SBIR Phase I:Automating Element-Level Inspection of Civil Infrastructures through Computer Vision

The broader impact of this Small Business Innovation Research (SBIR) Phase I project is to develop a damage detection analysis software platform to reduce the risk of failure for heavy civil infrastructure assets (e.g., bridges and dams) while optimizing the capital allocation for these vital systems using a unique combination of artificial intelligence (AI) and civil engineering. The public utilities responsible for civil infrastructureupkeep and assessment need fast, reliable solutions to collect massive amounts of data and rapidly assess the status of large amounts of heavy civil infrastructure.They also need to be able to make accurate and timely reports to stakeholders on the structural integrity of these assets. This project aims to facilitate data-driven decision making for civic leaders regarding infrastructure investments.This Small Business Innovation Research (SBIR) Phase I project will develop and commercialize technologies to increase the automation of the infrastructure asset management process and reduce the costs of this process for asset owners. This increase in automation may also lead to improved conditions and temporal change assessment through AI-powered analytics. The resulting analytics would improve inspection practices by increasing accuracy and objectivity, leading to safer infrastructure systems and fewer infrastructure failures.The technical innovation of this Phase I project is a 3D computer vision pipeline designed to process remotely sensed data and automatically transform it into a format that meets the reporting needs of engineers. The technology addresses long-standing challenges associated with using 3D remote sensing data for infrastructure asset management by developing the foundational technical capabilities to automatically segment and transform 3D point clouds into high-resolution 2D orthomosaics of infrastructure components. Existing approaches are not flexible or generalizable, either employing highly constrained geometric approaches or statistical deep learning models, for which relevant infrastructure data sets do not exist. The proposed approach will utilize computational geometric analysis to isolate and segment point clouds into individual structural components. The process will be prototyped and revised through user pilot studies and ongoing customer discovery and validation activities.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.