In the beginning there was analogue Closed Circuit Television (CCTV) where computer analytics did not exist, and it was up to those monitoring the pictures to spot as much as they could humanly do so – quite a tall order when you only had one or two operators watching banks of camera feeds! Then came IP video, which transformed surveillance systems by enabling video transmission over network cable, and IP cameras could be easily attached to networks.
Now that video was essentially digitised, developers soon realised that digital video could be analysed automatically by the clever use of algorithms. This became a growing feature of ever more sophisticated video management software (VMS), which within certain parameters could do things like detect movement or stationary objects, set virtual tripwires and count numbers of people in, for example, retail environments.
Originally, all these analytics were being carried out at the centre of the video surveillance system, at the server or NVR loaded with the VMS. A major drawback, however, was the amount of expensive bandwidth that was needed to transmit large amounts of video to the VMS, before they could be processed and analysed.
In the past few years, however, we’ve witnessed a trend towards more analytics taking place at the camera end – also known as the edge of the network. This approach has the immediate advantage of being able to analyse camera images locally, without having to send bandwidth-hungry video across the network. Instead, the results of the analysis are sent to the VMS as lightweight data. In practical terms, this means that a camera fitted with motion detection, for example, will only start sending images if and when any movement is detected. This has the advantage of reducing the amount of bandwidth and storage needed, and not needing an operator to monitor video when nothing is happening.
Another benefit of this distributed network architecture is recording at the camera end. This enables the live viewing of video at relatively low resolution, so conserving bandwidth, while recordings are made at higher resolution for post-event analysis and high quality evidence. Of course, the actual recording device does not have to be in the camera – NVRs can be used to record at the edge for a group of local cameras. And if an edge device should fail, only that part of the network will be affected. Because of its distributed architecture, recording at the edge also comes into its own in mobile and/or wireless applications, where bandwidth is at even more of a premium.
Edge-based systems can also be less costly, as unlike server solutions there is no need for server maintenance, and there are no proprietary software licenses to purchase and potentially no limits to scaling up the system.
What is making edge recording at the camera more feasible is the exponential growth in flash memory devices, such as SD Cards and USB drives, which can now hold several Terabytes of data.
Edge vs central
But one of the downsides of recording and analysing at the edge is the expense of cameras with sufficient processing power, whereas it may be less costly having cheaper cameras and leaving the processing to be done at the server. As most VMSs are camera-agnostic, there is also a wide range of cameras to choose from. While server-based systems do require more bandwidth, they have the advantage of being able to decide which analytics are assigned to which cameras. There’s also the consideration that recording and analysing at the edge could leave data more vulnerable to hackers and vandals, compared to the relative security of centralised storage.
As we know, IT does not stand still, and there’s already a third way to implement video analytics – via the Cloud. This method offers the advantages of server-based systems – such as centralised, top-down control and powerful AI analytics – but without the cost and maintenance needs of servers. It’s typically delivered as Video Surveillance as a Service (VSaaS), so obviating the need for upfront costs, and can cover video recording, storage, remote viewing, management alerts and cyber security.
“This top down approach is definitely influencing end users to believe that cyber security
issues can best be delivered here,” says a report from Memoori. “AI in the cloud looks to best offer improvements in analytics, supporting technologies around bandwidth availability and a stronger infrastructure. While AI and Cloud are still niches within video surveillance today, we expect them to be the major drivers of new product development in the 2020s.”
Asking security professionals whether they prefer edge-based or centralised systems does not elicit black and white answers.
Commenting on their relative merits, Andrew Sieradzki, Group Director for Security at Buro Happold says: “Choosing whether to go for an on edge-based or centralised surveillance system depends on a number of factors. The client’s aspirations, requirements and attitudes to risk and resilience, as well as cost and value for money are key considerations. There’s no ‘one size fits all’ solution and you need to work closely with the client, understand their business and objectives to engineer the solution around the client.
“It also depends on which features are required – video analytics, for example, need lots of processing power and an IT infrastructure to support that will need careful thought. The geography of the installation will need to be considered, as well as different legal jurisdictions if the solution covers several countries. A hybrid system is often a good way to go, as a system can be closely designed to the specific requirements of the client.”
Independent Security Consultant, Simon Lambert, has a slightly different take on the dilemma. “The issue I have with edge-based systems is when the recording is stored at the camera, it is potentially vulnerable to hackers and vandals. That also raises a question about the duty to protect data under data protection legislation.
“In addition, there’s potentially a lot more processing power at the centre than there is at the edge in a camera, and complex analytics at the edge might not be feasible. However, if you have sufficient processing power in the camera, early processing at the edge may be OK. In the end it depends on how sophisticated the requirement is.”
Of course, like most electronic security systems, there is a middle way – a hybrid of recording and/or analytics at the edge combined with a degree of further processing at the centre. This method could be especially beneficial to enterprises with several large sites and those with numerous remote locations. Choosing the best solution is also dependent on the kind of analytics required, and whether the camera will be sending video signals or just metadata.
Whatever option users go for, there is a whole host of alternative ways of designing a sophisticated networked surveillance system, so think bespoke rather than off-the-peg and get out that blank sheet of paper.