Build a Fire Detection with AI: YOLO, FastAPI & Next.js

Install conda from the official anaconda website and set up a python 3.9 virtual environment. Activate, deactivate, list, and remove environments, guided by the official documentation on basic usage.

Install and train YOLO using conda, configure data.yaml from Roboflow, and run predictions to detect fire images, saving results in runs/detect while testing with fire and non-fire samples.

Set up a FastAPI backend by creating an app with main.py in a conda environment, install FastAPI, configure Black formatter in VS Code, and run the dev server.

Connect to a database with sqlalchemy in fastapi, define models for user, session, and detection_log, and initialize sqlite with a sql_app.db, using a dedicated database client and explorer.

Adopt an endpoint-centric FastAPI structure with feature folders for each route, using router.py, schema.py, and CRUD modules to mirror Next.js API routes and enable automatic router registration.

Build an API that detects fire in images with YOLO using FastAPI, handling file uploads, temporary storage, and predictions with bounding boxes, fire_detected, and a timestamp.

Mount the log folder as StaticFiles in FastAPI to serve static image files at /log/image-filename, using os.path.join and os.path.dirname for paths. Prioritize security with appropriate access controls.

Log fire detection data to the database via crud module and create_detection_log, using SQLAlchemy add, commit, and refresh, update the sqlite file, and verify via the predict_fire endpoint with postman.

Implement a paginated fire detection log retrieval API using FastAPI components—crud, router, and schema—for DetectionLog, with sorting by created_at and JSON conversion, tested with Postman; Next.js frontend planned.

Install Next.js and pnpm, create a new Next.js app in the frontend folder, install pnpm globally, and run the development server.

Develop a Next.js frontend with TanStack Query that polls every 5 seconds, lets users select a camera, start and stop streaming, and send frames to /predict_fire for fire detection.

Implement real-time fire detection alerts by triggering a flashing screen overlay and an alert sound via the Web Audio API, with permission handling and oscillator gain control.

Present a fire detection logs page that fetches data from /get_detection_log with React Query, paginates results, and displays file name, image, and detections; includes home button and 0.7+ confidence filtering.