COMPARISON OF PHOTOGRAMMETRY TOOLS CONSIDERING REBAR PROGRESS RECOGNITION

Construction progress monitoring is an important process throughout the project timeline towards its successful completion. Among imaging techniques, photogrammetry is considered as economical and effective method. However, few studies can be found on construction progress monitoring via photogrammetry; thus, not much guideline is available for this domain. This study evaluated the photogrammetry tools for the progress assessment of the rebar grid framework. Photogrammetry tools were evaluated and analysed following defined criteria, and Agisoft Metashape, and 3DF Zephyr were identified as better options. This study aims to provide a guideline to construction industry professionals and stakeholders towards the adoption of photogrammetric progress assessment for construction activities.


INTRODUCTION
Construction progress monitoring is an essential and continuous activity, declared as a key factor towards project success (Pazhoohesh and Zhang, 2015). The evolution of Industry 4.0 (I4.0) and digital technologies have changed the attitude of construction industry stakeholders towards the adoption of advanced practices (Manzoor et al., 2021). Researchers have been working in digitising the process of construction project progress monitoring via data-acquisition technologies, as it helps to enhance accuracy by reducing human errors and the required effort (Mahami et al., 2019). Laser scanning, photogrammetry, and videogrammetry are renowned imaging techniques adopted for point cloud reconstruction (Rahimian et al., 2020). In comparison to laser scanning, photogrammetry stands as a major competitor with advantages, such as photogrammetry process is economical compared to laser scanning, digital images can be taken from any device, point cloud contains colour information, point clouds can be densified, frames can be intercepted from video streams for point cloud generation, etc. (Zhu et al., 2016;García-Gago et al., 2014).
In the domain of construction management, various studies have been performed for progress assessment of building components, such as RC elements (slab, columns, beams and walls), steel structures, shoring, formwork, masonry brickwork, tiles, etc. (Omar et al., 2018;Turkan et al., 2010); however, few studies have been focused on rebar (Alaloul et al., 2021). Rebar is the main reinforced concrete element, and its inspection is considered a rigorous and timely process, as rebar inspection requires close observation and experienced inspectors are preferred (Wang et al., 2017). Researchers have mostly adopted laser scanning for rebar detection, and very few studies have performed photogrammetry-based 3D point cloud reconstruction of rebar for the purpose of progress monitoring (Alaloul et al., 2021). Based on internet sources and literature review, more * Corresponding author than 37 photogrammetry-based software and tools are available. The aforementioned finding shows that there are plenty of options available for the application of photogrammetry, and there is a need to assess the better available tools considering the environment of the construction sector. The construction industry is enhancing towards digitisation, and to make this working theme successful, there is a need to provide less costly solutions (Qureshi et al., 2020). Likewise, in the domain of automated construction projects progress monitoring via digital data acquisition technologies, photogrammetry is a less costly solution with practical outcomes (Faltỳnová et al., 2016).
Various studies have been performed on the comparison of photogrammetry tools. However, each study compared the different combinations of tools considering varying target objects, i.e., most of the studies covered buildings, historical monuments Verykokou and Ioannidis, 2018;Murtiyoso et al., 2018), and aerial views (Alidoost and Arefi, 2017;Eltner and Schneider, 2015), whereas some also considered human figures, busts, fabric, etc. Wang et al., 2020). However, very few such comparison studies have been performed considering construction project elements for progress assessment (Pena-Villasenin et al., 2020). In light of the above discussion, this study is aimed to achieve the most suitable options among available photogrammetry tools by evaluating the point clouds considering the construction project monitoring theme. Furthermore, the photogrammetric testing and simulations have been performed by considering the rebar as a test subject, as rebar are distinct construction elements. Therefore, this study aims to improve the confidence of industry professionals in the adoption of photogrammetry tools for various construction processes in place of expensive detection technologies.

