With the extensive development in the infrastructures, road crack maintenance has become a critical issue in our day-to-day life. Specially, in concrete-structured constructions like roads, monuments, and bridges, cracking is a typical issue. Letting it to grow will increase the danger of accidents and cause considerable financial losses. numerous methods have been developed in these directions (road crack detection and segmentation) but there isn’t a proven technique for dealing with noisy, poor-quality real-world road crack photos. In this research paper a deep-learning based method has been proposed namely, Two-phase Convolutional Neural Network at pixel-level for road crack detection and segmentation. The first phase aids to remove noise and separate the small cracks whereas second phase labels the crack detected area and learn the actual context of crack. Hence, it shows higher impact on learning over the original noise image. The experiments have been performed done on two-publicly accessible benchmarks i.e., CFD dataset and Crack500 dataset. There are many methods and algorithms that are satisfactory in pavement crack applications, but there is no standard until today. Therefore, in order to know the developing history and the advanced research, we have collected a number of literatures in this research topic for summarizing the research artwork status, and giving a review of the pavement crack image acquisition methods and 2D crack extraction algorithms. The results on these datasets demonstrates that the two-phase CNN method outperform better results as compared to existing approaches, particularly for noisy and imbalanced datasets. Our analysis gives the precision of about 97.82% for the crack image detection and in pixel-level segmentation accuracy comes out to be approx. 95.40%.