Concrete Crack Detection Using UAV and Deep Learning

Abstract

In recent years, the number of infrastructures to be inspected or repaired has increased, because many of them are suffering from the degradation and deterioration. However, it is difficult to acquire the sufficient number of experienced engineers due to their aging and retirement. Usually, when judging the degree of damage of the infrastructure, a visual inspection is carried out by experienced engineers as a preliminary checking for non-destructive examination. Recently, it is expected to introduce robots such as Unmanned Aerial Vehicles (UAV) for the inspection because it can reduce the burden on engineers and collect data from the spaces where people are difficult to approach. However, it is often difficult to secure sufficient accuracy for the data collected by robots. This problem can be solved by adjusting the photographing angle and distance if possible to confirm the detection accuracy of the crack contained in the photographed image on the spot. Then, robotic inspection is available as a useful tool that can realize the real-time detection. In this research, an attempt is made to develop an efficient crack detection system for concrete bridge structure, using You Only Look Once (YOLO) method that is a method possessing the possibility of realtime processing. In order to demonstrate the applicability of the system with YOLO, an experiment was conducted by applying the system to the crack image of concrete wall. From the experiments, it was found that the detection time was highly small and further detection was possible even in real-time. Regarding the accuracy, although it is possible to identify the location where cracks occurred roughly, it is difficult to detect perfectly, and erroneous detection was also observed. For these reasons, further improvement in accuracy is necessary for practical use.