AI Intrusion Detection System Based on EBC7700

■ Project Completion

An Edge AI intrusion detection system has been completed that analyzes USB camera video on-board using the EIC7700 SoC mounted on the EBC7700 EVK.

The system detects human crossing of a virtual line and entry into restricted areas, and transmits the detection results and captured images through an MQTT Broker to Node-RED for viewing on PC and mobile devices.

■ Key Achievements and Features

- Real-time human object detection using the EIC7700 NPU and YOLOv3.

- Automatic judgment of virtual line crossing and restricted area entry.

- On-site status displayed in three security states: normal, alert, and intrusion alarm.

- Intrusion events and captured images transmitted via MQTT.

- Node-RED web dashboard enables remote monitoring via PC and mobile devices, manual image capture, and image storage.

- Processing performed internally on the board rather than on a central server, reducing latency and network load.

■ Future Improvement Plans

Future improvements will include adding real-time video streaming and video recording capabilities, with expansion to a multi-channel surveillance system capable of managing multiple cameras simultaneously. An integrated dashboard for managing video from multiple camera channels will also be implemented.