posted on 2021-02-23, 18:10authored byEric Bianchi, Matthew Hebdon, Pretap Tokekar, Lynn Abbott
Common Objects in Context for bridge inspection (COCO-Bridge) is an image-based dataset for use by unmanned aircraft systems (UAS) to assist in GPS denied environments, flight-planning, and detail identification and contextualization, but has far-reaching applications such as augmented reality. COCO-Bridge was introduced to augment an unmanned aerial vehicle (UAV) conducted bridge inspection process. UAVs have a notoriously difficult time operating near bridges because the signal can be lost between the operator and the UAV. This effort begins the process of building a publicly available dataset, examining model performance enhancements through image augmentation, and hosting a website repository of necessary code, raw images, and annotated data to access and contribute to the advancement of A.I in civil engineering. While there are datasets which have focused on detecting defects, this dataset focused on identifying specific parts of a bridge or structural bridge details to make educated autonomous decisions during flight. The image dataset consisted of 774 images and over 2,500 object instances to detect four structural bridge details. Methods to economize the predictive capabilities of the model without the addition of unique data were investigated to extend the performance of the training images. It was concluded that model performance was improved by selectively augmenting the training data about the y-axis.