Pothole Detection Using Image Enhancement GAN and Object Detection Network

Abstract

Many datasets used to train artificial intelligence systems to recognize potholes, such as the challenging sequences for autonomous driving (CCSAD) and the Pacific Northwest road (PNW) datasets, do not produce satisfactory results. This is due to the fact that these datasets present complex but realistic scenarios of pothole detection tasks than popularly used datasets that achieve better results but do not effectively represents realistic pothole detection task. In remote sensing, super-resolution generative adversarial networks (GAN), such as enhanced super-resolution generative adversarial networks (ESRGAN), have been employed to mitigate the issues of small-object detection, which has shown remarkable performance in detecting small objects from low-quality images. Inspired by this success in remote sensing, we apply similar techniques with an ESRGAN super-resolution network to improve the image quality of road surfaces, and we use different object detection networks in the same pipeline to detect instances of potholes in the images. The architecture we propose consists of two main components: ESRGAN and a detection network. For the detection network, we employ both you only look once (YOLOv5) and EfficientDet networks. Comprehensive experiments on different pothole detection datasets show better performance for our method compared to similar state-of-the-art methods for pothole detection.