UAS-Based Crack Detection Using Stereo Cameras: a Comparative Study

Structural health monitoring and inspection of bridges are paramount to evaluating their current conditions and identifying the severity of potential defects. These may be indicative of spalls or reinforcement corrosion. As of today, 25% of all the bridges across the United States are rated structurally deficient or functionally obsolete [1]. Currently, health monitoring of bridges is a human-based visual inspection process, which is labor intensive, costly and potentially unsafe. Safety issues, significant cost, as well as traffic interruption highlight the need to explore a reliable, low-cost, quantitative and safe solution for bridge condition assessment. Drone inspections using Unmanned Aerial Systems (UAS) have recently attracted significant industrial and academic interest. Various technologies are being explored including Lidar, photogrammetry and depth sensors, with the latter being a promising innovative approach for accurate 3D object reconstruction. In this study, we investigate the use of stereo cameras based on passive and active depth calculation for the structural integrity assessment of bridges. We conduct an experiment to determine the RMS error of two different vision sensors: The Intel Realsense D435i and the ZED stereo camera from Stereolabs. Our results show that for our application, the Intel Realsense D435i provides more accurate information.