Skip to the content.

Deep Learning in Science and Engineering

Youngjib Ham, Department of Construction Science
Deep Learning in Construction Science: Automated Contextual Information Analysis for Resource Allocation

Abstract

The construction industry has long been suffered from low productivity due to the nature of construction projects - unique and dynamic outdoor environments. Although previous studies have proposed earthwork productivity analysis methods to obtain a better productivity, the time-consuming and laborious data collection process remains as a major obstacle to implement it in practice. Consequently, the construction industry still relies on empirical decision making to arrange earthwork operations. In this talk, an automated productivity analysis method using vision-based jobsite monitoring with focus on an earthmoving process is introduced. Deep learning-based object detection model is used to analyze jobsite context information and then the produced context information is used as a simulation input to yield an optimal resource allocation plan. In addition, current state-of-the-art applications of deep learning and the underlying challenges of utilizing deep learning techniques in the construction domain will be discussed.