Machine learning with image tracking of a line

[member 428835]

Hi PF!

At the bottom of a transparent cup full of water is a hole where water drains. The cup is sloshed, so the gas-water interface is not flat. Are there any techniques you're aware of that implement machine learning to track the interface?

My current technique not using ML is:
1) splice video into still frames
2) turn each frame into a binary image. There are image filters I use, which kinda of require some specific oversight from my end.
3) Increment the region of interest
4) run a for-loop from the left most pixel column to the right most, storing the first pixel with a "zero" value in each column

This technique works pretty well, but obviously fails if the interface is "double-valued" within any given row (so it wouldn't work for tracking, say, a bubble). Is there a way for machine learning to better track the interface? I should stipulate I'm only interested in a machine-learning approach.

Thanks for your time!

 

[Aaron Bex]

Hi there,

I'm not familiar with any techniques that use machine learning to track the interface in a transparent cup full of water with a hole at the bottom. However, it's possible that there are techniques out there that use ML to achieve this.

You mentioned you have a technique that involves splicing video into still frames and turning each frame into a binary image. Have you considered using a convolutional neural network (CNN) to classify the frames? A CNN might help to identify the interface more accurately than your current technique.

Do you know of any other techniques or libraries that you could use to implement machine learning for tracking the interface?

Thanks!