Yeah, I’m still new to all of it. My plan of action is to make a machine that tumbles bricks onto a conveyor and takes pictures of them before cycling them back through. This is to collect a dataset for training a machine vision model. If that doesn’t work out I’ll just have to make it sort the bricks by color as that’s easy enough to do.
The hardest part I’ve encountered so far is trying to learn how to control lots of stepper motors, how to power everything, and of course learning the 3d modelling process so I can mix both Lego and non Lego in the machine’s construction.
I haven’t done any modelling myself, but if I ever start, I’ll probably try openscad. Lego are fairly uniform, so I would wonder if more traditional machine vision techniques would work.
For machine vision, I was thinking of using multiple cameras and then just stitching the images together and using that as the input. That way it gets images of the bricks from multiple angles, which when combined with bright lighting from LED strips should make it easier to identify.
I have basically no experience with any machine vision, I’ve done some edge/color detection once, but my intuition says that sounds like a plan. Because you have a pretty unique feature for each brick, its number of studs and size of cavity on the bottom, you might not need as full a picture of each brick as you might think. That’s pure speculation though. Have fun with the project.
Yeah, I’m still new to all of it. My plan of action is to make a machine that tumbles bricks onto a conveyor and takes pictures of them before cycling them back through. This is to collect a dataset for training a machine vision model. If that doesn’t work out I’ll just have to make it sort the bricks by color as that’s easy enough to do.
The hardest part I’ve encountered so far is trying to learn how to control lots of stepper motors, how to power everything, and of course learning the 3d modelling process so I can mix both Lego and non Lego in the machine’s construction.
I haven’t done any modelling myself, but if I ever start, I’ll probably try openscad. Lego are fairly uniform, so I would wonder if more traditional machine vision techniques would work.
For machine vision, I was thinking of using multiple cameras and then just stitching the images together and using that as the input. That way it gets images of the bricks from multiple angles, which when combined with bright lighting from LED strips should make it easier to identify.
I have basically no experience with any machine vision, I’ve done some edge/color detection once, but my intuition says that sounds like a plan. Because you have a pretty unique feature for each brick, its number of studs and size of cavity on the bottom, you might not need as full a picture of each brick as you might think. That’s pure speculation though. Have fun with the project.