AI is now a surveillance reality, but deploying it at the edge, in the cloud or hybrid is an individual organizational decision

Ask most companies about their future technology plans and the answers will most likely include AI. Then ask how they plan to deploy it, and that is where the responses may start to vary.

Every company has unique surveillance requirements that are based on market focus, scale, scope, risk tolerance, geographic area and, of course, budget. Those factors all play a role in deciding how to configure a surveillance system, and how to effectively implement technologies like AI.

At the Edge

One method that has been widely embraced by the industry is Edge AI.

The emergence of edge computing has transformed the way security and surveillance data is gathered, managed, processed and stored for efficient use. Instead of relying solely on centralized data centers in a virtualized cloud environment, edge computing involves processing data closer to the source, reducing latency and alleviating bandwidth constraints, among many other benefits.

Rapid advancements in edge computing intelligence and capabilities have also paralleled the maturation of cloud technologies, to the point where edge computing is now capable of performing tasks that, in the past, only the cloud could handle.

Depending on a company’s preferred model, some organizations will prefer to minimize network bandwidth and lower total costs by configuring an edge-based surveillance system. Others prefer a centralized cloud-based operation. Still others may opt for a hybrid approach, giving them the best of both worlds. In fact, many companies, including Hanwha, view this as an ideal ecosystem where edge and cloud solutions co-exist, with neither approach being “better,” and instead simply offering different sets of features and benefits.

Enter Edge AI

Occurring simultaneously with the acceleration of edge computing have been advances in deep learning (DL), a subset of AI. This trend has made it easier to bring AI capabilities on-board cameras, simply another way of saying it’s possible to enable more instances of AI locally, and effectively, at the edge.

Edge AI refers to running AI models on devices at the “edge” of a network, such as surveillance cameras, rather than managing them centrally using remote servers or in the cloud. With Edge AI, more functions are carried out on the camera itself.

Edge AI can be a long-term strategy on its own, enabling a company to migrate to AI at their own pace. Edge AI can also be a pathway to a hybrid on-prem/cloud alternative. In an Edge AI model, the majority of data is stored in the cloud, while only the most often used data is distributed over a network to the user’s fingertips. This significantly reduces bandwidth and storage requirements.

The key benefits of Edge AI include lower long-term operating costs by avoiding ongoing cloud service fees, as well as increased security and privacy benefits by keeping sensitive data processing on-premises and on the local network.

AI on a New Level

Another trend to consider also falls under the broader definition of edge computing: the Internet of Things (IoT), which is the entire universe of mobile and connected devices and technologies in use every day in our work and personal lives. The “things” can refer to any physical or software-based objects that feature a sensor, processing ability, and can store and send data to and from other connected devices – from “smart” home devices like thermostats and lights to fitness activity trackers, connected cars and surveillance devices.

All these IoT devices have one common characteristic: they are all basically minicomputers with increasingly powerful processing capabilities operating on the edges of a network.

IoT is not a new space, but it is continually growing. According to IoT Analytics, the number of connected IoT devices is expected to grow 13% to 18.8 billion by the end of 2024, and to more than 40 billion connected IoT devices by 2030.

Think of IoT as one gigantic edge computing network. A good example is the modern “smart” office building, where physical security and surveillance system devices may be networked and integrated with hundreds of commonly used building management systems such as lighting control, access control or fire control/suppression systems.

Each of these devices can be configured to work together resulting in a number of benefits: cost savings, increased energy efficiency and a more comfortable working environment.

Securing the Edge

Of course, once any degree of network connectivity or shared access is attached to an edge device, the potential for intrusion or vulnerability increases. This leads to a heightened need for securing the devices on the edge of a network, at the point where a company’s internet service comes onto the network at each endpoint before the traffic reaches a centrally orchestrated network.Security at the edge can be highly effective and the fact that it is decentralized gives organizations more options for managing their own unique security requirements. Securing the edge of an organization’s network computing also protects data and workloads in remote locations, which can be more vulnerable to threats and intrusions.Many customers running AI models on the edge will try to build a wall of security around their IoT devices, placing cameras in an isolated network that doesn’t have Internet access. That way, they are not as vulnerable to attacks. In this case, it is critical when you choose an IoT device or IP cameras to have many layers of protection against attacks. Edge security can also be configured in a layered approach, based on the idea that the more “walls” that potential bad actors must penetrate, the harder it is to ultimately reach the device.An Edge AI approach can help large enterprises with existing infrastructures of hundreds of cameras as well as smaller organizations just starting to adopt AI. Edge AI offers a flexible, cost-effective and phased approach for organizations to incrementally adopt and deploy AI capabilities in their video surveillance and monitoring systems.Edge AI may be better suited for applications that prioritize safety, efficient operation, and loss prevention, especially for markets like banking, retail, government, and military sectors that may have specific regulations about camera performance and data gathering.

