SMART OBJECTS CAMERAS

Working Package Versions

OAK-D Camera Setup — Raspberry Pi 5

Last updated: February 2026

PLATFORM

System Environment

Hardware: Raspberry Pi 5 — 16 GB RAM

Operating System: Debian GNU/Linux 13 (trixie) — aarch64

Python: 3.13.5

Virtual Environment: /opt/oak-shared/venv/ (shared, read-only for students)

Model Cache: /opt/depthai-cache/ (777 permissions, all users can write)

Full reference →

DEPENDENCIES

Critical Package Versions

These versions are confirmed working together for OAK-D camera detection with GUI display on Raspberry Pi 5.

Package	Version	Role
`depthai`	3.3.0	OAK-D camera SDK (v3 API)
`depthai-nodes`	0.3.7	Neural network parsing nodes
`opencv-contrib-python`	4.10.0.84	Computer vision + GUI support
`numpy`	1.26.4	Array math (must be < 2.0)
`scipy`	1.17.0	Scientific computing
`matplotlib`	3.10.8	Plotting & visualization

Full reference →

COMPATIBILITY

What Works vs. What Doesn't

Works

opencv-contrib-python 4.10 — GUI support, all attributes (COLORMAP_JET, imshow)
depthai 3.3.0 + depthai-nodes 0.3.7 — v3 API with ParsingNeuralNetwork
numpy 1.26.4 — compatible (< 2.0 required by depthai)

Doesn't Work

opencv-python-headless — Missing GUI functions (no cv2.namedWindow, cv2.imshow)

opencv 4.5.5.62 — Has GUI but missing cv2.COLORMAP_JET; causes AttributeError

depthai 2.30.0 — Incompatible with depthai-nodes 0.3.7; missing dai.Color

No system GTK libs — OpenCV requires libgtk-3-dev and libgtk2.0-dev via apt

Full reference →

INSTALLATION

Setup from Scratch

1. System Dependencies

sudo apt-get update

sudo apt-get install -y libgtk-3-dev libgtk2.0-dev pkg-config

2. Create Virtual Environment

sudo mkdir -p /opt/oak-shared

sudo python3 -m venv /opt/oak-shared/venv

sudo /opt/oak-shared/venv/bin/pip install --upgrade pip

3. Install Packages (order matters!)

pip install numpy==1.26.4

pip install opencv-contrib-python==4.10.0.84

pip install depthai==3.3.0

pip install depthai-nodes==0.3.7

pip install scipy==1.17.0

pip install matplotlib==3.10.8

Tip: Install numpy first — depthai requires numpy < 2.0.

Full reference →

TROUBLESHOOTING

Known Issues & Solutions

COLORMAP_JET error
module 'cv2' has no attribute 'COLORMAP_JET'
Fix: Upgrade to opencv-contrib-python==4.10.0.84 (not 4.5.5)

GTK rebuild error
Rebuild the library with GTK+ 2.x support
Fix: sudo apt-get install -y libgtk-3-dev libgtk2.0-dev pkg-config

depthai Color error
module 'depthai' has no attribute 'Color'
Fix: Upgrade to depthai==3.3.0 (not 2.30.0)

Qt platform warning
Could not find the Qt platform plugin "wayland"
Impact: Harmless — GUI still works. OpenCV falls back to X11.

Full reference →

VERIFICATION

Testing the Setup

1. Verify OpenCV GUI Support

source /opt/oak-shared/venv/bin/activate
python3 -c "import cv2; cv2.namedWindow('test'); cv2.destroyWindow('test'); print('GUI works')"

Expected: GUI works (no errors)

2. Verify DepthAI Camera Connection

source /opt/oak-shared/venv/bin/activate
python3 -c "import depthai as dai; print(dai.Device.getAllAvailableDevices())"

Expected: Device info with mxid (camera serial number)

3. Test Detection with Display

source /opt/oak-shared/venv/bin/activate
cd ~/oak-projects
python3 person_detector_with_display.py --display

Expected: Live camera window with green bounding boxes around detected people. Press q to quit.

Full reference →

HARDWARE

Hardware Configuration

Raspberry Pi 5

Model	Raspberry Pi 5 (16 GB RAM)
OS	Raspberry Pi OS (Bookworm) 64-bit
USB	Use USB 3.0 (blue) ports for OAK-D
Power	27 W official adapter or powered USB hub

OAK-D Camera

Connection	USB 3.0 (USB-C on camera side)
Platform	RVC2 (Luxonis Myriad X VPU)
Power	High draw during AI inference — use quality cable

Display Options

VNC — wayvnc (shows desktop of logged-in user)
Physical monitor — lightdm + lightdm-gtk-greeter
Headless — fully supported, no display needed

Full reference →