Shizuoka University
Okabe Lab
日本語

A Method for Generating Training CG Datasets
for Automatic Inspection of Distribution Lines by Drones

Toru Okamura

Shizuoka University

Makoto Okabe

Shizuoka University

研究のTeaser画像

3D models of utility poles generated by this method. The left two images show the pole in a broken state, and the right two images show the pole in a normal state.This method allows the simulation to be performed by randomly setting the support method of the poles, the location of the break, and the number of poles. The background is a Poly Haven image, and the texture of the insulators is reproduced using images from the NGK product information page. The method used these image data sets to train the disconnection detector.

Abstract

We are conducting research on a method for automatic inspection of power distribution lines by drones. Currently, the images taken by drones for inspection are checked visually, but in order to reduce this workload, we studied a method to automatically determine whether or not a pole in the image is disconnected by using a detector that determines whether or not the pole is disconnected (hereinafter referred to as a “disconnection detector”). One of the challenges of this research is that the frequency of pole breakage events is low, so there are few images of broken poles to train the breakage detector. Therefore, to compensate for the lack of data sets, this study expanded the data set by generating various patterns of utility poles in 3DCG and physically simulating the disconnection. In addition, we used the image dataset obtained from the 3D models of the poles generated by this method to train a disconnection detector and evaluate its accuracy.

Paper

Master's thesis

IPSJ SIG-CGVI, 197th conference

Video

BibTeX citation


@masterthesis{okamura.toru.19_master,
  title        = {A Method for Generating Training CG Datasets for Automatic Inspection of Distribution Lines by Drones},
  author       = {Toru Okamura},
  year         = 2025,
  month        = {February},
  address      = {3-5-1 Johoku, Chuo-ku, Hamamatsu City, Shizuoka Prefecture},
  school       = {Shizuoka University},
  type         = {Master's thesis}
    }