Automated image-based condition assessment of the built environment: A state-of-the-art investigation of damage characteristics and detection requirements

Abstract

Inspection activities are intended to detect damage, informing subsequent maintenance and repair actions. Considering the difficulties and limitations associated with traditional inspection methods, there have been growing interests in utilizing image processing and computer vision strategies. Despite several developments, however, the state of the practice still lacks necessary insights on how such advanced strategies should be utilized to realize their expected benefits, in terms of ease, coverage, and accuracy. Considering this critical gap, the current study systematically investigated various types of damage and how they can be evaluated in a condition assessment framework. For the automated detection, localization, and measurement of damage, various convolutional neural network, support vector machine, and classification-based methods were examined, including their advantages and limitations. This study’s recommendations are anticipated to assist researchers and practicing engineers with the proper selection and use of automated damage detection for improving how the built environment is inspected and maintained.