Camera board • RPI Hardware Requirements for Stream Processing with pose_estimation_tf

Hi

Summary

Please could you advise on a Raspberry Pi 4/5, AI HAT, AI camera setup for processing a video stream for posture recognition?

Details

I would like to (for example):
- process a video stream of a tennis player serving and returning balls.
- every n frames do posture estimation to find limb positions.
- train an AI model on "good practice".
- run the trained LLM on the rpi to identify a players "not good practice" and provide constructive feedback.

I read with interest here:

https://www.raspberrypi.com/documentati ... picam-apps

about the pose estimation capability (see pose_estimation_tf stage in the page).

Any guidance on resources about how to build this system and train my own models would be greatly appreciated.

These are a hardware platforms I'm currently thinking about:
Option A:
- RPI 5 8 or 16GB Ram
- AI M.2 or M.3 HAT+
- camera

Option B:
- RPI 5 8 or 16GB Ram
- AI camera

Note the Raspberry Pi AI Kit is no longer available.

Thanks for your thoughts.

J

Statistics: Posted by Jellybean — Sat Jul 11, 2026 4:51 pm

---