Object classification, position, and rotation angle


I’m working on pick and place solution for a box laying a table.
The environments will be:

  • Only one camera is used and it is in a fixed position
  • Camera looking at the table top view
  • Only one object at a time is placed on the table
  • a camera(openmv) will use a ToF sensor for distance calculation

The steps of the process that I imagine:

  1. object detection
  2. After detecting an object, calculate the object’s centroid position
  3. determine the orientation of an object
  4. send the object’s position info [cx, cy, z, degree] through modbus

I was thinking on adding AI model to openmv ecosystem to detect the boxes,
but I’m not sure if it determine the orientation and position of an object.

I read openmv documents and I found some function like
which,I guess, might help for this project, but I don’t know how to calculate the rotation
of an object like this (How to Determine the Orientation of an Object Using OpenCV – Automatic Addison) and how to get an object’s position.

Could you please give me some advice or better idea?

Thank you,

We don’t have generic object detection working yet. However, there’s a push to get this done by the end of the year. However, it will only run on the H7 plus… which is out of stock right now.

It’s ok about object detection. I have a STM32H7 discovery kit and already tested a model.
I wonder how to code to get the position and rotation angle of an object…

A model can tell you that with the right training.

Hello @kwagyeman !
I am really interested in this and would love to have this feature, the good news is I do have an H7 Plus!
I am wondering if there is any place where I may be able to see status updates on how this feature is progressing, considering the year is nearing the end.
Is there a github thread on this or something like that?

Thanks a lot! (and sorry for necromancing this thread)

Yeah, it’s just that I haven’t had time. I should be able to lock out the library on the OpenMV Cam soon. However, Edge Impulse has to release the models then afterwards.