Drone-Based Inspection of the Appearance Defects for a Large Object

Wen June Wang, Xiang Yin Dai, Chun Yuan Cheng, Shang Ming Ciou

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In general, the defect inspection of a large object, such as an aircraft, a bridge, or a building, etc., must need some tools or climbing high to achieve the inspection because the object is vast and high. However, climbing high is dangerous, and relying on other tools takes time and effort. Therefore, this paper aims to establish a drone system for detecting defects in the surface of a large object. In the system, the drone can fly along the object's exterior with the shortest path and adjust the angle of its gimbal such that the drone's camera can inspect the defects in the object's appearance. The shortest path is obtained from solving the Travelling Salesman Problem of the navigation points. The navigation points are built based on the normal vectors of the object's point cloud, which is established using OpenSfMThe shortest path is obtained from solving the Travelling Salesman Problem of the navigation points. The navigation points are built based on the normal vectors of the object's point cloud. The point cloud is created using OpenSfM (Structure from Motion). Adopting Visual Simultaneous Localization and Mapping (V-SLAM) as the drone's position control such that it can fly stably following the shortest path composed of navigation points. After the drone collects the whole image of the object's appearance, the network YOLOv4-P6 is used to recognizes the defects. This study finally proposed an experiment to inspect car defects and found three types of defects: paint loss, corrosion, and dent, successfully and efficiently.

Original languageEnglish
Title of host publicationProceedings of 2023 International Conference on System Science and Engineering, ICSSE 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages474-478
Number of pages5
ISBN (Electronic)9798350322941
DOIs
StatePublished - 2023
Event2023 International Conference on System Science and Engineering, ICSSE 2023 - Virtual, Ho Chi Minh City, Viet Nam
Duration: 27 Aug 202328 Aug 2023

Publication series

NameProceedings of 2023 International Conference on System Science and Engineering, ICSSE 2023

Conference

Conference2023 International Conference on System Science and Engineering, ICSSE 2023
Country/TerritoryViet Nam
CityVirtual, Ho Chi Minh City
Period27/08/2328/08/23

Keywords

  • deep learning network
  • defect inspection
  • drone flying
  • travelling salesman problenb point cloud
  • YOLOv4

Fingerprint

Dive into the research topics of 'Drone-Based Inspection of the Appearance Defects for a Large Object'. Together they form a unique fingerprint.

Cite this