Abstract
In today’s world, vehicle traffic is increasing on all types of roads, from single-lane streets to multi-lane highways. This rise in traffic emphasizes the need for improved road safety measures to protect drivers and other road users. Ensuring road structural integrity is essential for preventing accidents caused by road damage, such as potholes, landslides, and uneven surfaces, which are common contributors to traffic hazards. Road conditions often lead to media-reported incidents involving vehicle damage, visibility issues due to weather, suspension system damage, and unnecessary traffic congestion. Addressing these issues, this study presents an efficient solution using deep learning for real-time pothole detection, leveraging the YOLOv5 (You Only Look Once) algorithm. Unlike traditional methods like accelerometer-based, image-processing, or basic machine-learning approaches, YOLOv5 provides greater accuracy and is simpler to implement, yielding a promising mean Average Precision (mAP) of 84.5. Moreover, the performance of YOLOv5 can be enhanced by utilizing high-specification GPUs, thus enabling faster and more accurate pothole identification. This approach holds the potential to benefit both the public and government agencies by providing a highly precise and effective pothole monitoring solution. In addition to real-time monitoring, the proposed system allows for proactive maintenance measures, thereby contributing to safer roads and reducing the likelihood of roadrelated accidents.
Keywords: Computer Vision, Drivable Area Segmentation, Pothole Detection, Road Safety, YOLOv5