Automated Assessment of Concrete Cracks in Bridges by Exploiting UAV-Based Lidar and Camera Data

A collection of algorithms processing RGB image with Lidar data for detection and quantification of concrete cracks with high accuracy

Institute Reference: INV-22013

Background

Crack detection and characterization of concrete bridges is extremely important to define maintenance strategies and to optimize interventions. Existing crack assessment methods normally require each region of interest to be identified and segmented manually. Manual inspection methods are labor-intensive, time-consuming, and can even be life-threatening. Also, they are prone to human error, and they often require the use of expensive inspection means, such as under-bridge trucks. An automated computerized system based on image processing and laser scanning can reduce time, faulty inspection, risk, and cost. Various platforms are used to collect data, including, terrestrial/fixed platforms, ground-based robotic platforms, climbing robotic platforms, and unmanned aerial vehicles (UAV). UAV with high-performance vision sensors has received considerable attention among other platforms due to maximal mobility. However, data collected by UAV is noisy, which necessitates designing powerful algorithms to compute crack quantities and size.

Technology Overview

Researchers at Northeastern designed a new crack inspection method, comprising a collection of algorithms that can automatically process UAV-based lidar data and camera data. The proposed algorithms automatically extract individual concrete structural members from the sensor data and assess the concrete cracks in each structural member separately. The optimal locations and sizes of the candidate image patches are computed adaptively based on the depth maps retrieved from the lidar data. The resultant image patches perfectly cover the entire region of interest and have a consistent pixel density with the training datasets. This enhances the accuracy of crack detection and is an improvement to the sliding window-based crack detector. Exploiting the depth maps with the quantification algorithm also helps to compute crack lengths and widths in both the unit of pixels and the unit of measurements. This is an improvement over existing crack quantification methods, which either compute crack quantities only in the unit of pixels or rely on a reference object with a known size. The proposed algorithms offer flexibilities to address data with different data qualities, including those noisy data collected by UAV-based platforms. Consequently, this invention enhances the efficiency and accuracy of current bridge inspection practices.