Are bounding boxes required when using keypoints?

[Automateme2]

Hi all, probably a newbie question, but I'm going around in circles.
So, I run a large excavator and I'm trying to track the bucket size reasonably accurately, as it moves up & down, nearer and further away.
It's basically a moving rectangular box.

I successfully added bounding boxes to 800 images using Labelimg and trained a model to track my custom object. However the box sizes change and not very stable. I need more accurate sizes.

I found CVAT to do the annotating and am trying to setup 4 key points to define the buckets' shape.

• Should I be using individual images or mp4 ?
• Does it need a bounding box?
• How many images should I annotate as a trial?

Thanks

Brent