METHODOLOGY
The methodology was devised to evaluate the best options among available photogrammetry tools. Therefore, the criteria were defined for selecting the right tools based on three considerations, i.e., the tool should be offering close-range photogrammetry, most adopted by the research community (via literature review), and ease of availability of the tool via the internet. Figure 1 illustrates the strategy adopted to meet the study objective, where the overall methodology for selection and testing has been divided into three stages as follows: 1. Literature review and expert opinion 2. Photogrammetric model generation 3. Metadata and numerical analyses

Literature review and expert opinion
A literature review was performed to identify the most adopted photogrammetry tools by the research community. Articles were explored on Web of Science (WoS), and Scopus, for the last five years, considering specific keywords combinations, i.e., (photogrammetry OR point cloud) AND (image OR photo) AND (software OR tool OR technique). Following this, the most relevant articles were collected for review. Moreover, professional advice was also collected, and the expert opinions were taken from the website "ResearchGate" (https://researchgate.net/). The past literature and internet sources identified 37 photogrammetric open-sourced and paid software. Based on the defined criteria, nine tools were shortlisted for this study, which include VisualSFM, Meshroom, COLMAP, 3DF Zephyr, Regard 3D, RealityCapture, Autodesk ReCap Pro, Agisoft Metashape, PhotoModeler.

Photogrammetric model generation
In the second stage, 3D point cloud models were generated from the selected photogrammetry tools for testing purposes. A sample images dataset was developed for the rebar grid framework consisting of 50 images, as shown in Figure 2. The rebar grid was assembled, covering an area of 2.74m×2.74m (7.5 sq.m), with 16 rebars and each steel bar of 2.74m in length. The point cloud models were generated from the selected photogrammetry tools by following the developer guidelines and keeping high/ extreme software/ tool settings. The specifications of the camera and workstation used for data collection, simulation and cloud computation have been shown in Table 2.

Items Specification & Details Camera
Samsung SM-A225F Work Station Dell Precision 3630 Tower Intel Xeon CPU 64 GB RAM NVIDIA GeForce RTX 2060 Table 2. Camera and workstation specifications.

Metadata and numerical analyses
In the third stage, the attained point cloud models were evaluated and compared for information against five parameters, i.e., computational time, the number of dense points, mesh formation, percentage (%) completion of model elements (rebar), and % noise.
2.3.1 Time, dense points, and mesh The first three parameters were inspected via visual inspection and CloudCompare (https://danielgm.net/cc/). The computational time was noted for each tool separately until the complete photogrammetry process was achieved. However, the models were imported in CloudCompare and analysed to calculate the number of dense points cloud and mesh formation.

Percentage (%) completion of rebar model
The completion % of generated rebar model was evaluated via performing numerical analysis. The generated models were imported to CloudCompare and scaled up to the ground truth dimensions (GTD). In each model, generated rebars were measured for lengths considering all the 16 rebars individually. The GTD length of each rebar in the dataset was 2.74m±0.01m, and collectively 43.48m±0.16m running length for the all 16 rebars in the grid framework. To attain the percentage completion of generated point cloud model, the overall attained measured rebars length of each generated point cloud model was compared to GTD of the rebar dataset. Equation 1 was implemented to achieve the % completion of each model.
where %C = % completion of rebar Lc = calculated length of rebars LGT D = GTD of rebars 2.3.3 Percentage (%) noise To evaluate % noise, each scaledup model was cropped for 3.04m×3.04m (±0.01m), i.e., approximately 9.2 sq.m area around the rebar grid framework using CloudCompare. The overall number of point clouds were noted, and regions with noise were identified. Using Cloud-Compare, the noise was removed for each model separately, and the number of points cloud was noted again. Thus, % noise for each model was calculated by evaluating the difference between two readings by using Equation 2.
where %N = % noise Ni = Number of point cloud in initial model Nc = Number point cloud in cleaned model

