I’m building a project to detect work errors in a warehouse. The goal is to check if an employee follows the correct sequence (e.g., Pick item → Press button → Place in bin).
This requires detecting both the person’s actions (using their pose) and the objects they interact with.
For this combined task, what is the recommended Project Type in Roboflow? Is Keypoint Detection the right choice because it supports both pose skeletons and simple object bounding boxes?
If your use-case is simple, having one KeyPoint Detection project and model which does both (pose + object detection) makes sense. If you also want to detect individual products (toothpaste, toilet paper, etc), it will make sense to have separate projects; one object detection project and one keypoint detection project.
You can then use both models in a workflow to solve your problem of checking if an employee follows the correct sequence.
Hope this helps you get started with your exciting project!
Thanks for the very helpful information! I have a follow-up question. Could you please let me know approximately how many images are needed to train the model for workflow error detection?
