Yolo option for measuring in real time the size of stones being transported on a conveyor belt

Community Help
1

Good afternoon, I am contacting you to ask if there is any Yolo option to measure in real time the size of stones that are transported on a conveyor belt that feeds a vertical mill that reduces their size.
I am a teacher at the FIO UNICEN in Olavarría, Province of Buenos Aires, Argentina, and this work is part of a thesis for the industrial engineering degree. The idea is to detect when that size exceeds a certain limit to stop the line before that stone breaks the machine and puts it out of service. We are trying to program something but I still can’t find what I’m looking for…
If it is possible to use some version of yolo for this, it would be very helpful to have more information.
Thank you very much,
Geraldina.

2

Disclaimer: I am not an expert and this is my own interpretation of solving the problem.

Yes you can definitely use the YOLO object detection model to solve the problem.
How can yolo object detection model be used to solve the issue?
Well you will have to train the yolo model to detect the rocks and then you can do the post processing.

The whole process will look something like this

  1. Decide the yolo model version and size to use eg yolov8 , yolo5 and each yolo model comes in different sizes eg nano, small, medium, large and extra large. Model size will have impact on the accuracy of detections and how much processing resources are needed to do object detection. I find yolov8 medium to be a best of both worlds.
  2. once you have the model selected train it on custom dataset. Either you will have to prepare you own dataset or download from online site(might be hard to find dataset which meets your setup requirements)
  3. You can train on kaggle which gives free compute units.
  4. After training you will have to setup camera and use live stream to run run inference on the video. You can use inference pipeline from roboflow to do this.
  5. Setup the camera in close proximity and for each detected object(rock) get the bounding box of detection. You can map the width and height of bounding box to the actual size of the rock (will give you rough estimate). If the size is bigger than certain number you can trigger the stop condition for conveyer belt.
  6. While running inference on the inference to track the objects you can use bytrack

I missed so many points but this should give you a good starting point and overview of the whole thing
Checkout ultralytics docs here: https://docs.ultralytics.com/ you will find the information helpful

3

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.