| Abstract | This paper is investigating learning methods using depth as a cue measurement that can be used for bridge inspection. We investigate learning methods based on mono, stereo, and multiview image input and discuss the constraints that allow some methods to perform better than others in various scenarios. We go over the state-of-the-art deep learning methods, including supervised and unsupervised methods. These methods will be compared and evaluated, based on constraints, performance, and accuracy, and how top methods should be selected for each scenario. The same database should be used for fair comparison between all methods ensuring that evaluations are unbiased, replicable, and meaningful. |
|---|
