Inspiration

Ever seen a pile of garbage and wanted to shout out loud for god's sake that it should be removed from there? If no, then you are really lucky and otherwise, welcome to the real world problems. As humanity develops new technology and equipment, the level of waste increases every day. But what if we can use the boom in technology to solve the problem, imagine a futuristic world where flying objects are picking up the garbage, not just from the dustbins but from any place where garbage resides. We are trying to build a very baseline prototype to that futuristic world.

What it does

Aerial Imaging to identify unattended garbage on the streets, beaches and inform local bodies or government agencies about the trash hotspots.

How we built it

We are using drones to take aerial images of the area. These images are then fed into our machine learning model which learns from garbage images and tells the percentage of garbage it detects in the image. We have set the threshold to 30% as of now, that if in the image area we see 30% or more garbage we show that in our UI. We are using resNet to transform the images into vector representations and then are using Convolutional Neural Network to detect the garbage in the image. The model is trained on 4k images collected from trashnet repo.

The UI is built from bootstrap which demonstrates the images which crosses the threshold of too much garbage in an area.

Challenges we ran into

  1. Configuring the drone and integrating the drone feeds with our model and UI was really challenging.
  2. Training the model on so many images was really time-consuming and the model accuracy can be improved further.
  3. Doing all this in 24 hours was a challenging task.

Accomplishments that we're proud of

Building it. This was not an easy project to build as we had to use many tools that we had not used before and experimented with them. Incorporating APIs while using bootstrap and parsing the images to be fed to the CNNs was a complicated task as we were not used to the format of the calls.

What we learned

  1. How to work around drones.
  2. Training machine learning models on images.
  3. Image classification
  4. How to be collaborative efficiently when the workload is high.

What's next for Trash Buzzer

We have arranged a picker for the drone, so our goal is to identify the areas with garbage and identify the nearest dustbin and making the drone to pick up the garbage and dispose into the dustbin. The initial idea is to start with really small garbage like cans, plastic cups etc.

Built With

Share this project:

Updates