Performance Evaluation of different Machine Learning Techniques for Pothole Detection

A pothole is an oval-shaped depression in the surface of a road that is the result of the cyclical forces applied to broken pavement by traffic, causing shattered pieces of pavement to be pushed to the side. Pothole detection is an essential part of maintaining roads and ensuring safe driving conditions, along with estimating the road repair costs accurately. It is a challenging task that requires accurate detection and monitoring of road conditions in real-time. With advancements in computer vision, the video feeds of cameras can be analyzed using various deep learning models. This work proposes a road damage detection system with in-depth evaluation towards five deep learning architectures, namely Ultralytics YOLOv8, ResNet-50, MobileNet, EfficientNet and DenseNet121. It is critical to the cost evaluation of road repairs to examine whether these models can detect potholes and segment the shape of potholes. The results obtained aim to help develop a computerized system, which is accurate and high-throughput.