AI cameras use onboard AI analytics to process various data. Putting intelligence on edge devices such as AI cameras has its share of benefits and disadvantages. This article takes a closer look at the pros and cons of putting AI on the edge
Pros
Running analytics on cameras instead of in the cloud has various advantages. These are summarized as follows.
Lowered total cost of ownership
While the initial investment for purchasing AI cameras may be higher, the long-term operational costs are significantly reduced. “Cloud AI incurs recurring costs based on data transfer, storage, and compute usage, which can escalate rapidly with continuous, high-volume data streams. Edge AI agents, by processing data locally and only sending summarized insights to the cloud (if at all), dramatically cut down on these ongoing cloud expenses, offering a more predictable and often lower total cost of ownership,” said Shay Kamin Braun, Director of AIoT Product Marketing at Ambarella.
Higher quality of AI results
According to Braun, AI on-camera processing is done on raw video before encoding – hence preserving all the details of the scene and at the full resolution of the sensor. “This is in contrast to the cloud where AI is done on a video stream that has been scaled/encoded/decoded, which usually leads to some quality degradation, especially when trying to detect smaller objects at longer distances,” he said.
Reduced latency and bandwidth usage
Another key benefit of running AI analytics directly on cameras is providing real-time intelligence without increased bandwidth or cloud processing requirements. “Edge-based analytics reduce latency in mission-critical applications and reduce the need to stream large volumes of video offsite. Instead of sending raw video for analysis, they perform AI analytics at the source, generating only lightweight metadata for important events. This can dramatically reduce bandwidth, storage, and compute cycles on back-end servers wherever they may be located – on-premises or in the cloud,” said Adam Lowenstein, Americas Product Director at i-PRO.
Other advantages
Putting intelligence on the edge also allows increased privacy, security and reliability. “Increased privacy enables compliance with stringent data sovereignty and privacy regulations; higher security minimizes exposure to network vulnerabilities and centralized data breaches; and higher reliability is achieved by not relying on cloud connectivity and availability. This also brings greater efficiency and resilience in disconnected or intermittently connected environments,” Braun said.
Cons
While there are benefits to running analytics on the edge, there are also drawbacks and limitations which are listed as follows.
Compute resource constraints
One of the disadvantages is compute resource limitations in AI cameras. “Some cameras have constrained compute resources, which may restrict the complexity or number of AI models that can run simultaneously. Thermal management and power consumption are also considerations when packing powerful processors into compact form factors,” Lowenstein said.
“Cameras are small, meaning their hardware is limited, and they often use low-power processors to conserve energy and reduce heat. Complex analytics, like facial recognition and behavioral analytics, require more computational resources,” said Sean Um, GM and Head of European Business Development Team at IDIS. “Edge devices also have storage limitations. Complex analytics require storage of large datasets or deep learning models, which isn't feasible on many of today’s cameras. They may also lack sufficient RAM to handle large computations.”
Camera analytics upgrade challenges
Another challenge is upgrading analytics on cameras. “Updating analytics algorithms on numerous edge cameras is harder than updating a centralized surveillance solution. Security patches and model upgrades can also become logistically challenging,” Um said.
Higher initial investment
Finally, as aforementioned, AI cameras carry a higher price tag. While the TCO may be lower down the road, user needs and requirements will be taken into consideration.
“Most users don't require complex analytics. Essential detection capabilities, such as those that mitigate typical security threats and health and safety risks, will meet the needs of most customers. Therefore, there's currently no significant demand for common applications or run-rate type installations that need advanced features on the edge,” Um said. “Some high-end model cameras may offer advanced capabilities at a significantly higher upfront price point, and they may be useful for niche applications; however, it’s the trade-offs between cost, power, and performance that are major considerations for stakeholders.”