Enhanced YOLOv5s Model for Improved Multi-Sized Object Detection in Road Scenes

Object detection plays a crucial role in visual scene understanding, particularly in identifying and classifying multiple objects with varying sizes. This project focuses on enhancing the performance of object detection models using improved versions of YOLO (You Only Look Once), specifically YOLOv8 and YOLOv9, to accurately detect small, medium, and large-sized objects in structured road environments. The model is trained and evaluated using the publicly available “Road Traffic” dataset from Roboflow, which contains diverse traffic scenarios. The integration of enhanced model capabilities allows for better localization and recognition of objects that vary in scale, shape, and position. The system is built using Flask as the back-end framework and HTML, CSS, and JavaScript for the front-end interface. It includes user-oriented modules such as registration, login, home, user dashboard (supporting image and video input), and logout functionality. The overall aim is to develop an efficient detection system that addresses the limitations of size-based detection errors. The proposed system can be accessed through a web interface, allowing users to upload and view detection outputs interactively.

Keywords: YOLOv8, YOLOv9, Object Detection, Flask, Multi-sized Objects, Road Scenes, Image Detection, Video Analysis, Deep Learning, Traffic Dataset