Publications / 2019 Proceedings of the 36th ISARC, Banff, Alberta, Canada

Deep Learning Detection for Real-time Construction Machine Checking

Bo Xiao and Shih-Chung Kang
Pages 1136-1141 (2019 Proceedings of the 36th ISARC, Banff, Alberta, Canada, ISBN 978-952-69524-0-6)

Construction sites require many efforts to be well organized due to the complicated tasks and various construction machines. Recently, computer vision technology has gained success in the construction research field. By deploying single or multiple cameras, we can extract construction information from videos and then help the project manager to understand what happened in real-time. This paper presents a method of checking construction machine status automatically by cameras. We assume that the camera is installed in a high position, which provides clear views for the whole site. This research focuses on extracting the machine information from videos, comparing with construction schedule and feedbacking to project manager for further decision making. In the preliminary stage, a deep learning detector has been employed for detecting active construction machines. Meanwhile, a construction image dataset has been organized for training deep learning models precisely and robustly. This dataset also helps to promote this method to generalized construction scenarios. Comparing the number of active machines of video and the expected number of machines from the schedule, the project manager will get real-time feedback and alerts when missing construction machines. In the future steps, we will develop a method to understand construction activities from videos and highlight important activities automatically. Once the reliable method has been developed, it will benefit the project manager to monitor construction sites from an easier way.

Keywords: Construction Machine; Computer Vision; Deep Learning