Search

Purpose: Contractors must check the provisions that may cause disputes in the specifications to manage project risks when bidding for a construction project. However, since the specification is mainly written regarding many national standards, determining which standard each section of the specification is derived from and whether the content is appropriate for the local site is a labor-intensive task. To develop an automatic reference section identification model that helps complete the specification review process in short bidding steps, the authors proposed a framework that integrates rules and machine learning algorithms. Design/methodology/approach: The study begins by collecting 7,795 sections from construction specifications and the national standards from different countries. Then, the collected sections were retrieved for similar section pairs with syntactic rules generated by the construction domain knowledge. Finally, to improve the reliability and expandability of the section paring, the authors built a deep structured semantic model that increases the cosine similarity between documents dealing with the same topic by learning human-labeled similarity information. Findings: The integrated model developed in this study showed 0.812, 0.898, and 0.923 levels of performance in NDCG@1, NDCG@5, and NDCG@10, respectively, confirming that the model can adequately select document candidates that require comparative analysis of clauses for practitioners. Originality/value: The results contribute to more efficient and objective identification of potential disputes within the specifications by automatically providing practitioners with the reference section most relevant to the analysis target section. [ABSTRACT FROM AUTHOR]

Tower cranes are generally used in modern construction projects due to their exclusive and flexible performance for material transportation in complex workspaces. Although traditional deep learning vision-based methods have proven their effectiveness in construction site monitoring, a shortage of realistic training images and inefficiencies in them have been an obstacle in achieving effective tower crane monitoring. To deal with this issue, this paper presents a database-free approach for timely productivity monitoring of tower crane operations. Using a synthetic model, the authors construct a research framework to effectively generate training images on deep learning–powered object detection. Additionally, an enhanced tracking algorithm collaborating with unsupervised clustering-driven postprocessing enables stable performance for tracking resources. Further, logical ontology-based activity recognition constructed on domain-specific knowledge takes place to estimate time-to-time operation of the tower crane. The curtainwall installation was selected to represent the dynamicity and variability of tower crane operations for which it has been difficult to apply traditional deep learning technologies. Results displayed that the intended framework successfully produced 300,000 training images in 1 h, about 500 times faster than human resources, and an enhanced tracker performed well even in the case of unsatisfactory detection results. It is also proved that a performance of activity analysis was acceptable to 93.58% accuracy. The results demonstrate that the proposed framework has proven remarkable data generation speed with little manual input and good performance of operational-level activity analysis. These findings can automate the monitoring process of tower cranes' operations and the productivity of curtainwall installation, assisting construction project decision making. Furthermore, such efficient methods are less time-consuming and can handle background scene changes effortlessly, building customized models in less time and to less cost, thus buffering the practical gap of applying vision-based technologies into visually dynamic construction sites. [ABSTRACT FROM AUTHOR]

This paper presents a systematic method to create universally applicable synthetic training image datasets for computer vision-based construction object detection. The synthetic images created by inserting a virtual object of interest into a real site image allows us to minimize the time and effort for training image data collection and annotation. In addition, the use of synthetic images has an additional benefit that training images can be easily customized for a target construction site by considering the context of the site (e.g., different background scenes, camera positions, and angles) and the possible variability of target objects to be detected (e.g., different sizes, locations, rotation angles, and postures) on images. An automated approach proposed in this study attempts to systematically create the synthetic images using the Unity game engine in which context- and variability-related parameters can be controlled. The proposed method was validated by training a deep learning-based object detection algorithm [i.e., a faster regions with convolutional neural network (R-CNN) model] with synthetic images and testing it on real images from earthwork construction sites to detect an excavator. The CNN models trained with synthetic images showed an average precision value of more than 90%; in particular, the classifier using synthetic images outperformed the one using real site images. The detection results also demonstrated an improved capability to capture the high irregularity of a construction object on images when using techniques of context customization and variability randomization. The findings from this study demonstrate the feasibility and practicality of the use of synthetic images for vision-based approaches in a construction domain. Ultimately, the proposed approach serves as an alternative way to build comprehensive image datasets for construction entities, contributing to facilitating vision-based studies on construction. [ABSTRACT FROM AUTHOR]

This paper proposes a vision-based action recognition framework that considers the sequential working patterns of earthmoving excavators for automated cycle time and productivity analysis. The sequential patterns of visual features and operation cycles are incorporated into the action recognition framework, which includes three main processes: excavator detection, excavator tracking, and excavator action recognition. The first process uses a pre-trained detector to localize earthmoving excavators on sites from all of captured images. Next, the detection results are associated and the excavators are tracked by Tracking-Learning-Detection algorithm. Lastly, to recognize the operation types of the excavators, their sequential patterns of visual features and operation cycles are modeled and trained with a hybrid deep learning algorithm, i.e., Convolutional Neural Networks and Double-layer Long Short Term Memory. Three experiments were performed to validate the proposed framework and confirm the positive effects of sequential modeling: (1) an experiment without sequential pattern analysis, (2) an experiment with the sequential pattern analysis of visual features, and (3) an experiment with the sequential pattern analysis of visual features and operation cycles. In the experiments, the research team used a total of 72,365 images collected from actual earthmoving sites. The average accuracies of the excavator action recognition in the three cases were 79.8%, 90.9%, and 93.8% respectively. The results demonstrated the applicability of the proposed framework and the significant positive impacts of sequential pattern modeling on the recognition performance. This research contributed to the identification of critical elements that explain sequential working patterns and to the development of a novel vision-based action recognition framework. In addition, the findings of this study can help to automate cycle time analysis and productivity monitoring of earthmoving excavators. • Vision-based action recognition of earthmoving excavators for automated cycle time and productivity analysis • Modeling and learning sequential working patterns of an excavator • Convolutional Neural Network and Double-layer Long Short Term Memory for sequential pattern analysis • A novel action recognition framework that incorporates sequential patterns of visual features and operation cycles. • Acceptable performance with 90.9% precision and 89.2% recall rate [ABSTRACT FROM AUTHOR]

