Phone Detection

Real-time driver phone detection using a Roboflow model and OpenCV.

Prerequisites

Create a .env file in this directory:

ROBOFLOW_API_KEY=your_key_here

---

Option 1 — Docker Compose (headless, no window)

Best for servers, background processes, or running against an RTSP camera stream. The container always runs with --headless, so no display is needed.

RTSP stream

Add RTSP_URL to your .env:

ROBOFLOW_API_KEY=your_key_here
RTSP_URL=rtsp://user:pass@192.168.1.100/stream

Then build and run:

docker compose up --build

Webcam

In docker-compose.yml, swap the command and uncomment devices:

command: ["--source", "0"]
devices:
  - /dev/video0:/dev/video0

Then:

docker compose up --build

Note: webcam passthrough requires Linux. On macOS, Docker Desktop does not expose USB/webcam devices to containers — use the native approach below instead.

---

Option 2 — Native macOS (with live window)

Run the script directly in a virtual environment to get the cv2.imshow preview window.

Setup (first time)

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Webcam

source .venv/bin/activate
python src/detect.py --source 0

Press q in the video window to quit.

RTSP stream

source .venv/bin/activate
python src/detect.py --rtsp rtsp://user:pass@192.168.1.100/stream

Headless (no window) — native

python src/detect.py --source 0 --headless

---

Options

Flag	Default	Description
--source	0	Webcam index or path to a video file
--rtsp	—	RTSP URL; overrides --source
--conf	0.5	Confidence threshold (0–1)
--headless	off	Disable the preview window