YOLOv8 Based Deep Learning Method for Potholes Detection

Potholes are a common road hazard that may cause significant damage to vehicles and pose risks to drivers and pedestrians. The detection and classification of potholes in images can aid in effective maintenance and repair of road infrastructure. In this study, the training of YOLOv8, an advanced object detection algorithm, on a pothole image dataset is presented. The pothole image dataset consists of diverse images captured from various road conditions and environments. The dataset is annotated with bounding boxes around the pothole instances to facilitate supervised learning. YOLOv8, known for its accuracy and efficiency, is utilized as the object detection framework due to its real-time object detection ability. To train YOLOv8, a two-step procedure is followed. Firstly, a pre-training phase is conducted on a large-scale COCO dataset. This pre-trained model is fine-tuned on the pothole image dataset. The fine-tuning process adapts the pre-trained model to accurately detect and classify potholes. The performance of the trained YOLOv8 model is evaluated. This proposed approach is able to accurately detect and classify potholes in images. In this work YOLOv8 is used for pothole detection. Different variants of YOLOv8 (YOLOv8x, YOLOv81, YOLOv8m, YOLOv8s, YOLOv8n) are used and then the results on the dataset “Pothole detection Image Dataset” is analysed. Overall, this study presents a comprehensive framework for training YOLOv8 on a pothole image dataset.