Build Your Own AI Security Camera with WebRTC (Step by Step)

By Naresh · October 13, 2026 · Tutorials

Smart cameras with "AI person detection" usually come with a monthly cloud fee. You can build the same thing at home with an ordinary IP camera, a small computer and free software. You get a live view in your browser with sub-second latency, plus alerts when a person shows up.

What you need

• Any IP camera or CCTV DVR that gives an RTSP stream (most do)
• A computer to run it: an old laptop, a mini PC or a Raspberry Pi 5
• Python 3.10+

How it fits together

Camera (RTSP) ──► MediaMTX ──► WebRTC ──► Browser (live view)
                     │
                     └──► Python + YOLO ──► "Person detected" alert

MediaMTX does the video plumbing. A separate Python script reads the same stream and runs object detection. Keeping them separate means the live view stays smooth even when detection is busy.

Step 1: Live view in the browser

Download MediaMTX, then add your camera to mediamtx.yml:

paths:
  frontdoor:
    source: rtsp://user:pass@192.168.1.20:554/stream1

Run ./mediamtx and open http://localhost:8889/frontdoor. You now have a live, low-latency stream. (New to this? Read how RTSP-to-WebRTC works first.)

Step 2: Add AI person detection

Install the libraries:

pip install ultralytics opencv-python requests

Create detect.py:

import time, cv2, requests
from ultralytics import YOLO

model = YOLO("yolov8n.pt")          # small, fast model
cap = cv2.VideoCapture("rtsp://localhost:8554/frontdoor")
last_alert = 0

while True:
    ok, frame = cap.read()
    if not ok:
        time.sleep(1); continue
    results = model(frame, classes=[0], conf=0.5, verbose=False)  # class 0 = person
    if len(results[0].boxes) and time.time() - last_alert > 60:
        last_alert = time.time()
        cv2.imwrite("alert.jpg", results[0].plot())
        print("Person detected!")
        # send yourself a message, e.g. a Telegram bot:
        # requests.post(f"https://api.telegram.org/bot<TOKEN>/sendPhoto",
        #               data={"chat_id": "<CHAT_ID>"}, files={"photo": open("alert.jpg","rb")})

Notice the script reads from MediaMTX (localhost:8554), not the camera directly. Many cheap cameras only allow one or two connections, so MediaMTX shares a single connection with everything.

Step 3: Make it practical

• Save CPU: process one frame every second instead of every frame.
• Fewer false alarms: only alert when the person is inside a zone (a doorway, a gate).
• Recording: MediaMTX can record streams to disk with a setting in the same config file.
• Remote access: use a VPN like Tailscale rather than opening ports on your router.

Need something like this built? I take freelance projects in streaming, web apps and automation. Get in touch.

More in Tutorials

• How to Stream an RTSP Camera to Your Browser with WebRTC
• 10 VS Code Extensions That Save Time Every Day
• Run AI Models on Your Own Computer with Ollama (Step by Step)