Bridge Inspection Using Drone & AI

Bridge inspection is a vital aspect of civil infrastructure maintenance, ensuring structural integrity and public safety. However, traditional inspection methods are often time-consuming, labor-intensive, and pose safety and accessibility challenges. This thesis presents an innovative, cost-effective solution that leverages drone technology and artificial intelligence to automate and enhance the bridge inspection process.

The project utilizes a DJI Spark drone, selected for its portability and affordability, in combination with a DJI Osmo Pocket 3 camera equipped with an in-built gimbal for image stability. Automated flight paths and camera operations were programmed using DroneBlocks, enabling consistent and precise image capture, even in hard-to-access under-deck areas. A scaled-down physical model of the Columbia Avenue Bridge was created using a concrete slab with real and artificial cracks to evaluate inspection accuracy. The slab was elevated using a welded sawhorse structure to simulate real-world under-bridge conditions.

Images captured during drone flights were processed using Structure from Motion (SfM) techniques in software such as MeshLab and Polycam to reconstruct a detailed 3D point cloud of the bridge surface. Two approaches were used for crack detection: a U-Net-based convolutional neural network trained on a labeled dataset of concrete cracks, and a heuristic method based on surface curvature analysis of the point cloud. The AI model mapped crack locations and dimensions, while the heuristic approach provided a lightweight and interpretable alternative for detecting surface anomalies.

Despite limitations related to drone payload and on-site processing capabilities, the system effectively visualized structural defects and enabled semi-automated condition assessments. This work demonstrates the practical integration of consumer-grade drones, open-source tools, and AI-driven analysis to modernize and streamline structural health monitoring in a safe, efficient, and scalable manner.