The number of single-person households (SPHs) has been consistently increasing owing to various social issues, such as separation by death, declining marriage rate, and increasing divorce rate. Unfortunately, this demographical change is creating a new social problem, namely, lonely death. In response to this problem, many researchers have attempted to develop wearable sensor-based and computer vision-based systems that monitor occupant behaviors and detect possible emergency events in indoor environments. However, existing approaches face challenges in monitoring SPHs owing to their technical disadvantages; for instance, if the occupant is not wearing the electronic sensor, or if the signal is occluded by other objects, it is not possible to monitor SPHs. Moreover, as existing studies focus only on classifying the occupant's daily activities, such as eating, sitting, and talking, the emergency events that are significant for SPH monitoring are still unclear. To address these challenges, this study investigates emergency events that have a critical impact on the occupant's health and proposes a deep learning-based sound recognition model to monitor occupant behaviors and detect possible emergency events in SPH environments. Experiments are conducted using audio data collected from actual SPH home environments and online data-sharing websites. The average precision and recall rates of the developed model are 78.0% and 90.8%, respectively. The results demonstrate that the developed model could successfully distinguish emergency sound events from the sounds of regular human activities. The findings can not only secure and rescue SPHs in danger but also provide new research directions for indoor occupant and event monitoring. Image 1 • Major emergency events were monitored in single person household environments. • Proposed a deep learning-based sound recognition model to monitor occupant behavior. • Average precision and recall rates were 78.0% and 90.8%, respectively. • Baselines for occupant behavior and abnormal event monitoring. [ABSTRACT FROM AUTHOR]

Since most state-of-the-art vision technologies have recently originated from machine learning or deep learning algorithms, it has become very important to build a large, high-quality training database (DB). To this end, this paper proposes an automated framework that creates images using web crawling and virtual reality techniques, labels target objects, and generates a training DB for vision-based detection models. The framework contains three main processes: (1) image collection and labeling using web crawling; (2) image producing using a 3D modeling tool; and (3) foreground–background cross-oversampling. As a result, the framework constructed a training DB composed of 99,800 images in 42 min. The deep learning model was trained by the generated DB and showed macro F1-scores of up to 96.99%. These results imply that the framework successfully constructed a high-quality training DB within a short period of time. The findings can contribute to reducing time and effort in developing vision-based monitoring technologies. • Minimized the number of target site images required to develop a training DB. • Developed a training image DB using web crawling and a 3D modeling tool. • Constructed a training DB of 99,800 images in 42 min automatically. • Successful performance with macro F1-scores of up to 96.99%. [ABSTRACT FROM AUTHOR]

As most of the state-of-the-art technologies for vision-based monitoring were originated from machine learning or deep learning algorithms, it is crucial to build a large and rich training image database (DB). For this purpose, this paper proposes an automated framework that builds a large, high-quality training DB for construction site monitoring. The framework consists of three main processes: (1) automated construction image collection using web crawling, (2) automated image labeling using an image segmentation model, and (3) fully randomized foreground-background cross-oversampling. Using the developed framework, it was possible to automatically construct a training DB, composed of 5864 images, for the detection of construction objects in 53.5 min. The deep learning model trained by the DB successfully detected construction resources with an average precision of 92.71% and a recall rate of 88.14%. The findings of this study can reduce the time and effort required to develop vision-based site monitoring technologies. • Automatically developed training image DB using web crawling for construction monitoring • Automatically labeled web crawled images using a semantic segmentation model • Foreground-background cross-oversampling to synthesize photorealistic site images • Validated acceptable performance with 92.7% precision and 88.1% recall rate [ABSTRACT FROM AUTHOR]

In order to achieve database-free (DB-free) vision-based monitoring on construction sites, this paper proposes a deep active learning approach that automatically evaluates the uncertainty of unlabeled training data, selects the most meaningful-to-learn instances, and eventually trains a deep learning model with the selected data. The proposed approach thus involves three sequential processes: (1) uncertainty evaluation of unlabeled data, (2) training data sampling and user-interactive labeling, and (3) model design and training. Two experiments were performed to validate the proposed method and confirm the positive effects of active learning: one experiment with active learning and the other without active learning (i.e., with random learning). In the experiments, the research team used a total of 17,000 images collected from actual construction sites. To achieve 80% mean Average Precision (mAP) for construction object detection, the random learning method required 720 training images, whereas only 180 images were sufficient when exploiting active learning. Moreover, the active learning could build a deep learning model with the mAP of 93.0%, while that of the random learning approach was limited to 89.1%. These results demonstrate the potential of the proposed method and highlight the considerable positive impacts of uncertainty-based data sampling on the model's performance. This research can improve the practicality of vision-based monitoring on construction sites, and the findings of this study can provide valuable insights and new research directions for construction researchers. • Deep active learning selects most meaningful-to-learn training images. • The approach achieved over 80% mean Average Precision (mAP) only with 180 images. • It eventually resulted in 93% mAP for construction object detection. • Human effort can be reduced and more powerful object detectors can be trained. • This is the first attempt to apply deep active learning in the construction domain. [ABSTRACT FROM AUTHOR]