The condition of roads is crucial for both user comfort and safety, and regular maintenance and inspections are necessary to identify and repair pavement distress. This paper presents a new approach to collect high-resolution pavement distress data and deep learning models to detect and classify pavement distress accurately. The study aims to achieve two main objectives. Firstly, to enhance image quality and consistency in collecting high-resolution pavement distress data a common limitation of previous research. Secondly, to develop and compare deep learning models that can accurately detect and classify pavement distress.To accomplish the first objective, a custom data collection apparatus is designed that captures only the relevant portion of the pavement surface and minimizes irrelevant details. The study utilizes two datasets to make a comparison, RDD2020, which is a publicly available dataset with pavement distress images, and EV22, a dataset collected using our own custom apparatus.To achieve the second objective, the study utilizes three state-of-the-art deep learning models, namely Mask R-CNN, Cascade Mask R-CNN, and Hybrid Task Cascade. These models are trainedand tested on both the RDD2020 and EV22 datasets, and their performance is evaluated using metrics such as precision, recall, F1 score, and IoU. The results indicate that the approach is effective. The EV22 dataset has higher image quality and consistency, with a larger proportion of relevant pavement area in the images compared to RDD2020. In terms of pavement distress detection, all three deep-learning models outperformed RDD2020.