A Road Damage Detection Based on Contrast Limited Deep Convolution Neural Networks for Urban Roadways

Road crack detection is a crucial safety measure for any country, especially in regions with complex road networks. In India, most roads are well-connected to cities and urban areas. However, urban roads often suffer frequent damage due to various factors. This study focuses on detecting road damage in Indian urban areas using a Deep Convolutional Neural Network (DCNN). We have developed a model specifically designed for identifying cracks in Indian urban roads. To evaluate the proposed model, we collected over 700 images of damaged roads from different urban locations across Tamil Nadu. In this work, we employed the DCNN algorithm for road crack detection, which has proven to be an efficient approach. The proposed method incorporates Histogram Equalization (HE) and Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance the contrast of road images. This is the first initiative aimed at developing a road damage detection system tailored to urban roadways in India. A comparative analysis was conducted during the preprocessing stage by applying both HE and CLAHE techniques. The model was trained using 5,000 roadside images, including both cracked and non-cracked surfaces. During training, image enhancement was performed using HE and CLAHE, and the processed images were then used to train DCNN. For testing, 700 roadside images (with and without cracks) were utilized, following the same preprocessing steps. The model's accuracy was determined based on its ability to correctly identify road cracks. Results indicate that the proposed model performs effectively for crack detection on CLAHE-enhanced images of Indian urban roads. Additionally, the proposed model's performance was compared with existing models such as ResNet, VGG16, and VGG19 using the same dataset. Evaluation metrics including accuracy, precision, recall, and F1-score were used. The proposed model achieved 98.6% accuracy, 98.5% precision, 99.6% recall, and a 99% F1-score.