Feature Description:
I would like to classify electricity poles according to their diameter. This diameter varies in increments of 5 mm.
Would the FOMO-pro be able to do this job, for example? Project ID:
[Provide the project ID, if applicable]
Context/Use case:
classify electricity poles
Problem to be Solved:
classify electricity poles
Proposed Solution:
Fomo-pro?
Expected Benefits:
[Explain the benefits of this feature, including how it will improve the user experience or workflow]
Potential Impact:
[Describe how this feature will impact your project or other users]
Related Issues or Requests:
[List any related issues or feature requests]
Additional Information:
[Any other information that might be relevant]
5mm Pole is a very small change in diameter, do you mean wire? and we dont have a model called FOMO-pro. Do you mean Yolo-Pro? If so can you share some image examples?
yes, only 5 mm between 5 types of poles.
Sorry about Fomo-pro, yes, I am talking about Yolo-Pro.
The diameter differences between the poles range from about 5 to 10 mm. For example, type 1 has a diameter of 232 mm, while type 2 measures 227 mm…
The idea is that when a car passes by on the street, it takes pictures of the pole and an edge device on the car will process and classify, identifying the type of pole in the image based on its diameter
Regardless of methodology, Edge AI, or traditional machine vision using e.g. contour finding and measuring, I find this very challenging. That is, unless you are at the exact same distance from the poles every time when you drive by. Or, there are some other features than only the diameter distinguishing the poles. With 5 mm at a distance of several meters seems the pole diameter difference would be a few pixels or so wide, of course depending on camera resolution.
But I’d love if you can prove me wrong!
I have some experience with using OpenMV cameras, lately the OpenMV Cam RT1062. I’ve used both Edge Impulse and traditional computer vision with it. You could also ask the same question in their forum.
Edit1: A year ago I made an nut counting project with that camera and Edge Impulse. Even with a static distance to the conveyor belt and controlled lighting, it had some issues in classifying M10 and M12 nuts correctly as they look too similar even to the naked eye. In that project, adding much more training data would probably have helped, but as said, that was in controlled conditions.
Edit2: If you use a camera/lens which zooms in “a lot!” so the pole is big in pixels AND you can measure the distance between the camera and the pole with a range sensor, it might be more feasible.
Could always hook up a stereo camera like the new OAK 4 D, I suspect you can connect and ingest directly to Edge Impulse, but fair warning, we haven’t tried!
