A Proposal for Pothole Detection with Machine Learning at the Edge
Urban infrastructure, edge machine learning, YOLOv8, FOMO, TinyML, pothole detection, model optimization, real-time processing.
Potholes in urban roads represent a significant issue, impacting both user safety and vehicle durability. This study introduces an innovative approach integrating machine learning on edge devices, with an emphasis on YOLOv8 and FOMO models in the context of TinyML. We utilize a specialized dataset, containing annotated images, to effectively train these models for accurate pothole detection. The focus of the research is on optimizing the performance of these models for devices with limited computational resources, aiming for real-time efficiency and reduced energy consumption. This work not only provides effectively trained models but also introduces an adaptable framework for pothole detection, ensuring practical and efficient implementation. Furthermore, we present a complete pipeline for pothole detection at the end, validating the models' accuracy and efficiency. This approach proposes a robust and energy-efficient solution for the automatic recognition of potholes, significantly contributing to improvements in urban infrastructure maintenance.