Insights
13 February 2023
Be Aware Of AI’s Appetite For Bandwidth: The Challenges Of AI In Video Surveillance
In our recent article, How AI Is Changing The CCTV Industry, we highlighted how UK installers may be missing out on the new business opportunities presented by AI. However, when installing a new system, there are challenges of AI in video surveillance that you need to take into consideration.
This possibility was supported by a report produced by Memoori Smart Building Research which predicts video surveillance product sales could increase at 5.7% CAGR and reach $44.8 billion by 2028. The report also estimates that the percentage of IP network cameras sold with built-in AI capabilities will rise steadily from around 18% in 2022, to over 50% by 2028.
Manufacturers such as Dahua and Hanwha Vision appear to be confident that going forward there will be high demand for AI supported electronic security solutions. They have demonstrated this by heavily investing in research and development (R&D) in order to harness the power of technology. In doing so, they intend to provide businesses, local authorities and other organisations with an affordable tool to keep one step ahead of intruders. As mentioned in the previous article (How AI Is Changing The CCTV Industry), as well as helping combat criminal activity and making remote monitoring more efficient, there are also countless ways in which AI can assist businesses to become more efficient and make best use of their human resources.
The Bandwidth Challenges Of AI In CCTV Systems
Visit any of the leading manufacturers’ websites and you are likely to find a bandwidth calculator which will help you estimate the network capacity required for a specified number of cameras, to truly understand the challenges of AI in video surveillance. The bandwidth requirement, i.e. megabits per second (Mbps), as well as the amount of data storage that will be needed, is calculated by taking into account factors such as the resolution of the camera, frame rate and the number of days the end-user wishes to retain the data.
As an example, an online bandwidth calculator may suggest that a network video recorder (NVR), which the manufacturer claims can handle 80 Mbps, should (without AI deployed), be able to adequately support up to 16 video streams and sub-streams if the installed cameras are 4 megapixel models or less. Similarly, a more powerful recording solution with a potential throughput of 512 Mbps, should be able to support 102 x 4 megapixel cameras. These calculations on based on an average stream size, but in reality, the amount of throughput capacity needed is likely to be affected by other factors, such as the varying activity taking place in the field of view and the impact of AI being deployed.
AI functionality adds a new dimension to a video surveillance solution which, whilst offering significant benefits, can exponentially impact on bandwidth and storage requirements. It could, in fact, potentially halve the number of video channels that can be supported, if the user is not prepared to sacrifice frame rate or quality of the images.
The latest compression technologies can help reduce video bandwidth requirements to leave the remaining process capacity available for handling the AI related data, but the use of compression also inevitably risks the loss of some video quality.
The overriding factor to take into account is that video file sizes are not linear or constant and you cannot assume that a 16 channel NVR will fully support any 16 cameras, regardless of what resolution is selected. With some NVRs set by default to operate at a constant maximum bit rate, to prevent them from being overloaded and crashing, other parameters are likely to have to automatically change, e.g. frame rate or image quality.
New technology to the rescue
The good news is that video surveillance manufacturers are responding to the challenge presented by AI’s appetite for bandwidth.
For example, the latest generation of Dahua IPC WizMind S cameras, which are equipped with the highly impressive AcuPick video search technology, can be configured to only transmit metadata and thus avoiding using up large chunks of bandwidth in order to transmit unwanted video. Instead, an operator can remotely retrieve recorded video of any activity or incidents if and when necessary.
Help Is At Hand
If you have previously installed AI equipped CCTV systems, you are likely to be aware of the challenges of AI in video surveillance and the need to accurately predict bandwidth and storage requirements at the design stage, in order to avoid disappointing your end-user client by underestimating the cost of taking full advantage of this exciting technology.
If, however, you have limited previous experience of installing these kinds of systems and do not currently employ technicians with this skill set, we would still urge you not to be deterred from competing for projects which involve AI. Instead, we would encourage you to take full advantage of the free design advice available from most of the leading manufacturers, who in a crowded and very competitive marketplace, will be keen to win your business and will therefore willingly guide you through the bandwidth calculation process.
Editors Note: This article has been updated to reflect the latest Memoori research and product lines developed by leading manufacturers