Need Guidance for Building a Real-Time CCTV Surveillance Web Application Using Axelera Metis

Howdy ​@Nanthu! Loving this project - there’s simply no denying that for all the many and wonderful AI vision use cases, security and CCTV are the ones that are driving the way forward!

Maybe the best first step is to just get something up and running, and focus on the front end once that’s all confirmed and you’ve seen it in action: one model, one camera. YOLOv8n is already in the model zoo, and inference.py takes RTSP sources directly, so from your SDK directory try:

./inference.py yolov8n-coco-onnx rtsp://<your-camera-url>

First run will compile the model, so give it a few minutes. If that shows live detections from your camera, we know your card, install and RTSP feeds are all confirmed working, and everything else you’re planning then builds on top of that. It’s easy to jump ahead to the fun design work, but a solid foundation pays dividends here!

The model zoo reference and the application integration tutorial cover the input source formats and how to move from inference.py to your own Python application, including multiple streams and RTSP latency tuning.

For inspiration once that's running, there are a bunch of community projects in similar territory: a Home Assistant CCTV object detection build using Metis and RTSP cameras, and a deeper dive into a multi-model, multi-stream security system. There's also an earlier thread on streaming annotated video back to a web frontend that's close to your dashboard use case. And although right now you’re focusing on facial recognition, there’s some outstanding work on license plate recognition here, too.

And RetinaFace and FaceNet variants do exist in the model zoo, so the cascade face detection plus recognition piece will be a solid follow-up once your basics are rock solid and working. Can’y wait to see all this in action! Keep us posted on all your progress, won’t you!