Abstract
Automated pavement distress recognition is a key step in smart infrastructure assessment. Advances in deep learning and computer vision have improved the automated recognition of pavement distresses in road surface images. This task, however, remains challenging due to the high variations in road objects and pavement types, variety of lighting condition, low contrast, and background noises in pavement images. In this dissertation, we propose novel deep learning algorithms for image-based road condition assessment to tackle current challenges in detection, classification and segmentation of pavement images. Motivated by the need for classifying a wide range of objects in road monitoring, this dissertation introduces a Multi-Scale Convolution Neural Network (MCNN) for multi-class classification of pavement images. MCNN improves the classification performance by encoding contextual information through multi-scale input tiles. Then, an Attention-Based Multi-Scale CNN (A+MCNN) is proposed to further improve the classification results through a novel mid-fusion strategy for combining multi-scale features extracted from multi-scale input tiles. An attention module is designed as an adaptive fusion strategy to generate importance scores and integrate multi-scale features based on how informative they are to the classification task. Finally, Dual Attention CNN (DACNN) is introduced to improve the performance of multi-class classification using both intensity and range images collected with 3D laser imaging devices. DACNN integrates information in intensity and range images to enhance distinct features improving the objects classification in noisy images under various illumination conditions. The standard road condition assessment includes determining not only the type of defects but also the severity of detects. In this regard, a pavement crack segmentation algorithm, CrackSegmenter, is proposed to detect crack at pixel level. The CrackSegmenter leverages residual blocks, attention blocks, Atrous Spatial Pyramid Pooling (ASSP), and squeeze and excitation blocks to improve segmentation performance in pavement crack images.
Notes
If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu
Graduation Date
2022
Semester
Summer
Advisor
Yun, Hae-Bum
Degree
Doctor of Philosophy (Ph.D.)
College
College of Engineering and Computer Science
Department
Civil, Environmental and Construction Engineering
Degree Program
Civil Engineering
Identifier
CFE0009167; DP0026763
URL
https://purls.library.ucf.edu/go/DP0026763
Language
English
Release Date
August 2022
Length of Campus-only Access
None
Access Status
Doctoral Dissertation (Open Access)
STARS Citation
Eslami, Elham, "Analytical Study of Deep Learning Methods for Road Condition Assessment" (2022). Electronic Theses and Dissertations, 2020-2023. 1196.
https://stars.library.ucf.edu/etd2020/1196