Garbage in bins around cities are constantly overflowing. Our goal was to create a system that better allocates time and resources to help prevent this problem, while also positively impacting the environment.
What it does
Urbins provides a live monitoring web application that displays the live capacity of both garbage and recycling compartments using ultrasonic sensors. This functionality can be seen inside the prototype garbage bin. The bin uses a cell phone camera to send an image to the custom learning model built with IBM Watson. The results from the Watson model is used to classify each object placed in the bin so that it can be sorted into either garbage or recycling. Based on the classification, the Android application controls the V-shaped platform using a servo motor to tilt the platform and drop the item into it's correct bin. Once a garbage/recycling bin nears full-capacity, STDlib is used to notify city workers via SMS that bins at a given address are full.
Machine learning is applied when an object cannot be classified. When this happens, the image of the object is sent via STDlib to Slack. Along with the image, response buttons are displayed in Slack, which allows a city worker to manually classify the item. Once a selection is made, the new classification is used to further train the Watson model. This updated model is then used by all the connected smart garbage bins, allowing for all the bins to learn.
Challenges we ran into
Integrating all components
Learning to use IBM Watson
Providing the set of images for IBM Watson (Needed to be a zip file containing at least 10 photos to update the model)
Accomplishments that we're proud of
Integrating all the components.
Getting IBM Watson working
Getting STDlib working
Training IBM Watson using STDLib
What we learned
How to use IBM Watson
How to effectively plan a project
Designing an effective architecture
How to use STDlib
What's next for Urbins
Algorithm for optimal route for shift
Dashboard with map areas, floor plans, housing plans, and event maps
Heat map on google maps
Bar chart of stats over past 6 months (which bin was the most frequently filled?)
Product Information and Brand data