A streaming camera and clip recorder for FRC robotics data collection (and streaming).

Built by Nico and Ethan from FRC 4419 Team Rewind with the help of OpenAI Codex.

• Live MJPEG preview in the browser
• One-button start/stop recording from the web UI
• Hardware H.264 encoding (no FPS drop while recording)
• Timestamped MP4 clips with auto-cleanup (keeps last 5)
• One-click clip downloads for Roboflow labeling
• Team logo upload

The app runs a single Python process on the Raspberry Pi. A background thread captures frames from the camera in a continuous loop. Two things happen with those frames:

CSI Camera
    |
    v
Picamera2 capture loop (1280x720 @ 50 FPS)
    |
    |--- downsample to 640x360 --> JPEG encode (25 FPS) --> /stream.mjpg --> browser
    |
    '--- H.264 hardware encoder (1280x720 @ 50 FPS) --> .h264 --> ffmpeg remux --> .mp4

Preview stream -- Every ~40ms the latest frame is downscaled and JPEG-encoded with an FPS counter and REC indicator overlay. The Flask /stream.mjpg endpoint sends these JPEGs as a multipart HTTP stream, which the browser renders as a live video feed via a plain <img> tag.

Recording -- When you press Record, the Pi's hardware H.264 encoder starts writing a raw .h264 file directly from the camera's full-resolution stream. This happens in hardware so it doesn't slow down the preview or eat CPU. When you stop recording, ffmpeg remuxes the .h264 into a downloadable .mp4.

Web UI -- Flask serves a single HTML page. JavaScript polls /api/status every second and /api/clips every 10 seconds. Recording start/stop are simple POST requests to /api/start and /api/stop.

The camera must be connected via the CSI ribbon cable (not USB). Make sure the camera interface is enabled -- run sudo raspi-config, go to Interface Options, and enable the camera if it isn't already.

1. Install dependencies:

sudo apt update
sudo apt install -y python3-opencv python3-flask python3-picamera2 ffmpeg

2. Clone this repo:

cd ~
git clone https://github.com/zredlined/frc-ai-camera.git
cd frc-ai-camera

3. Test that it runs:

python3 app.py

Open http://<pi-ip>:5000 from your laptop. You should see the live feed. Press Ctrl+C to stop.

4. Set up as a system service (starts automatically on boot):

sudo tee /etc/systemd/system/frc-pi-camera.service > /dev/null << 'EOF'
[Unit]
Description=FRC Pi Camera Recorder
After=network.target

[Service]
Type=simple
User=admin
WorkingDirectory=/home/admin/frc-ai-camera
ExecStart=/usr/bin/python3 /home/admin/frc-ai-camera/app.py --output-dir /home/admin/frc-ai-camera/recordings
Restart=always
RestartSec=2

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl daemon-reload
sudo systemctl enable frc-pi-camera.service
sudo systemctl start frc-pi-camera.service

Change User and WorkingDirectory if your Pi username isn't admin.

Check that it's running:

sudo systemctl status frc-pi-camera.service

From your laptop, run the deploy script:

./deploy.sh <pi-hostname-or-ip>

This copies the code to the Pi and restarts the service. Default host is pi3.local if you don't pass one.

• Set a label (e.g. yellow_ball) in the web UI before recording.
• Record short clips (10-45 seconds) so they're easy to label in Roboflow.
• Vary distance, lighting, and backgrounds between clips.
• Download clips directly from the Recent Clips list in the web UI.

Clips are stored in recordings/ on the Pi. Only the last 5 are kept -- download any you want to keep before they get auto-deleted.

From your laptop:

scp -r admin@<pi-ip>:~/frc-ai-camera/recordings ./recordings_from_pi

python3 app.py --help

Camera defaults: 1280x720 @ 50 FPS capture, 640x360 @ 25 FPS preview, JPEG quality 65.

Click Upload Team Logo in the web UI. Supports .png, .jpg, .jpeg, .webp.