Edge AI Safety System That Independently Determines Restricted Area Intrusion and Issues Immediate Alerts

Practical Value of AI Direct Judgment Technology Recognized, Moving Beyond Simple Video Surveillance

An edge AI-based safety management system that independently determines restricted area intrusions and issues immediate alerts has been recognized for its excellence at a domestic technology contest.

Bang Jeong-bae, Senior Engineer at Asung, was selected as an outstanding work at the 'TechDay Contest 2026' hosted by e4ds for the 'EBC7700-based AI Intrusion Detection System' and received the Mouser Award, presented by the official sponsor.

This award is particularly significant in that it represents recognition of the practical value of technology in which artificial intelligence directly judges and responds to dangerous situations on-site, moving beyond simple video surveillance.

The project developed by Bang analyzes USB camera video in real time to independently determine whether a person has breached a restricted area, and when intrusion is detected, immediately transmits the situation through a web dashboard and warning system—an edge AI-based intrusion detection solution.

The starting point of the project was the structural limitation of existing CCTV systems.

General CCTV systems are limited to recording functions, requiring operators to continuously monitor multiple screens. In cases where large facilities or multiple areas must be managed simultaneously, there is a risk of missing dangerous situations due to reduced attention span.

To address this problem, Bang initiated development of an intelligent surveillance system in which artificial intelligence automatically recognizes people and generates alerts only in actual dangerous situations.

The core of the award-winning project is on-device artificial intelligence implementation leveraging the AI computation capabilities of StarFive's EBC7700 board.

The EBC7700 is a RISC-V-based single-board computer that provides AI computational performance of up to 13.3TOPS, allowing video input from a camera to be analyzed directly within the board.

Since there is no need to transmit video to a separate cloud server, the system offers the advantage of ensuring fast responsiveness and security.

The system operation method is also intuitive.

When camera video is input, the AI model detects people and distinguishes between safe state, approach state, and intrusion state based on a preset 'Cross Line.'

When a person approaches the boundary line, an alert is generated, and when entering the forbidden zone, it is determined as intrusion, transmitting status information and captured images via MQTT-based communication.

Subsequently, the Node-RED-based web dashboard displays intrusion alerts and captured images, with warning sounds output as needed.

This project received high praise particularly for integrating AI object recognition technology and real-time alert systems required in industrial sites into a single platform.

Rather than simply recognizing people, the system implements a structure that analyzes recognition results and selectively alerts only to actual dangerous situations.

This reduces unnecessary false alarms and helps operators manage sites more efficiently.

The development process was not without challenges. Bang unified the development environment to resolve errors caused by SDK and library version differences between the development PC and board, and completed the system by adding cross-line judgment and MQTT communication functions to the provided reference code.

Additionally, by separately managing status data and image data and conducting stage-by-stage verification, stability was enhanced. Through this trial-and-error and improvement process, the project achieved a level of completion suitable for actual on-site application.

This award is particularly noteworthy as it represents a case demonstrating the industrial applicability potential of edge AI beyond mere product technology verification.

In spaces where access control is critical, such as manufacturing plants, logistics centers, research facilities, and power plants, the system can reduce the burden of continuous human monitoring while raising safety levels.

In particular, the approach of performing inference directly on-site is expected to contribute to establishing smart safety management systems by reducing dependence on network environments and enabling immediate response.

Bang plans to further advance the system in the future. The plan includes adding real-time video streaming and recording capabilities and expanding to a multi-channel dashboard capable of managing multiple cameras simultaneously, aiming to integrate wider facilities. Through this, the goal is to develop the system from simple intrusion detection into a smart control platform.

Meanwhile, Mouser Electronics served as a platinum sponsor of 'e4ds Tech Day 2026' and supported this contest.

Bang Jeong-bae, Senior Engineer at Asung's 'EBC7700-based AI Intrusion Detection System' can be viewed at the link below.

https://www.e4ds.com/makers/project_detail.asp?id=1497