Join or Sign in

Register for your free asmag.com membership or if you are already a member,
sign in using your preferred method below.

To check your latest product inquiries, manage newsletter preference, update personal / company profile, or download member-exclusive reports, log in to your account now!
Login asmag.comMember Registration
INSIGHTS

AI thrust edge vs. cloud debate into spotlight

AI thrust edge vs. cloud debate into spotlight
This article looks at edge and cloud architectures and the merits/disadvantages for each, as well as whether other options, for example hybrid, can be considered.
Security now relies on a range of analytics, a lot of them AI-based, for different purposes. This then leads to the question whether the analytics should be run on the edge or in the cloud. This article looks at each architecture and its merits/disadvantages, as well as whether other options, for example hybrid, can be considered.
 
Video analytics now play a more important role to fulfill end users’ security and non-security objectives. For example, object detection can be used to detect whether unusual individuals/vehicles are approaching, while people counting can aid retailers seeking to meet capacity requirements.
 
The question then becomes where should these analytics, a lot of them based on AI and deep learning, be run – on the edge or in the cloud. Each has its advantages and disadvantages. Below we take a closer look.
 

Cloud

 
A major advantage of cloud is it can handle any amount of processing load. “If there is a sudden burst of events, edge systems won’t be able to keep up and there will be a backlog of events with high latency. This will never happen with cloud processing as cloud systems are designed to handle varying loads,” said Sam Joseph, Co-Founder and CEO of Hakimo. “Cloud allows you to use state-of-the-art sophisticated AI models as there are hardly any processing constraints for cloud systems. This allows the usage of advanced algorithms that would be impractical to run on an edge device. These algorithms can then be used to enable complex use cases for security teams.”
 

Edge

 
Edge processing, meanwhile, wins out by low latency and low bandwidth consumption.
 
“On-camera, or edge-based analytics, have the advantage of low latency, and can reduce bandwidth on the network because only the resulting metadata and relevant footage needs to be passed on to central systems, as opposed to a full video stream which is best for real-time use cases. Processing the video on a central server requires more bandwidth, as the full video stream needs to be transmitted to the server, which can require large amounts of bandwidth,” said Fabio Marti, VP of Marketing at Azena.
 
“Edge processing gives the user complete control over data security, data management and encryption processes. The system works independent of public Internet services, and can rely on internal networks. Processing times might be faster, data more readily available,” said Elke Oberg, Marketing Manager at Cognitec.
 
In fact, edge processing is gaining traction as cameras become increasingly powerful.
 
Using powerful AI chipsets from manufacturers like Ambarella, a world leader in automotive, IoT, robotics, and security technology, camera manufacturers like i-PRO have incredible AI-power at their disposal which even enables multiple analytics to run on a security camera simultaneously. Edge-processed analytics are the definitely the future for all AI-based use cases,” said Adam Lowenstein, Director of Product Management at i-PRO Americas.
 
The so-called edge appliances, meanwhile, can also be a viable edge solution if the user wants to keep their existing cameras.
 
“Traditionally, AI analytics are run directly on the camera; however, there are retrofit solutions, in the form of an AI appliance connected to the camera, that can enable these more traditional IP cameras to run edge-based analytics as well. For an end user starting from scratch or looking to upgrade their system, it may be better to purchase cameras that can run AI analytics directly on the camera, but for those not looking to invest in a new camera system, an AI box may be the best choice. The benefit of AI boxes is that they can be integrated with security solutions that would not otherwise be able to run AI analytics,” Marti said.

What to choose?

 
As mentioned, cloud and edge each have merits and disadvantages. So how to know which architecture to use? The user should decide by the application or use case that they plan to implement.
 
“Generally speaking, edge analytics are favorable for real time use cases and contexts where bandwidth limitations apply. Cloud or server analytics typically make more sense for after the fact analytics, such as forensics, or use cases where bandwidth is not a blocker or cost factor,” Marti said.
 
“Initial object detection, attribute extraction, and event information should be processed on the edge. If the use case requires the use of a large data set to compare with, and if that data set is frequently updating, then it can be further processed in the cloud,” Lowenstein said. “If it is something simple such as ‘find all images of a blue truck over a specified time period,’ then this type of searching can be handled completely on the edge in concert with a capable VMS. The user should decide based on the type of analytic that is required to be run.”
 

Hybrid

 
In fact, the line between edge and cloud is no longer so clear-drawn. An equally important question is how the user should use their IT resources most effectively. In this sense, a hybrid architecture can be ideal.
 
“You can have an on-premise appliance that does the basic processing and sends only the important parts of a video to the cloud for processing. That is a good solution for sites with bandwidth issues,” Joseph said.
 
“Users could store large databases on a cloud server, and then have the software and smaller databases on an edge system. They can send queries to the cloud system when necessary, otherwise use the smaller system locally,” said Oberg.
 
Marti, meanwhile, raised an example. “A hybrid approach can make sense in certain contexts. For example, one could use edge analytics on a camera for some simple detection capability, such as detecting a car, and real time license plate recognition to open a barrier. The detection of a car could then trigger footage to be passed on to a server for more sophisticated analysis, such as car model identification or passenger analysis. This way you can combine the benefits of low latency analysis and bandwidth conservation through edge analytics with the more scalable compute power of server side analytics,” he said.
 
In the end, analytics can be processed in the cloud or on the edge. Each has its own benefits and drawbacks. Ultimately, the user should choose based on their unique needs and requirements. From an IT resource optimization perspective, a viable option is hybrid, which is expected to see more demand and use cases in security.


Product Adopted:
Others
Subscribe to Newsletter
Stay updated with the latest trends and technologies in physical security

Share to: