TY  - BOOK
AU  - Anyaegbuna,Ikenna Emmanuel
AU  - Özgünalp,Umar
TI  - POTHOLE DETECTION AND REPORTING SYSTEM USING IMAGE PROCESSING
PY  - 2024///
KW  - Electronics and Communication Engineering
KW  - Dissertations, Academic
N1  - Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Electronics and Communication Engineering
N2  - Road damage detection is a pivotal aspect of infrastructure upkeep ensuring the safety and efficiency of transportation networks. Current methods often fall short in precisely categorizing distinct types of road surface irregularities and restricting maintenance strategies. This research aims to revolutionize road damage detection by leveraging cutting-edge computer vision techniques, specifically employing YOLOv8 models—YOLOv8n, YOLOv8s and YOLOv8m.
Methodologies encompass the rigorous training and optimization of these models using a comprehensive dataset sourced from the "2020 IEEE International Conference on Big Data." The primary objective is to enable these systems to detect and categorize various types of road surface damages accurately. Performance evaluations and comparative analysis among the YOLOv8 variants form the core of the study to determine their applicability in practical scenarios.
YOLOv8n demonstrated moderate performance with reasonable precision and recall rates albeit displaying relatively lower mAP50 and mAP50-95 scores across diverse damage classes. Conversely, YOLOv8s showcased amplified precision, recall and mAP scores signifying superior object detection capabilities compared to prior models. Nonetheless, YOLOv8m exhibited noteworthy advancements in precision and recall albeit at the cost of increased computational demands.
In conclusion, while YOLOv8 models prove effective in detecting and categorizing road surface damages their suitability varies based on computational requirements. YOLOv8n is fitting for devices with limited computational resources, YOLOv8s improves accuracy for moderately capable devices and YOLOv8m, although accurate demands higher computational power. This study offers crucial insights for selecting appropriate YOLOv8 models tailored to diverse computational scenarios thereby contributing to advancements in road infrastructure maintenance and safety
ER  -