pdf
Published
2023-12-06
Section
Engineering
How to Cite
An Efficient Computer Vision and Machine Learning Model for Real-Time Litter Detection on Raspberry Pi
Bernard Okyere
Kumasi Technical University
Linus Antonio Agyekum
Kumasi Technical University
Abstract
Improper waste disposal and littering pose significant environmental challenges in urban areas worldwide. Existing regulations and technologies often fall short of effectively addressing this issue. To provide a cost-effective solution, this paper demonstrates the feasibility and practicality of using computer vision and machine learning for litter detection, contributing to environmental preservation. This involves designing and deploying a Convolutional Neural Network (CNN) and Computer Vision model on a Raspberry Pi for real-time litter detection. It employs the SSD-Mobilenet-v2-FPNLite-320x320 Model for improved accuracy in detecting different types of litter. An overall mean average precision (mAP) of 76.5%, indicates the system’s effectiveness in detecting and classifying litter objects. Furthermore, the deployed model achieved significant feet by identifying and classifying various types of litter objects, including plastic bottles, paper, and polythene, achieving high precision and recall scores for these classes, and facilitating prompt detection and response in practical settings. This research provides a valuable solution for tackling littering and waste management challenges in public spaces.
