Processing data on the edge and in the cloud each has pros and cons. Combining AI cameras and cloud together results in hybrid solutions, which offer the best of both worlds.
Putting AI on edge devices such as AI cameras has various advantages and disadvantages. Processing data in the cloud, meanwhile, also has its share of pros and cons. Combining AI cameras and cloud together results in hybrid solutions, which offer the best of both worlds.
Edge advantages
As reported in a
previous article, edge offers various benefits. Processing can be done close to the data source, reducing latency and bandwidth consumption. The quality of AI analysis is also better, as processing is performed on raw video before it’s encoded.
A word on edge AI boxes
It should be noted that increasingly, the so-called edge AI boxes are used, especially in deployments where legacy systems are still used.
“We see a trend where companies are delivering AI boxes to enable the quick addition of AI for installations where some of the cameras are not yet AI-enabled (or are not GenAI-enabled). Adding one or several AI boxes to a security installation is a lot less costly than having to upgrade all of the installed cameras, or upgrading the backend NVR/storage infrastructure,” said Shay Kamin Braun, Director of AIoT Product Marketing at Ambarella.
“Edge AI boxes can handle real-time processing, which is essential for security and safety operations. They can also process data from multiple cameras, integrate with other sensors, and support wider business applications, such as ERP systems, offering scalability. For example, a small retail store can achieve complete coverage with a single fisheye camera and a 4-channel AI box, providing the essential in-store and customer behavior analysis they need at an incredibly competitive price point,” said Sean Um, GM and Head of European Business Development Team at IDIS.
Cloud pros and cons
Processing data in the cloud has various advantages as well. Among them is the ability to leverage the cloud’s almost infinite compute resources.
“It gives you access to far greater processing power than what’s available on individual cameras or local servers. This makes it possible to run more advanced models, integrate data from multiple sources, and perform higher-level reasoning, like identifying individuals across multiple cameras or correlating video with access control and sensor data,” said Florian Matusek, Director of AI Strategy and MD at Genetec Austria.
Matusek adds: “Cloud-based analytics also make it easier to manage and scale your system. As long as your system is based on an open architecture, you can centrally configure rules and analytics across sites, regardless of the camera brand or model, which simplifies operations and future-proofs your deployment.”
Yet cloud processing has its share of drawbacks and limitations, too. “There are clear drawbacks for video analytics in the cloud: latency, bandwidth costs, and uptime dependencies. Sending high-resolution video to the cloud for real-time analysis introduces delays and raises costs quickly, especially at scale. For time-sensitive, mission critical applications, a hybrid approach which utilizes the edge is often the best solution for many customers,” said Adam Lowenstein, Americas Product Director at i-PRO.
Hybrid architecture
This is where a hybrid architecture can come in handy. Hybrid systems are now seen in most real-world security deployments, where the cloud handles the more complex and compute-intensive analyses and the edge performs real-time detection and related tasks close to the source.
“A hybrid architecture offers the best of both worlds. Edge devices handle local, real-time decision-making – critical for safety, operational efficiency, and uptime. The cloud then plays a supporting role for centralized access, compliance storage, and strategic insights,” Lowenstein said. “Hybrid architectures are inherently more resilient, cost-effective, and scalable. They also simplify compliance with data sovereignty laws by keeping sensitive footage on-prem while pushing only necessary data or metadata to the cloud. For i-PRO, this hybrid model – edge-first, cloud-supported – is the most realistic and sustainable approach to building modern video surveillance systems.”
Braun echoes Lowenstein’s comments. “A hybrid approach – that is, running some of the analytics on-camera, some on an on-premise edge AI box, and/or some on the cloud – allows vendors and service providers to optimize camera networks to benefit from the advantages of each of those,” he said.
Braun adds the following are examples of where each element in this hybrid approach is best suited:
On-camera analytics are implemented to process input video with high privacy, high security, low latency and high availability;
Edge AI boxes are used for quick installations that upgrade AI without having to replace all cameras, as well as for advanced/complex analytics, and for multi sensor/multi-modal inference;
Cloud resources are used to run analytics that don’t require the same level of privacy, security and latency, as well as for newer and larger models that may not fit on previously deployed hardware or that may require frequent updates.