A Deep Learning Approach for Pothole Detection Using Yolov8 Model

Pothole The removal of pavement pieces on the road due to high traffic is known as Pothole. One of the major reasons for tedious travel and the major cause of accidents is due to the presence of potholes on roads. Hence pothole detection has been an interesting area of research for monitoring and maintaining road conditions in developed cities. Several systems exist for automatic detection of potholes but are either time consuming or less accurate. Hence an automatic pothole detection system has been developed that leverages Deep learning and Internet of things. Pothole images are captured using the CCTVs installed, and the images are transferred to the cloudfor further processing. YOLOv8 model has been used in detecting Potholes in images which is processed in Open VINO for deploying the optimized system. The core part of YOLOv8 is the CSPDarknet53 architecture which is known for its accurate object detection features that work in real-time. Our solution is optimized for lightweight operation using OpenVino, guaranteeing scalability across variety of IOT devices, in contrast to existing system that frequency struggle with deployment and efficiency. The model is trained on pothole dataset from Kaggle and Roboflow based on which a precision score of 89.2% is achieved.