Detection and Classification of Road Damage Using Deep Learning Approach With Smart Phone Images

Abstract

In cities, road surface monitoring is mostly done by hand which is a time-consuming and labor-intensive procedure. One of the most critical responsibilities is infrastructure maintenance work for traffic safety. To keep the road network safe, it must be assessed on a regular basis to identify potential threats and risks. We work on detecting and classifying road damage using deep learning approach in this research, which is a low-cost intelligence system. The goal of this work is to investigate the detection and categorization of road damage from road surface photographs using deep learning concept. This study used different transfer learning algorithms to categorize road damage in order to determine which algorithm performed better at detecting and classifying road damage. We divide damages into four groups: potholes, cracks, and revealing and rutting. For this research, we used a smartphone camera to collect data from the streets of Dhaka and processed with it. Our work uses various transfer learning deep neural network algorithms including VGG16, VGG19, ResNet50, MobileNetV2, EfficientNetV2 for classifying road damages, as well as for detection, and it outperforms earlier research. We got the highest 97.15% accuracy for ResNet50 and lowest accuracy 94.88% for MobileNetV2 and EfficientNetV2, 94.31% accuracy for VGG16 and 93.18% for VGG19.