Picamera2 example

This example performs object detection on live video from a connected Raspberry Pi Camera. Its reported pipeline FPS includes camera capture and detection. Both headless and preview modes request a 30 FPS camera stream by default.

The camera examples run until interrupted and do not record video. Pass CAMERA_ARGS="--frames 100" to stop after a fixed number of detections.

Usage

Run these commands from the yolo26-raspberry-pi-5/ runtime directory.

Prerequisites

Install the required packages:

sudo apt update
sudo apt install --yes --no-install-recommends \
  ca-certificates \
  curl \
  git \
  make \
  python3 \
  python3-venv \
  python3-picamera2 \
  rpicam-apps \
  tar \
  gzip

Confirm camera access:

rpicam-hello --list-cameras

For a graphical preview, also install the Qt and OpenGL Python bindings:

sudo apt install --yes --no-install-recommends \
  python3-opengl \
  python3-pyqt5

To use your own compatible model, see Acquiring a model. Otherwise, the pinned standard YOLO26n checkpoint is downloaded and converted automatically.

Live camera without a preview

Run detection without rendering a preview:

make example-camera

This mode avoids display and overlay-rendering overhead. Select another camera with CAMERA=1, and pass additional application options through CAMERA_ARGS:

make example-camera CAMERA=1 CAMERA_ARGS="--frames 100 --print-detections"

Live camera preview

Run from a terminal in an active desktop session:

make example-camera CAMERA_PREVIEW=qtgl

Rendering the annotated preview adds performance overhead. Preview dimensions do not change the model's detector resolution.

When connecting over SSH to an existing desktop session, the desktop user and SSH user must be the same:

DISPLAY=:0 WAYLAND_DISPLAY=wayland-0 XDG_RUNTIME_DIR="/run/user/$(id -u)" \
  make example-camera CAMERA_PREVIEW=qtgl

Use CAMERA_PREVIEW=auto to let Picamera2 select a preview backend. Stop the graphical desktop and use CAMERA_PREVIEW=drm for direct console display.