Inspection of concrete structures is essential to ensure the safety and structural integrity of civil infrastructures such as bridges, tunnels, dams, and buildings, etc. The conventional manual inspection approach of sending an on-site inspector to target locations is labor-intensive and time-consuming. Besides, it also causes a huge risk of generating untrackable data and inaccurate metric measurements. This thesis addresses two major challenges in the automatic inspection of concrete structures: 1) data collection using robots and visual sensors, and 2) visual data analysis algorithms. The contributions of the dissertation research are summarized in the following perspectives.
First, we develop a new depth sensor calibration method and a depth enhancement algorithm to enhance the performance of the perception sensor. For depth sensor calibration, we assume the pixels are spatially independent, thus the depth calibration can be formulated as a per-pixel fitting and approximation problem. Besides, a new depth calibration system is developed for per-frame pose estimation. For depth image completion, we introduce a new hybrid filtering algorithm that consists of a stride-varying Gaussian filter and a bilateral filter completion algorithm to fill large holes.
Second, we propose a depth adaptive image pyramid network for region detection and an edge-sensitive defect segmentation network for contour generation, for concrete inspection. For region detection, we propose a new depth adaptive window selection algorithm, and we also introduce an image pyramid network to increase the detection performance with varying resolution input. To obtain a fine defect contour, we propose an edge-sensitive model to alleviate the loss contribution from the spalls, by introducing side layers to capture the crack area.
Third, we create a concrete defects data set and make it publicly accessible to benefit the research community. This data set consists of two major defects, i.e., concrete cracks and concrete spalls. The data set contains 298 spalls image frames in which all images are labeled over the rebar and the complete area, and 954 crack images in which 522 images are labeled. We provide the detection and inspection baselines to the public for comparison.
Fourth, we develop a robotic inspection system that consists of software with a pose estimation algorithm, smart defect detection algorithm, and a metric measurement algorithm to cover the needs for metric inspection and apply them on a drone for fast, coarse inspection and on a wall-climbing robot for detailed close-up inspection. Field test results demonstrate that our robotic inspection system could provide comprehensive inspection in the field.
|Commitee:||Uyar, M. Umit, Xiao, Jizhong, Zhu, Zhigang, Tian, Yingli, Jiang, Hao|
|School:||The City College of New York|
|School Location:||United States -- New York|
|Source:||DAI-B 81/8(E), Dissertation Abstracts International|
|Subjects:||Robotics, Electrical engineering, Artificial intelligence|
|Keywords:||Automated data collection, Concrete inspection, Crack/spalling detection, Metric Measurement, UAV Inspection system, Wall-climbing robot|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be