This research explores the integration of advanced deep learning techniques for automated pothole and speed bump detection, highlighting various methodologies from recent literature. A systematic review reveals that models utilizing Convolutional Neural Networks (CNNs), including YOLOv3, VGG16, and EfficientNetB0, have achieved impressive accuracy rates in real-time road surface monitoring. In particular, the EfficientNetB0 model was fine-tuned using a comprehensive dataset comprising 400 annotated images of potholes and speed bumps, collected under diverse environmental conditions. The model achieved a validation accuracy of 91.91%, demonstrating robust performance in identifying road anomalies. Notably, the implementation of advanced data augmentation and regularization techniques, such as dropout and L2 regularization, contributed to preventing overfitting and enhancing generalization across varying contexts. This study underscores the potential of deep learning frameworks in improving road safety and maintenance efficiency, paving the way for future enhancements, including dataset expansion and real-time application integration.