RESULTS AND DISCUSSION
The 3D point cloud models have been generated for the same images dataset, following the guidelines given by the developers and considering high/ extreme software settings for the best as well as detailed outcomes. The evaluation of selected photogrammetry tools was performed based on metadata, % model completion, and % noise. Table 3 illustrates the generated point cloud models from the selected photogrammetry tools.
It can be seen that 3D point cloud generation performance varies between selected tools. VisualSFM, Meshroom, COLMAP, and Regard 3D generated partial point cloud models. However, 3DF Zephyr, PhotoModeler, Agisoft Metashape, and Reality-Capture gave better outcomes. The model attained via Autodesk Recap Pro was average, as incomplete element generation was observed for rebars. Table 4 shows the metadata analysis of the generated point clouds.
Metadata on generated point cloud models were collected on computational time, generated number of the point cloud, and mesh development. The computation time represents the time taken by each photogrammetry tool for the generation of the 3D point cloud model by following all the available processes by that tool. The number of point cloud generated in each model were calculated for the cropped area, i.e., 9.2 sq.m, in which 16×2.74m±0.01m rebars were placed. Likewise, the mesh generation was observed and noted for each model to assess the extend of photogrammetry pipeline provided by the each developer of the selected nine photogrammetry tools.
It can be observed that all the tested photogrammetry tools offered mesh formation except VisualSFM, which only performed dense point cloud generation. Moreover, the maximum number of point cloud was attained by Agisoft Metashape (10,138,227), followed by RealityCapture (4,894,717). Likewise, the computational time of the COLMAP was the highest (35 minutes), followed by Autodesk Recap Pro (27 minutes). However, minimum computational time was taken by VisualSFM (3 minutes), but it doesn't offer mesh generation. Whereas, RealityCapture (7 minutes), Meshroom (9 minutes), and Photomodeler (9 minutes) offered the lowest computational time with mesh generation.  Table 3. Generated 3D point cloud models. completion rate. 3DF Zephyr also attained a good outcome with a 97% completion rate. However, most of the photogrammetry tools have accomplished % completion between 89% to 60%. Where lowest model completion was achieved by COLMAP (33%), and Regards3D (52%). Table 6 illustrates the % noise formation in generated point cloud models. Point clouds obtained with imaging reconstruction techniques are often corrupted with a significant amount of outliers and noise (Rakotosaona et al., 2020). In this study, the scaled-up point cloud models for each tool were analysed for noise using CloudCompare, and noisy areas were removed. As already discussed, models were overviewed for the area of 9.2 sq.m, which was cropped from the originally generated point cloud model covering 16×2.74m±0.01m region. For each tool, noise-cleaned model was compared with uncleaned model for % noise assessment by following Equation 2. It can be noted that most of the photogrammetry tools offered less noise generated models. Minimum noise were observed in Agisoft Metashape (0.003%), RealityCapture (0.005%), and 3DF Zephyr (0.042%). However, high noise were reported for COLMAP (20.465%), and Autodesk Recap Pro (7.366%).

DISCUSSION
This study has been performed to evaluate better performing photogrammetry tools considering construction environment and materials for purpose of progress monitoring. From the literature review and expert opinion, nine photogrammetry tools were selected for testing, i.e., VisualSFM, Meshroom, COLMAP, 3DF Zephyr, Regard 3D, RealityCapture, Autodesk ReCap Pro, Agisoft Metashape, and PhotoModeler. Rebar was selected as a testing material, and the rebar grid framework was assembled on a 9.2 sq.m area with 16 rebars and 2.74m ±0.01m in length. The image-based dataset was prepared using a Samsung SM-A225F camera, and 50 images were captured. Photogrammetric models were generated by each selected tool for the same dataset. To gain the best and most detailed outcome models, high/ extreme model generation settings were adopted depending on each tool.
Attained models were analysed and evaluated against computation time, number of the point cloud, mesh formation, % model completion, and % noise. It was observed that for computational time VisualSFM (3 minutes), RealityCapture (7 minutes Therefore, in light of this performed study and past literature, it can be determined that every photogrammetry software has some benefits over others, depending on the type of the targeted object, nature of the job, and site conditions. Hence, it cannot be declared that some particular software is the best; software should be selected or chosen depending upon the job description, and guidance may be taken from the available literature.

CONCLUSION
This study was devised to evaluate the best available photogrammetry tools for close-range progress assessment of the rebar grid. Nine photogrammetry tools were selected for testing, and 3D point cloud models were generated. Attained point cloud models were analysed and evaluated against computation time, number of point cloud, mesh generation, % model completion, % noise, via performing visual inspection and numerical analyses. The comparison revealed that Agisoft Metashape, and 3DF Zephyr are better options for close-range outdoor testing adopted for construction activity progress monitoring/ assessment. Moreover, there is no hard and fast rule that a higher number of point cloud assures excellent models. This study aims to motivate the construction sector towards the adoption of photogrammetry as data-acquisition technology as a part of progress monitoring process.