Performance Analysis of Banner and Poster Detection using Machine Learning Algorithms

Abstract

This research project, conducted within the Computer Science and Engineering Department at Daffodil International University, focuses on the implementation of YOLOv8, a real-time object detection system, for the purpose of Banner and Poster Detection in the unique visual context of Bangladesh. Leveraging diverse datasets collected physically from local areas and annotated using the Roboflow website, the study explores the key elements contributing to the high accuracy of YOLOv8 in detecting banners. The model's architecture, including advancements in YOLOv8's latest version, bounding box regression, and confidence scoring, facilitates precise localization with confidence scores reaching 99.99%. The use of normalized coordinates and probability distribution further enhances the model's ability to generalize across different image sizes. Multiobject detection capabilities, training on diverse datasets, and post-processing strategies implemented by Ultralights' engine contribute to the model's robust performance. The research project attains a remarkable accuracy of 99.99%, validating the efficacy of YOLOv8 in automated banner detection tasks. The outcomes not only showcase the model's strengths but also hold significant implications for real-world applications, offering a reliable and accurate system for detecting banners and posters in the dynamic visual landscape of Bangladesh.