i-PRO highlights how gen-AI on the edge can elevate system intelligence without rip-and-replace
Date: 2026/11/17
Source: Editorial Dept.
Generative AI has had a large impact on video security in the past two years, most notably through natural-language search in forensic investigations.
i-PRO is now going beyond that. Its latest edge AI cameras from the X-series Fisheye lineup no longer just enable users to search footage using queries in natural language, but also set up automated system responses. Effectively, they can tell their system in natural language to take a certain action if they detect a certain event, for example send an alert.
“Gen AI can now be used to describe situations that the camera should detect in real time,” Hiroshi (Huey) Sekiguchi, Executive VP for the Americas, i-PRO, told a recent webinar from i-PRO headquarters in Tokyo. “It marks a shift from natural language search to natural language detection powered by edge AI.”
In the webinar, the company showcased several use cases to demonstrate the new capability. In the first demonstration, the video camera detected a (staged) fight between two people and, in the second, it detected a person falling to the ground. The system had previously been set up using natural language to send alerts in such cases—which the camera did in the demo without significant latency. The third demo included a crowded scene in which all actors wore safety vests. When another actor entered the room without wearing one, the system flagged it as a safety breach and sent a notification.
The three examples are typical of what i-PRO suggests the technology is for: situations that are hard to define with line-crossing or intrusion rules, but easy to describe in a sentence.
“Security teams need faster understanding, not more screens and complicated workflows,” Sekiguchi said.
Five AI triggers
i-PRO also gave insight into how the AI in the new X-series cameras—powered by Ambarella CV72 SoCs and 12.4 megapixel image sensors—works in situations where the economical use of compute power is key. In general, this is an often-underestimated factor in setting up AI video security systems that offer tangible benefits.
The analytics on the new i-PRO cameras rely on “five types of triggers,” which Philippe Henaine, Manager, Strategic Partners, i-PRO, listed as “people, vehicles, bicycles, motion and area.”
“We are able to use any type of area in the image and request the camera to analyze it in order to detect events and send notifications without the involvement of server-based analytics,” Henaine explained, adding that data generated through real-time event detection on the edge feeds back into Active Guard, i-PRO’s VMS plugin for forensic investigation.
AI analytics for non-AI cams
Henaine estimates that the new CV72-powered devices have “six times more power than our existing performance,” which gives each camera ample processing headroom. Thanks to such efficient analytics, i-PRO can introduce another key innovation: Each X-series camera can be paired with up to three non-AI cameras and conduct analytics based on their images in real time.
“Each new Fisheye can effectively turn three legacy cameras into AI cameras,” Henaine said. “This brings great benefits to end users. They only need to update 25% of their camera infrastructure and get 100% of the AI performance.”
Other highlights
In the webinar, i-PRO also highlighted that the X-series’ AI capabilities are pre-installed on the camera. Users have access to the AI’s full capabilities without purchasing any additional subscription or license.
Meanwhile, i-PRO also reiterated a point it has made in previous webinars: that it seeks to ensure that its AI is used for ethical purposes. While there are no hard limits on what terms can be used in a natural-language prompt to the camera, Henaine pointed out that some factual limits on the results that can be achieved exist as the AI has been developed for ethical use.
“It’s not about limiting the use of AI, but about enabling the ethical use of AI,” Sekiguchi added.
Asked about whether the natural-language functions are limited to English, Sekiguchi said the functions work most accurately when the operator uses English. However, he also pointed out that “i-PRO is here to help” users who want to unlock the capabilities in other languages, for example by linking it up with one of the common large-language models from ChatGPT or Copilot.