The study presents a novel approach for road crack detection that leverages the ResNet-50 architecture alongside advanced preprocessing methods and machine learning algorithms. To ensure robust learning, the model was trained on a large dataset of 40,000 photographs that were evenly distributed between intact and cracked road surfaces. Using this dataset, advanced convolutional neural networks, in particular ResNet-50, were trained and optimized to accurately detect cracks. Methods like data augmentation, which broadens the variety of training images, and stochastic gradient descent, which maximizes learning, were used to improve model performance. The validation accuracy metric showed that ResNet-50 had the highest accuracy of 92.13 when compared to five other architectures: VGG-16, VGG-19, Faster R-CNN, DenseNet, and YOLO v2. This performance advantage shows how well ResNet-50 detects cracks because it outperformed the other models in terms of accuracy and computational efficiency. The model’s ability to discriminate between intact and cracked surfaces was improved by the preprocessing methods, such as noise reduction and contrast enhancement, which were crucial in honing the input data. The study also looked at the automated crack detection system’s scalability and deployment potential in relation to current road maintenance technologies. This research offers a significant advancement in infrastructure management by enabling proactive road monitoring, which could lead to enhanced road safety and reduced maintenance costs.