Project Details

Abstract

Visual inspection at regular intervals has traditionally been the primary method for assessing the condition of transportation assets to ensure they meet performance objectives. However, this method is labor-intensive, costly, poses safety risks to inspectors, and may suffer from quality inconsistencies. These challenges have driven the adoption of new inspection technologies such as drone imagery and LiDAR. However, the abundance of data generated from these technologies motivates the development of automatable and reliable methodologies for data processing to understand asset conditions and performance. Computer vision (CV) techniques offer an efficient means to process visual data and extract a high-level understanding of images and videos. However, the current CV-based techniques ignore the "context" of collected data, limiting their applicability and generalizability. This study aims to develop robust, context-aware CV models with low inference times that provide actionable insights on asset conditions. The proposed models will be applied to steel bridges, and the impact of various spatial and temporal contexts on CV model performance will be examined. The project outcome will advance the state of the art of using CV models for bridge inspection and provide opportunities for integrating these technologies into integrated asset management systems.

Project Word Files