Point Cloud-Based Concrete Surface Defect Semantic Segmentation

IF 4.7 2区工程技术 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Journal of Computing in Civil Engineering Pub Date : 2022-06-20 DOI:10.7146/aul.455.c227

Neshat Bolourian, M. Nasrollahi, Fardin Bahreini, A. Hammad

引用次数: 3

Abstract

Structural inspection is essential to improve the safety and sustainability of infrastructure systems, such as bridges. Therefore, several technologies have been developed to detect defects automatically and accurately. For example, instead of using naked eye for bridge surface defect detection, which is subjective and risky, Light Detection and Ranging can collect high-quality 3D point clouds.This paperpresents the Surface Normal Enhanced PointNet++ (SNEPointNet++),which is a modified version of thewell-knownPointNet++methodapplied to thetask of concrete surface defectdetection. To this end, a point clouddatasetfrom three bridges in Montreal was collected, annotated,and classified into the three classesofcracks, spalls, and no-defects. Based on the comparison between the results(IoU)from the proposed method and similar researchdoneon the same dataset, there areat least 54% and 13% performance improvementsin detecting cracks and spalls,respectively.

查看原文本刊更多论文

基于点云的混凝土表面缺陷语义分割

结构检查对于提高桥梁等基础设施系统的安全性和可持续性至关重要。因此，为了自动准确地检测缺陷，人们开发了多种技术。例如，用肉眼检测桥梁表面缺陷具有主观性和风险性，光检测与测距可以采集到高质量的三维点云。本文提出了表面法线增强PointNet++ (SNEPointNet++)，它是对著名的npointnet ++方法的改进版本，用于混凝土表面缺陷检测任务。为此，我们收集了蒙特利尔三座桥梁的点云数据集，对其进行了注释，并将其分为裂缝、碎片和无缺陷三类。基于该方法与相同数据集上的类似研究结果(IoU)的比较，在检测裂纹和碎屑方面，该方法的性能分别提高了54%和13%。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Journal of Computing in Civil Engineering 工程技术-工程：土木

CiteScore

11.90

自引率

7.20%

发文量

审稿时长

6 months

期刊介绍： The Journal of Computing in Civil Engineering serves as a resource to researchers, practitioners, and students on advances and innovative ideas in computing as applicable to the engineering profession. Many such ideas emerge from recent developments in computer science, information science, computer engineering, knowledge engineering, and other technical fields. Some examples are innovations in artificial intelligence, parallel processing, distributed computing, graphics and imaging, and information technology. The journal publishes research, implementation, and applications in cross-disciplinary areas including software, such as new programming languages, database-management systems, computer-aided design systems, and expert systems; hardware for robotics, bar coding, remote sensing, data mining, and knowledge acquisition; and strategic issues such as the management of computing resources, implementation strategies, and organizational impacts.