A low-cost, video-based visual pavement condition index (VPCI) for network-level pavement condition screening using smartphones and computer vision

Abstract Pavement agencies require frequent condition information, whereas repeated network-level visual surveys are resource-intensive. This study develops a low-cost smartphone-video workflow for qualitative pavement condition screening. Pavement videos collected from selected road sections were sampled to produce more than 58,000 images, from which 30,000 images were retained after removal of unsuitable frames and region-of-interest preparation. Seven distress classes—weathering, raveling, alligator cracking, potholes, patching and utility cuts, longitudinal and transverse cracking, and block cracking—were labeled at low, medium, and high apparent severity levels for distress recognition. Because image data do not represent all variables required for exact pavement condition assessment, only four visually observable distress groups—alligator cracking, potholes, longitudinal and transverse cracking, and block cracking—were used to define the Visual Pavement Condition Index (VPCI). A YOLOv8 detector was used for visual distress identification, and a decision-tree model was used to predict qualitative pavement condition categories from visual distress deduct-value features. The reported precision was 58.47% on the primary study data and 51.20% on an independent validation dataset comprising 18 road sections. These results indicate that the proposed VPCI is a preliminary visual screening indicator rather than a replacement for conventional field-based pavement condition assessment; additional data development and validation are required before operational application.