Getting More out of Your Tech Spend

Edge AI can allow for easier deployment, as a company’s existing camera infrastructure can be used without the need to replace every camera all at once or invest in expensive servers. The users still receive help from full AI functionality and performance without the potentially burdensome maintenance costs associated with total camera replacement or VMS licensing – while at the same time getting the most use of their current technology expenditures.

It is a similar transition strategy to what the industry saw with the migration from analog to IP, when many companies used encoders to “convert” existing cameras, ultimately getting more lifespan out of current infrastructure without having to do a complete upgrade all at once.

There are also more solutions available to make the logistics of performing edge AI more cost-effective and efficient. Purpose-built AI hardware, like NVIDIA’s Jetson devices, further enables deploying powerful AI models on the edge. NVIDIA’s Jetson platform can be used to run specialized AI models and applications tailored to the user’s needs. With Jetson, customers can accelerate modern AI networks, easily roll out new detection models, and use the same software for different products and applications.

For example, Hanwha multi-sensor cameras feature an option to include an NVIDIA Jetson module to run AI models. The Jetson is a small board that can be inserted into a device and offers anywhere between 20 tops to 100 TOPS (Tera Operations Per Second), a significant upgrade when added on top of the camera’s original specs.

Hanwha’s AI Box also runs on the Jetson platform. The AI Box converts any camera supporting ONVIF/SUNAPI into an AI-enabled edge device with object classification and attribute extraction, avoiding the added cost of “ripping and replacing” an entire system all at once.

Devices like the AI Box are designed to support AI apps developed by third parties and targeted toward different verticals. Customers who have those applications can pick and choose new AI detections, functionalities and analytics to load into a device on a camera-by-camera basis. And it is still done at the edge, allowing them to use their existing infrastructure.

Customers can use solutions like these to customize their edge networks, again, without taking on the potentially significant investment of overhauling their entire infrastructure all at once. This approach allows companies to pick the best camera that suits a specific application and then supercharge it with an add-on solution, performing upgrades and replacements slowly over time at a pace that makes sense for their organization.

Even after a system has been deployed, a company can evaluate it holistically. Then they can make an informed decision about which cameras or views would be the most beneficial to receive added AI capabilities whether they are looking for analytics loitering, attribute extraction, clothing or color.

When to Consider Cloud

From a hardware perspective, if a company is satisfied with its current hardware infrastructure and does not plan to upgrade anytime soon, edge can be an ideal model. However, any hardware you buy is limited to what it can process. Of course, there will be improvements on the model and potentially new detections added, but it cannot compare to the cloud’s resources and ability to deliver immediate updates.

It is important to weigh the pros and cons based on organizational needs. Again, hybrid solutions provide a great deal of efficiency since users can run most of the needed AI models on the edge and only pay subscriptions for high-processing AI models.

A current hardware investment may run certain object detection models perfectly well and will continue to do so throughout the life of a device. But if a company wants to run more models, based on updated classifications, an edge device may only be limited to its own ability as of the time of its original design.

Cloud-based AI offers access to a deeper pool of resources and flexibility, and ongoing downloads of new firmware and updates. Edge AI models also produce valuable metadata, which can be uploaded to cloud-based platforms or imported into AI metadata visualization software programs. This helps generate valuable business insights to support data-driven decisions that can affect operations across an entire organization – offering a view into statistics that can optimize the customer experience, increase productivity, enhance profitability and more.

Companies today face increased security threats, and they are managing their operations with fewer resources and tighter budgets. They need options when choosing the types of solutions that will work best for them – and that solution may be found on the edge, in the cloud or a hybrid combination of both.