Development of a Deep Learning-Based Workforce Activity Tracking Method for the Prefabricated Construction Sector
Project no. 22644 · Status: Kapanmış (Yayın bekleniyor) · 2021
Summary
<p>With the advent of the Industry 4.0 era and the digitalization wave, a digital transformation process has commenced across all sectors. Technologies such as artificial intelligence, deep learning, and image processing, which have become commonplace in our daily lives, are increasingly being employed in the industries. The construction industry, like others, is striving to adapt to these advancements. However, the utilization of information technologies in the Turkish construction sector is notably limited compared to other countries. Particularly, practical applications involving object detection or activity recognition using computer images are almost non-existent. Therefore, research endeavors in this field hold significant importance in pioneering the digitalization of the sector. This research aims to employ deep learning methods to identify and temporally express activities related to the workforce, a substantial cost component in construction projects. Within this scope, an algorithmic solution will be developed to automatically detect actions performed by construction workers during their work. This solution will eliminate the need for time-consuming, often observer-dependent workforce activity tracking, enabling an automated process. As part of the research project, a camera system will be set up to record video footage of construction workers at a prefabricated building component manufacturing facility for three months. Video footage of workers engaged in the manufacturing process will be labeled according to their respective activities. The Faster Regional Convolutional Neural Networks (Faster-R-CNN) algorithm will be utilized for activity recognition. This study not only aims to develop a solution for automatically identifying construction workers' activities from video recordings but also to create a new and annotated dataset specific to construction tasks.</p>