A critical review on pavement distress detection using images and point clouds from visual features to geometric modeling

Pavement distress detection plays a pivotal role in ensuring roadway safety, serviceability, and cost-effective infrastructure management. With rapid advancements in intelligent transportation systems, computer vision, and sensing technologies, non-contact detection approaches based on images and point clouds have become increasingly prominent due to their efficiency, objectivity, and scalability. This review systematically examines both image-based and point cloud-based methodologies, structured along the complete detection pipeline encompassing data acquisition, preprocessing, distress extraction, and geometric quantification. Image-based techniques rely on visual cues, such as texture, color, and edge continuity, to identify surface-level anomalies efficiently, benefiting from mature deep learning frameworks for classification, object detection, and pixel-level segmentation. In contrast, point cloud-based methods capture rich three-dimensional geometric and structural information, enabling detailed modeling of crack depth, rutting deformation, and surface irregularities. Although each modality can independently achieve satisfactory performance, their complementary strengths have driven a growing trend toward hybrid frameworks, combining image-based rapid screening with point cloud-based precision modeling, to enhance detection accuracy, robustness, and adaptability across varying conditions. Furthermore, this paper highlights persistent challenges, including multimodal data fusion, high equipment and labeling costs, computational complexity, and the need for standardized benchmarks. By synthesizing current progress and identifying key technical bottlenecks, this review provides a comprehensive foundation and forward-looking perspective for developing intelligent, efficient, and scalable pavement distress detection systems. • Image-based and point cloud-based pavement detections are systematically reviewed. • Strengths, weaknesses, and applicability of current approaches are compareed. • Key challenges, such as blurred boundaries, costly labels, high computation are identified. • Hybrid trends about image screening with precise 3D point cloud modeling are highlighted.