Extract Pose Key Points Using rpicam-vid/rpicam-hello

Hi!

I recently bought a Raspberry Pi 5 and a HAILO8 AI accelerator, alongside with a case that supports the PCI express 3 interface.

I was trying all kinds of variation in the past couple of days to extract the key points based on the yolov8s_pose_h8l_pi and yolov8s_pose_h8 models. Initially, after setting the Raspberry Pi 5 stock OS (latest stable), I installed the rpicam-apps, the hailo8 tools and all the required drivers.

The example commands with raspicam-vid and raspicam-hello both work, providing me an overlay in the UI with some purple lines connecting the key points of my body.

While the overlay is cool, I need a practical tool to extract that data and output it to the console, preferably. I tried modifying the post process json file and added several attributes in the “hailo_yolo_pose” section, like “output”: “stdout”, or “dump_results”: true, but none of those increases the console verbosity while running the command.

I also tried with the python script, called basic_pipelines/pose_estimation.py from here, however the script is very unstable. The original version doesn’t even run, I had to make some optimisations in the code to make it work, but after a few seconds it freezes.

From the way I see, there are 3 options:

  1. There is an attribute that makes the rpicam-apps output the key points
  2. There is a stable python script (I don’t need UI or overlays btw) that is reliable enough
  3. Writting a plugin or modifying the C++ source code for the post processor (or make my own libcam module, which would take a lot of time)

Can you please help me with this issue?

Welcome to the Hailo Community!

I recommend you have a look at our Application Examples repositories.

GitHub - Hailo Application Code Examples

And we have one especially for Raspberry Pi

GitHub - Hailo RPi5 Examples

When you use GStreamer there are other plugins that may help you create a pipeline and extract the data. Please have a look at the TAPPAS User Guide (available in the Developer Zone).

1 Like