The problem Smart-Waste solves

There are a multitude of people living in the NIT Silchar campus, be it students, guards or faculties. Naturally, this will generate a lot of waste. Waste is best treated when its properties are known. Presently, there is no centralised monitoring of waste disposals and therefore it is difficult to make stratergies for dynamic and effective management of waste. Our solution automates the process of tracking disposals and categorisation according to its properties by using IoT and machine learning. Smart-Waste is not just a piece of software, but a complete ecosystem in itself.
The ecosystem contains IoT aided dustbins containing weight sensors and camera that edge computes a ML model on the images fed by the device's camera module, mobile application for users, and monitoring dashboard for admins. Users have to scan a QR code on the dustbin after disposing waste to be able to track disposals. The IoT device sends data to backend servers once it detects disposal by change in weight and composition by the ML model. Admin can view data of all the dustbins and can add or remove dustbins. Admins will also recieve notifications when any dustbin exceeds its capacity to maintain cleanliness.

Challenges we ran into

Our project involves IoT integration but due to time and budget constraints it was not possible to implement the IoT part of the project. We figured out a workaround to this by creating a python script that replicates the behaviour of the IoT device by sending dummy data to the backend server after edge computing the ML model for classificaton of categories of waste in disposal.
We faced a lot of difficulties in development of the model. The task at hand to classify types of waste by image processing and multiclass object detection was not a straighforward and trivial problem. After trying multiple algorithms such as neural networks for image processing, and YOLO and SSD for multiple objects detection, we finally implemented SSD with COCO dataset.