I am the leader of a team of 5 passionate students
We practiced Design thinking process and product prototyping process
Project Details
Problem
We discover that in Hong Kong when you throw 100 plastic bottles into recycle bin, only 4 of them is actually recycled
We interviewed different industry leaders to get more insights
We found 3 mean causes
It is expensive to process plastic - Multiple types of plastic requiring different processing plant to properly recycle. Too expensive to filter out other types of plastic and then send to another processing plant.
High transportation cost - The bottles are not compressed when they are being transported, leading to the truck transporting mostly air, instead of plastic. Plastic per truck is too low
Contamination - Small contaminents like water or coke will ruin the entire batch of plastic bottles
We asked “Why not install compressors in recycle bin? It will increase the ‘Plastic bottle per truck ratio’.”
Ans: If we compress the bottle, and it has liquid inside, it ruins the entire batch of plastic and also cuase hygene problems.
Solution
We went through the ideation process iteratively to arrive in the final solution
A smart recycle bin consist of 3 main component to solve the problems
Components
Computer vision system (CV)
Computer vision (CV) system that identifies contaminants inside bottles and would reject the bottle if they are detected
The CV system combines with weight sensor to estimate the expected weight of the bottle
If the actual weight exceeds the expected weight, we reject the bottle
This systems works for non-transparent bottles
NIR spectrometer
We use NIR spectrometer to identify the type of plastic
We sort the plastic at the source, so that there is not “Second stage transportation”
Mechanical Compressor
We compressor that shrinks the volume of bottles after it is identified to be clean
Due to safety concerns, we did not implement it in the lab prototype
Computer Vision System
I was in charge of developing the computer vision system
I have built my own dataset of 5000+ images
Built and trained my own model using Tensorflow and Keras
The following is a snippet of my technical research report