AA+TrashSort

Inspiration

We all very well know how beneficial it can be for our planet if we recycle. We use papers, plastic bottles, aluminium cans, cardboard, glass on a daily basis, all of which can be recycled if we wanted to. But often times the thing that is stopping us from doing what is right is simply the additional steps we would have to take to get our items recycled. Would i throw my recyclable trash in a general waste bin that is 2m away or find a recycle bin further away? And even in cases whereby there are 4 recycle bins side by side one for each type of material, we would have to pause for a few seconds before we are able to decide which bin we would like to throw our items in. In the round the clock work life we live in these days, that few seconds of additional wait might be a compelling enough reason for someone to just throw his/her items in the general waste bin rather than recycle them responsibly

What it does

Our prototype is a smart bin, one that would hopefully replace all rubbish bins in the world, one that can sort the item thrown in automatically into plastics, general waste, paper and aluminium cans. So it basically shifts the hard work of sorting waste from humans to machines.

How we built it

We built the rubbish bin out of a cardboard box with a platform controlled by the servo motor to perform the trash sorting process. It runs on the Raspberry Pi with PiCamera module to take images of the trash and analyze it as well as utilizing a servo motor to perform the actual sorting. We trained our own neural net using Tensorflow on our self-created data set to accurately classify the trash that we placed on the platform.

Challenges we ran into

Challenges we faced were mainly in the form of hardware limitations. Since we could only work with the hardware components we ordered at the beginning, we didn't foresee potential hardware that we could have used. Such sensors included, weight sensors and proximity sensors. Additionally, due to time constraints, we had limited time to collect more data of the trash and consequently train our neural net on a more extensive dataset.

Accomplishments that we're proud of

We were able to successfully come up with a beta version of the prototype, one that classify the items thrown into 2 groups. Scaling it to more than 2 groups is not very hard as all it requires is a more complicated Rubbish bin design with more servo motors to control direction of sort as well as a larger dataset to classify a wider range of trash seen on a daily basis.

What we learnt

Hardware and software integration poses its own set of challenges that are very hard to debug. We spent countless hours messing with the voltages and the PWM outputs of the Raspberry Pi (without a multimeter) to get it to work with our servo motors. Additionally, we had to design our own power sources to split up the power between the raspbery pi and the servo motor as well as the picamera as all these components took a significant power draw. The limited computing power of the raspberry pi also forced us to reconsider our strategies to utilize more efficient algorithms to minimize

What's next for TrashSort

We would like to extend it to be able to sort the items into the 4 standard categories used by the government to classify trash instead of the 2 that our current prototype is able to work with.

Built With

• raspberry-pi
• tensorflow