processing and analysing of visual data through machine learning and artificial intelligence, is perceived as a significant opportunity for edge computing. The market for edge-enabled video analytics is projected to be worth $75bn by 2030, the second largest revenue opportunity behind cloud gaming.1
Over the past few years, investment in companies building video analytics software and platforms has boomed as AI models become increasingly mature and applicable across a wide range of use cases. From specific applications which manage production of specialised goods, to wide-range surveillance platforms monitoring crowds and workers, video analytics use cases are considered an early adopter of edge computing.
In this article we explore some of the more mature video analytics use cases and why edge compute is required for their proper function.
Distributed computing for security and surveillance
There are around one billion surveillance cameras in the world.2 Historically, the vast majority of this video is processed on the camera itself or watched by a person who manually triggers a response when necessary. As AI models mature, these CCTV cameras can be equipped with algorithms that automatically trigger an output in real time when certain conditions are met. Birmingham City Council has installed new AI-equipped CCTV cameras in Aston, Erdington, and Edgbaston to combat fly-tipping. The technology identifies dumped rubbish and notifies the authority, eliminating the need for manual video reviews after incidents.3
These networks of cameras use edge compute deployments to host these algorithms. To process live footage in real-time requires a level of computing which is difficult to house within the camera itself. Although current deployments of regional edge are few and far between, over the coming years surveillance use cases such as these will increasingly leverage distributed edge infrastructure to run more advanced algorithms. Given the wide-area, public nature of public surveillance, placing these workloads at the network edge makes logical sense.
Security use cases also offer significant value to the private sector. Advancements in facial recognition technology now allows CCTV cameras to identify individuals with up to 90% accuracy.4
Monitoring wide-area, outdoor, or remote locations lends itself to video analytics running on distributed compute infrastructure, allowing for the real-time analysis of upstream video. For industries such as oil and gas, where valuable assets and sensitive processes are protected by stringent authorization protocols, using automated video analytics software to cut down the risk of unauthorized entry for critical areas can offer significant ROI.5
In the short term, security, and surveillance use cases such as these will represent the largest revenue opportunity. The AI models which process the images are generic, meaning they can use broadly the same logic for multiple use cases, decreasing the need for alterations and therefore boosting profitability.
Worker safety for enterprise
Enterprises with stringent safety measures, such as manufacturing or oil and gas, are already using video analytics to improve their practices. With the annual cost of workplace accidents in the EU totalling EUR 476bn, and worldwide deaths totalling 2.8 million, using technology to reduce this number presents a real social and financial opportunity for enterprises.67
Worker safety in manufacturing is paramount. Video ingest and analytics have become essential in managing risks posed by hazardous settings and machinery, offering cost savings for businesses. Through high-definition cameras and real-time analysis, dangers and safety breaches can be quickly spotted. These systems allow rapid hazard detection without needing costly equipment, using easily-installed cameras. Real-time footage analysis can prompt immediate actions, like machine shutdowns, to prevent injuries. Some systems even identify workers without proper safety gear. However, challenges exist. Camera installation must be minimally invasive, and false alarms can hinder production. Despite challenges, the solution's applications are broad and fairly mature.
Smart city traffic management
Despite increasing moves to pedestrianize modern cities – alongside significant investment in public transport – urban areas such as London are still experiencing an increase in the number of drivers on its streets. Between 2011 and 2019, the number of miles driven on London roads rose by 3.5 billion, leading to an increase in carbon emissions and traffic, ultimately resulting in reduced productivity. Overall, this costs the UK economy £5.1bn of productivity every year.8
Companies like Yunex Traffic have been developing multi-modal systems that leverage video ingest for optimised traffic flow. Working with Transport for London, they have built an adaptive traffic control system that uses a variety of data, including video, to provide policy-driven traffic management to reduce journey time and reduce traffic. When using data from multiple sources, on-device compute quickly becomes inefficient and a shared infrastructure is required. Edge nodes offer the required compute within a short round-trip, delivering low-latency response times.9
Video analytics for production and process management
Although general security and monitoring systems are a significant short-term opportunity, video analytics has the potential to drive efficiencies in specialised systems across a whole host of verticals. Implementation of cameras across a site provides enterprises with tools to enhance operational and management efficiency, foresee production interruptions, and swiftly optimise manufacturing flow. One of the most valuable use cases today is inventory management, where analytics aids in accurately managing vast inventories and streamlining storage based on real-time data. Additionally, video analytics can detect defections, employing advanced machine vision techniques to spot imperfections, surpassing the accuracy of human inspections. Importantly, with the shift towards edge computing, these cameras can process video data in an edge infrastructure environment, maximizing privacy, reducing bandwidth usage, and offering real-time insights, thereby underscoring the value of video analytics in modern manufacturing environments.
The video analytics sector is evolving, with profit-generating use cases today and potential for significant growth in the future. Video analytics derives real benefits from edge computing, with advantages like reduced latency to create real-time insight. For the video analytics domain to optimise its capabilities, considering both telecom infrastructures and the nuances of edge computing is important.
 STL Partners – Edge computing market sizing forecast
 Comparitech, https://www.comparitech.com/vpn-privacy/the-worlds-most-surveilled-cities/#:~:text=At%20the%20end%20of%202021,to%20IHS%20Markit's%20latest%20report
 BBC, https://www.bbc.co.uk/news/uk-england-birmingham-66375533
 Recent Advances in Video Analytics for Rail Network Surveillance for Security, Trespass and Suicide Prevention—A Survey
 UN Global Impact: https://unglobalcompact.org/take-action/safety-andhealth
 Axelent, https://www.axelent.co.uk/world-of-axelent/knowledge/costs-for-work-related-accidents/