AI-Driven Matching of Pavement Surface Images Captured by In-Vehicle Smartphones

Abstract

Pavement monitoring is essential for ensuring the safety, efficiency, and longevity of transportation infrastructure. By regularly assessing pavement conditions, authorities can identify hazards such as cracks, rutting and potholes, enabling timely repairs to prevent accidents and injuries. Moreover, monitoring allows for the optimization of resources, ensuring that maintenance efforts are targeted where they are most needed, thereby extending the lifespan of pavements and reducing long-term costs. Non-destructive testing methods valuable data without harming the pavement structure, though they may vary in complexity, cost, and the depth of information they provide. Imaging technology has been widely applied in pavement detection field, there are several types of pavement inspection devices for different pavement distress, for example, digital camera to detect the pavement crack, rutting and potholes. This study presents the chronological evolution of pavement surface images got by a smartphone over a three-year duration. Initially, video footage of the road surface is recorded using a smartphone securely affixed to a vehicle's interior windshield at an oblique angle. Given that the recorded video is not captured from a top-down perspective, its distortion gives it suboptimal for analysis; thus, it is transformed into a bird's-eye view. Subsequently, these images undergo sorting and alignment based on GPS segments. However, it is noted that sorting by GPS may lack precision due to inherent errors and variations in vehicle speed. Each measurement yields its sequence of images, making exhaustive comparison computationally impractical. A method for finding potential pairs for each image is developed to solve this. This involves a preliminary screening of a select subset of images, which are then matched utilizing AI-based feature matching. A stitching procedure is implemented upon identifying the best pairs from two distinct measurements. Following this method, variations in pavement surfaces, such as the proliferation of crack intersections, enlargement of potholes, emergence of new potholes, or instances of repair work, are observed.