Gear.Ai

Inspiration

The inspiration for this project stemmed from the topic of pollution specifically, incorrect disposal of material. From our experience, many people find it hard to sort waste and therefore end up placing materials in the wrong compartment. This makes reusability, conservation difficult and usually leads to pollution due to unusable waste. We believe that in order top solve the world's waste management problem, we need to reach the general consumers and start at an individual level. By targeting the root cause for improper waste management, we can tackle this environmental problem which affects the beautiful earth we live on.

What it does

Gear.Ai is able to detect objects and categorize into 4 categories: Hazardous Waste, Electrical Waste (E-waste), Recyclable and Green Waste. The app is able to approximate show the percent of certain "type of waste" in a particular object. For example, if a cell phone is shown, it should detect certain percentage of Electrical-waste, certain elements as recyclable etc.

How I built it

The first step in machine learning is gathering data. Using google Images and locally captured pictures, we created a dataset. This dataset was stored on Azure and then "pulled" into the application using Azure API. Computer vision functionality was added using python-based computer vision framework, OpenCV. We also used a local web server called Flask

Challenges I ran into

Since my team was quite inexperienced, we encountered many problems related to training our dataset and getting accurate predictions. The first challenge we come across is collecting enough images for test objects. We had to manually create a new dataset to suit our apps needs and therefore was a rather time-consuming.

Accomplishments that I'm proud of

The app is able to distinguish between recyclable materials such as plastic as well as electrical waste such as phones. In addition it provides and approximate estimate of the types of materials present going beyond the basic categorization of of items based on type. We have a trained model to around 70-80% accuracy

What I learned

Throughout the process of developing our hack, we gained many insights about team collaboration and types of challenges we may face, From the basics such as collect sufficient data, we were able to work on techniques to speed up data collection and utilize Azure's pre-trained models to develop this application. In addition to this, we learned about using OpenCv2 and how we can optimize computer vision to solve real world problems.

What's next for Gear.Ai

This concept has a huge potential in the future. Gear.Ai can be developed to analyze materials and provide waste management analytics and derive consumer waste-type data from across the world. Using this, we can analyze materials and bring waste management at the fingertips of the general user.

Next Steps:

Creating a functional iOS app and waste token.
reward system for good waste management practices.
Special features for large-scale parties and events which can make them eco-friendly