search

ethantrokie / water-detection

14 stars 0 forks source link

Detecting water live #2

Open Thomas-Markhorst opened 3 years ago

Thomas-Markhorst commented 3 years ago

Dear developer,

I'm part of a Delft University of Technology student group working on making a robotic boat that measures water quality and removes algea, plastic and other debris from water. For this we also use an Intel Realsense depth camera for obstacle avoidance, to distinguish water from actual obstacles it would be great to use your algorithm based on the UVA-paper. I have two questions:

  1. Would the algorithm be able to perform live? So when it gets a new frame immediatelly analyse this frame for presence of water?
  2. Is this algorithm capable of detecting the parts of the image in which water is present?

With best regards. Thomas Markhorst

ethantrokie commented 3 years ago

Hi, it's been a while since I worked on this, but I'm glad it might be put to some use for something good! In regards to your questions.

  1. Its not able to do it live as the code currently is (might be able to do it some tweaking) currently you have to take a short video (1-2 seconds is probably fine) then run it through the algorithm.
    1. It is capable of detecting the parts of the image where the water is present. In the .readme file in the "Machine Learning Algorithm" section and it should be able to explain which functions to show which parts of the image have water.

But also as a side note, my algorithm is useful if you are limited to a single RGB camera without depth perception, there's a lot of really good papers out there that do research on two camera systems/ones with depth perception. I highly recommend checking those out too. Theres a lot of great ways to detect water, so if you have flexibility on camera type I'd look into all your options because other types of cameras might be better suited for detecting water (such as a thermal cameras).

Best of luck and hope this helps.

Thomas-Markhorst commented 3 years ago

Hi Ethan,

Thanks for the quick reply! We will look into papers using depth perception for water detection. If we would need to use your algorithm, do you have any indications where we could tweak?

Best regards, Thomas

Thomas-Markhorst commented 3 years ago

Hi Ethan, I found another approach that will most likely solve our problem: https://www.researchgate.net/publication/280882993_Ground_Plane_Detection_Using_an_RGB-D_Sensor. This paper describes how to detect the ground level using the depth image. Thanks for your time anyway! With regards, Thomas