Abstract

According to the UK Royal Academy of Engineering and the US National Academy of Engineering, mapping our existing infrastructure and its condition inexpensively, and automating construction through computer science and robotics advancements are two of the engineering grand challenges for this century. This seminar outlines a series of research projects that help address these challenges based on a common computer vision framework; a general approach to creating visual pattern recognition models that automate the recognition of infrastructure objects based on their visual and spatial characteristics. The projects presented create such models to a) measure the size of objects, such as dimensions of roof panels, for as-built geometric modeling and quantity take-offs, b) count objects, such as bricks, workers, and trucks, for comparing the as-built with the design and measuring sustainability and productivity, c) spatially track materials, personnel, vehicles and equipment to automate project and congestion monitoring and control systems, and d) detect defects/damage on objects, such as cracks, spalling, air pockets, and potholes, for structural damage assessment and maintenance decision making