For years, surveillance upgrades were relatively easy to see. A better camera meant improved image quality, stronger low-light performance, or a wider field of view. More demanding analytics usually meant adding processing capacity somewhere else in the system.
More processing is now taking place inside the camera itself. Axis identifies the growing importance of edge computing as one of its key security technology trends for 2026, with more powerful cameras capable of analysing scenes and generating useful metadata locally.
For integrators, the design conversation is increasingly about how intelligence at the camera works with the wider surveillance environment.
The camera is becoming part of the architecture
A modern surveillance camera can carry considerably more responsibility than simply capturing a stream for analysis elsewhere.
Axis describes edge processing as increasingly capable of producing both actionable information from a scene and metadata describing the objects and activity within it. Its 2026 Perspectives report also points to lower latency, reduced bandwidth demand, and less dependence on central servers as benefits of moving appropriate processing closer to the camera.
The change is also visible in the hardware itself. Hanwha Vision’s Wisenet 9 system-on-chip uses dual neural processing units, with separate resources for image processing and AI object detection and analytics. That allows cameras built around Wisenet 9 to perform more AI processing locally without forcing image quality and analytics to compete for the same processing resource.
Milesight takes a different approach with its Open Platform, which allows supported cameras to run third-party algorithms and applications directly on the device.
Together, those developments show why camera selection increasingly involves more than imaging performance. The processing architecture and the ability to extend what happens at the edge can influence how the system develops over its lifetime.
Some decisions are better made close to the scene
Moving processing to the edge makes particular sense when an event needs to be recognised quickly and the decision can be made using information available at that camera. There is little value in sending every frame across the network to another system before deciding whether a clearly defined event has occurred if the camera can make that determination itself.
Bandwidth is part of the consideration, especially across larger camera estates or sites where connectivity is constrained. So is resilience. Local analytics can continue to perform appropriate functions without every decision depending on a constant round trip to the central processing infrastructure.
Scaling also looks different. Axis notes that edge processing increases available compute as edge devices are added, because each capable camera brings processing resources.
VMS turns edge intelligence into operational context
A camera recognising an event is only useful if the organisation can act on that information. Milestone provides a good example of how edge analytics can feed into a wider operational workflow. Its 2026 guidance on line-crossing analytics describes cameras performing the detection locally and sending structured metadata into XProtect. The VMS can then turn that event into alarms, rules, operator workflows, and searchable information.
The camera can handle immediate scene-level analysis, while the VMS provides a broader view across devices and events. Milestone also supports metadata generated by cameras and third-party systems for searching within XProtect, allowing operators to work with information about objects and recorded scenes rather than relying entirely on manual video review.
Cloud resources add another layer where broader analysis, management, and insight across distributed environments can make sense.
Hybrid surveillance is becoming a design decision
Axis expects hybrid architectures to continue evolving as edge and cloud capabilities become more significant, though it notes that most surveillance environments remain heavily on-premises today.
Where the processing sits should follow the job it needs to do. Some decisions are best made at the camera, while the VMS and cloud handle the work that requires a wider view. The architecture needs to reflect the workload rather than an assumption that one layer should perform every task.
Camera selection now includes a compute decision
For South African integrators, this adds another dimension to surveillance design. Resolution, imaging performance, storage requirements, and network capacity remain fundamental. Increasingly, the processing available inside the camera and the way that intelligence integrates with the rest of the environment also need attention.
Duxbury’s surveillance portfolio brings those different parts of the edge-intelligence story together. Axis provides an established ecosystem around intelligent edge devices and analytics. Hanwha Vision is pushing more AI processing into the camera through hardware such as Wisenet 9, while Milesight’s Open Platform extends what supported cameras can do at the edge. Milestone XProtect provides the VMS layer that can bring camera-generated events and metadata into a broader operational environment.
For partners, the value lies in being able to design around the customer’s requirements rather than forcing every project into one architecture. The amount of intelligence required at the edge, the way that information is used centrally, and how the environment may need to evolve can all influence the right combination.
The next surveillance upgrade may still involve replacing a camera. What has changed is how much of the system’s intelligence can now arrive with it. For integrators, deciding where that intelligence should live is becoming part of the design itself.
