Pose Estimation and Face Landmark Tracking with Raspberry Pi and OpenCV

Hi Edward,

Welcome to the forum!!

If you run the following line in the terminal does it fix your error?
sudo apt-get install python-opencv python3-opencv opencv-data

Liam

1 Like

Hi Tim
This is amazing!!
I’m planning to use it so that my wife and I (and only us) can open and close our garage my raising or lowering our arms. I’m hoping to combine this with vehicle model and number plate recognition so that it also opens when we drive up.
Only snag I had getting it up and running was that the software load script installed numpy 1.16.2 and warned that this was not compatible with matplotlib 3.5.3 which requires numpy >= 1.17. This was confirmed by an error when I first ran your python code but it ran perfectly after setting python3 as my default python guided by Python Default Version - Raspberry Valley
and upgrading numpy using: sudo pip3 install numpy upgrade. The upgrade to numpy version 1.21.6 was conformed by pip show numpy.
Thanks for yet another great guide.

1 Like

Hey mate :blush:

I love your idea! Please take photos along the way as you build it because that project definitely belongs on our projects page - Electronics Projects Australia

Also give the Oak-D Lite a squizz. When you start stacking AI systems on top of each other that AI camera definitely provides some Computing Ooomfph. And I have seen it used to track vehicles and number plates too :blush: Could be right up your alley - depthai-experiments/gen2-license-plate-recognition at master · luxonis/depthai-experiments · GitHub.

Do stay safe though and provide yourself and the system some safeguards. I definitely don’t let any nefarious agents sneaking into your place.

Kindest regards,
Tim

Thanks for your support and advice Tim.
The RPi 4 is certainly struggling a bit at very low fps with quite some lag, so the Oak-D lite looks well worth exploring. I’ve watched your Oak-D lite video and read the guide and lots of other stuff but I’m still not entirely sure how my plan would all play out with the Oak-D lite camera?? Guess I’ll just have to get one and try it out. I’ve got the hand raising (posture) recognition working, as well as face identification (which I’ve migrated to use opencv exclusively, without the imutils functions involvement, without any real change in fps) so I can run both face identification and posture position recognotion in the same python script, but maybe I’ll have to ditch all this code to use the Oak-D lite and adapt their computer vision AI tools and samples?? Now looking at ground up opencv number plate recognition, since ALPR apparently now charges $5/month.

1 Like