ABSTRACT
Traditional inspections are typically performed manually. They require specialized equipment, which usually is expensive, time-consuming and can often be a source of risk. Unmanned Aerial Vehicles (UAV) provide an alternative to overcome the challenges of traditional manual inspections. This has been stud- ied as well as the use of machine learning and image analysis algorithms for detection and quantification of damages, in particular cracks. This paper presents a framework for inspections that combines data acquisition, crack detection, and the quantification of essential parameters of concrete cracks. From this method, the width and length of cracks are determined and compared with control measurements to estimate the accuracy of the method. The results show that using the pre-trained network to detect cracks, only a few to no false negatives are obtained. From the images classified as containing cracks, the quantification methodology is performed obtaining measurements of crack width down to 0.13 mm.
