In this paper, we developed techniques to identify and quantify the damage (crack) to bridges based on images obtained by the unmanned aerial vehicle (UAV). The scope of the research includes image acquisition using UAV, the classification system of crack based on Deep-learning and algorithms of detection and quantification using improved Image Processing Techniques (IPTs). A conventional crack detection method using only IPTs can be applied marginally according to the image acquisition environment (lights, shadows, etc.), so we proposed the techniques based on Deep-learning to find the crack part in the region of interest (ROI) from the other types of damage or non-crack. After classifying the crack part in the ROI, improved IPTs are applied to the detected regions to quantify cracks at 300 micrometers. Performances of the technique were evaluated through preliminary test and field test. The non-contact bridge damage detection technology using UAV can be applied to the actual bridge inspection field It is expected to have more performance than existing bridge inspection methods.
It is expected that bridge inspection using unmanned aerial vehicles (UAVs) equipped with imaging devices is able to improve public safety and structural reliability by providing the close detail of a bridge appearance. Thus, interests in bridge inspection using UAVs are increasing worldwide. However, at present, most of them simply use commercially available UAVs to acquire images of parts of the bridge that are difficult to access (e.g., upper parts of the pylons in long-span bridges). It cannot meet the final goal of bridge inspection that is to assess the condition of the bridge. Therefore, this approach is still considered to be at an early stage from a practical point of view and a more systematic and reliable approach is needed. In this paper, challenging issues of bridge inspection using UAVs are identified and their solutions are presented. To this end, a recently launched research project is introduced by describing the developing core technologies such as a new UAV-localization algorithm without GPS, noncontact inspection techniques based on data fusion of hybrid images and a bridge condition evaluation technique based on the processed data originally obtained from UAVs. Some interim results of field tests are also presented.