Deep Neural Network based Estimation of Missing Measurement Data for Construction Path Optimization

Yuki Yoshino, Chieko Hoshino, Yutaka Uchimura

Research output: Contribution to journalArticlepeer-review

Abstract

With a rapidly aging population, declining birthrate, and decreasing number of skilled workers, automation of construction machinery is expected. On construction sites, automated earth moving work by a bulldozer requires the measurement of the soil pile. However, measuring entire pile data using sensors mounted on the bulldozer is difficult, since the back side of the soil pile becomes blind spot. To solve this problem, it is required to generate the path of the pile spread only using the image of the part visible from the bulldozer. This paper proposes a method to generate digital evaluation model (DEM) of the soil pile from occluded measurement pile data using a convolutional neural network. Since deep learning-based methods require a large amount of training data, we generated pile data and captured images via simulations. Considering the sensing device, three image patterns and their estimation accuracy were evaluated. By using the trained network model, construction path optimization for earth moving tasks was performed using the estimated DEM data. The results of DEM estimation and filling performance of soil moving tasks are also shown.

Original languageEnglish
Pages (from-to)171-177
Number of pages7
JournalIEEJ Journal of Industry Applications
Volume13
Issue number2
DOIs
Publication statusPublished - 2024

Keywords

  • construction automation
  • construction path optimization
  • deep neural network
  • digital evaluation model (DEM)

ASJC Scopus subject areas

  • Automotive Engineering
  • Energy Engineering and Power Technology
  • Mechanical Engineering
  • Industrial and Manufacturing Engineering
  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Deep Neural Network based Estimation of Missing Measurement Data for Construction Path Optimization'. Together they form a unique fingerprint.

Cite this