Trash Bin Prediction using Object Detection

Trash Bin Prediction using Object Detection

Author Name : Sonia Malik

ABSTRACT

Improper waste management has been a problem in India for a very long time. There is low awareness among Indian citizens regarding proper segregation of garbage waste. This has not only caused severe mismanagement of garbage, but has also resulted in fatal accidents, poor health conditions of workers at landfill sites and significant environmental pollution. Efficient and accurate object detection has been an important topic in advancement of computer vision systems. This work presents an online web-app that detects garbage objects in real-time with the help of a state-of-the-art object detection model and decide accurately the color of the dustbin in which the object should be thrown and also decide whether the object belongs to dry or wet waste. The selection of the model to be used is based on extensive research of the available object detection models that are suitable for this application of detecting garbage objects in real time with high precision and less inference time. Based on the research, the YOLO class of object detection models is identified as the best fit for this project. The several architectures of this class – YOLO V3, YOLO V4 and YOLO V5 are chosen for the task. A rich dataset consisting of expected garbage objects is prepared. Before using it for training, this dataset is made to pass through several preprocessing and augmentation stages to reduce training time, artificially increase the size of the dataset, increase model performance and eventually create different versions of the dataset. Each of these versions of dataset is used to train each of the models using transfer learning. Based on several performance metrics the best combination of the dataset version and the object detection model is selected, fine-tuned and implemented in an easy-to- use and attractive web application deployed on the internet which is freely available to everyone. This effort will make waste segregation extremely easy, improve waste management and prevent life threatening health diseases and accidents at garbage dumps and landfill sites.

Keywords: Garbage detection and classification, object detection, waste management.