In construction sites, safety issue has always been a threatening risk that must be taken seriously. In order to improve the safety management capability of construction site, this paper establishes a construction site fall hazard management system. This system takes computer vision instead of sensor systems as an information exchange bridge between the physical construction site and the virtual model. The main contribution of this paper is to propose a deep learning-based approach to achieve bi-directional information exchange from physical construction sites to digital models. To train the deep learning model, this paper builds a dataset containing approximately 4000 images from multiple sources to detect on-site safety risks. This paper also proposes a method for acquiring coordinates of workers and risk sources, then establishes an automated safety management framework based on the real-time locations of risk sources and workers. The effectiveness of this safety management framework is verified through a case study comparing the differences between safety officers and the automated system. In addition, the method proposed in this paper integrating building information modeling (BIM) with the real construction process, achieving bi-directional coordination between the digital level and the physical level, providing data basis for intelligent construction.

Tower cranes can cover most of the area of a construction site, which brings significant safety risks, including potential collisions with other entities. To address these issues, it is necessary to obtain accurate and real-time information on the orientation and location of tower cranes and hooks. As a non-invasive sensing method, computer vision-based (CVB) technology is widely applied on construction sites for object detection and three-dimensional (3D) localization. However, most existing methods mainly address the localization on the construction ground plane or rely on specific viewpoints and positions. To address these issues, this study proposes a framework for the real-time recognition and localization of tower cranes and hooks using monocular far-field cameras. The framework consists of four steps: far-field camera autocalibration using feature matching and horizon-line detection, deep learning-based segmentation of tower cranes, geometric feature reconstruction of tower cranes, and 3D localization estimation. The pose estimation of tower cranes using monocular far-field cameras with arbitrary views is the main contribution of this paper. To evaluate the proposed framework, a series of comprehensive experiments were conducted on construction sites in different scenarios and compared with ground-truth data obtained by sensors. The experimental results show that the proposed framework achieves high precision in both crane jib orientation estimation and hook position estimation, thereby contributing to the development of safety management and productivity analysis.

Tianjin 117 is a high-rise building comprised of a reinforced-concrete core and perimeter steel framing that includes megacolumns and megabraces, with a height of 597 m. As the construction process goes on, the influences of the external environment and loads are more crucial to structural safety. In order to verify the structural properties and behaviors in the construction stages, a structural health monitoring (SHM) system is developed in this paper by the authors’ research group. Various sensors are used in this SHM system to monitor different structural parameters that significantly affect the structural safety. The collected data can be analyzed using specifically developed software, thereby used for assessment of structural performance during construction. Interlinked with finite-element software, structural monitor/analysis/evaluation software is developed based on the idea of modular design, which can realize data exchange and result comparison between on-site monitoring and numerical mod...

The location information of entities in construction sites, such as workers and construction machines, is valuable in project management and safety. Therefore, as nonintrusive and accurate solutions, various vision-based methods have been proposed to track entities in construction sites and obtain their three-dimensional (3D) coordinates. However, most existing vision-based methods realize 3D localizations by basing entity matching on the epipolar line, which brings instability in entity matching due to the calculation error of the epipolar line or failure to match entities when multiple entities are located on the same epipolar. To solve this problem, a novel framework based on reidentification is proposed to automatically match workers across two camera views, thereby obtaining their 3D coordinates in construction sites. In this framework, deep-learning-based computer vision algorithms are firstly used to detect and track workers in two camera views. Then, the reidentification (ReID) algorithm is applied to utilize tracked workers’ visual features to match the workers across both two camera views and different frames. As a result, for every matched pair, the worker’s pixel locations in two camera views can be obtained to calculate the 3D coordinates through triangulation. The implementation of videos recorded from a real construction project proves the feasibility and accuracy of this framework. Specifically, through utilizing the ReID algorithm to match workers, the framework achieves competitive results on workers matching with precision, recall, and accuracy of more than 99%, 93%, and 93%. Furthermore, it also effectively addresses the practical problems of ID repetition and ID switching. Meanwhile, this paper extends the application scenarios of reidentification algorithms in construction sites, thereby contributing to the future application of multiple-camera vision-based methods in construction sites.

Traditional approaches to modular construction progress monitoring and quality control with stringent and tight tolerances for on-site and off-site assembly processes are usually based on 3D laser scanning, but the high equipment costs associated with acquiring point clouds have economic ramifications. This paper provides the details of a new and inexpensive method through the integration of AprilTags and an ordinary phone. By using AprilTags instead of QR codes to label modules, progress management is achieved through the rapid identification and association of multiple modules based on a single image. Moreover, a virtual multi-view vision algorithm based on AprilTags is proposed to generate 3D reverse models of the construction site; the quality result can be acquired by comparing the offset and rotation values of the reverse model and the BIM model. Finally, all the algorithms are validated through comparing the reverse models with the reference models made with 3D printing and 3D laser scanning, which verifies the accuracy and efficiency of the proposed method.