It is important to acquire geometric data for infrastruc- ture management (Concrete Construction 2005). The present method for collecting geometric structure data for inspection of the structure relies on manual data collection methods, which require time and could be prone to error (Rens et al. 1997). Inspectors that are close to the hard-to-reach parts of the bridge (Kuang et al. 2009) may face safety issues, making the data collection process cumbersome and time-consuming (Sanford et al. 1999). Furthermore, the instrumentation can take measurements only in discrete and sparsely populated areas. In addition, most of the collected data is recorded in the form of paper-based or written files without any support, so the process of interpreting and retrieving geometric infor- mation requires a large amount of data transmission work to be done manually by a human (Estes and Frangopol 2003). Therefore, it would be advantageous to record the final location of reinforcement as a future document to be acces- sible if needed during the service life. This would become a permanent record of the field. When a forensic investigation is needed, a potentially effective tool for measuring concrete cover after construction could be ground penetrating radar (GPR), which uses penetrating wave. Researchers and com- panies use GPR to measure the actual cover, spacing, and bar diameter (Tosti and Ferrante 2020). Not only are GPR and other means expensive and only justified in specific cases, but also it is not always possible to use them due to traffic or other functionality restraints (De Souza et al. 2004). The cost of not knowing the actual quality of the structure will translate into negative consequences both in cost and safety (BrÜhwiler 2009). Hence, an automated way of quantifying reinforcement in structures such as bridge decks using innovative technologies coupled with artificial intelligence (advanced processing and decision-making concepts) is agreed to be useful for construc- tion quality (Akinci et al. 2006). The proposed method can inform the inspector of the reinforcement during construction and become a permanent record of the quality that can be accessed in the future. Methodology This section describes the technical capabilities of the Azure Kinect camera in relation to data collection, as well as examples of RGBD applications used in the past. Subsequently, the capability of using RGBD cameras for creating 3D point cloud data for rebar inspection is explained. RGBD Camera The RGBD camera is a color-depth camera that has the poten- tial of reconstructing the 3D shape of an object. There are different types of RGBD cameras that can be used for this study. In 2020, Microsoft launched the third generation of Kinect, called Azure Kinect, with time of flight technology. This sensor is offered at a relatively low price (currently less than US$1000) but offers practical performance and depth accuracy. 3D scanning and measurement capabilities bring low-cost Kinect sensors to a wider range of applications such as gesture recognition (Ren et al. 2013 Patsadu et al. 2012 Biswas and Basu 2011), AR (Ren et al. 2013 Ma et al. 2013 Vera et al. 2011), robot navigation (Correa et al. 2012), and the construction of the building (El-laithy et al. 2012). Figure 2 shows the RGBD camera that is used in this study. Figure 2 shows Microsoft’s newest sensor technology that is incorporated into the Kinect hardware, which is a single USB- connected device. The camera dimensions and weight are 103 × 39 × 126 mm and 440 g, respectively. This RGBD camera ME | RGBDCAMERAS Figure 1. Rebar mat inspection on a big bridge deck construction site. Figure 2. Azure Kinect camera: (a) front view (b) lateral view. 103 mm 39 mm 126 mm 48 M AT E R I A L S E V A L U AT I O N • J A N U A R Y 2 0 2 3 2301 ME Jan New.indd 48 12/20/22 8:15 AM COURTESY: XINXING YUAN

includes a 1-megapixel time of flight depth sensor that provides different modes of scanning. Different fields of view (FOV), res- olution, and operating ranges can be selected to optimize the application. Table 1 summarizes the modes that this camera can support. The FOV, resolution, and functioning range are the features of interest for the automatic geometric measure- ment proposed in this paper. As can be seen in Table 1, as the functioning range increases, the resolution will decrease. However, as long as the camera records a video in the functioning range mentioned in Table 1, the picture would be of good quality and the camera can detect the depth. The RGBD camera also has a 12-megapixel color camera with a maximum resolution of 3840 × 2160 that can be used with 0, 5, 15, and 30 FPS frame rates. Additionally, an inertial measurement unit (IMU) is embedded in the camera that consists of an accelerometer and a gyroscope that can be utilized for sensor orientation and spatial tracking. The mentioned camera contains a developer kit that uses artificial intelligence (AI) sensors to develop computer vision and speech model applications. This developer includes a cross-platform that can read data from the RGBD camera. The software development kit enables users to develop their own applications. Furthermore, the recorded data can be analyzed in other software such as MATLAB. In the next sections, the authors propose to use MATLAB to analyze the 3D point cloud data that has been generated by the RGBD camera. RGBD Camera for Rebar Inspection Instead of visual inspection as the conventional approach, the RGBD camera can be used to scan the entire rebar of the structure. In this regard, the authors present an algorithm to measure the diameter of the rebar in RC structures. So, instead of the inspector doing all the measurements with a tape or caliper, the RGBD camera can be an alternative for visual inspection of the rebar. This can also be accomplished at a distance, which can increase safety during the inspec- tion, as can be seen in Figure 3. The process of data collection starts by connecting the camera to a computer, then a video is recorded by walking around the object and scanning it. The recorded video is imported to software such as MATLAB or CloudCompare to generate the 3D point cloud data. At the construction site, the camera needs to be plugged into a power bank. The inspecting crew can estimate the size/area to inspect based on the power bank capacity. The authors successfully tested using camera using a regular portable power bank at a construction site. The structure consisted of three 16 ft (4.9 m) tall walls with an approximate length of 50 ft (15.2 m). Future research efforts include the testing of longer reinforced cages in civil engineering infrastructure under construction (such as bridge decks, industrial slabs, and parking garages). Further analysis can be applied to the data to get the required information from the rebar before pouring the concrete for example, spacing, diameter, and elevation (cover). However, to estimate the cover from the generated 3D point cloud data before the construction, the space between the outer edge of the rebar and the ground or mold should be cal- culated, which can be done at any point where the inspector is willing to measure the spacing, so the camera and the algo- rithm can detect the cover even if it is uneven. The algorithm measures the diameter of the rebar using a Pratt fitting circle algorithm (Chernov 2022). To have an accurate fitting implementation, all points in single rebar that are perpendicular to its orientation should be projected to a plane, and then the algorithm should be executed. Experimental Setup The experiments presented in the next section aim to evaluate the potential application of an RGBD sensor approach for rebar size placement inspection. Rebar Selection Rebar diameter increases by 1/8 in. (3.175 mm) in the United States. To get the nominal diameter in inches, the bar size should multiply by 1/8 (i.e., #4 rebar diameter is 0.5 in. [4 × 0.125 in.]). The purpose of this experiment is to recognize rebar #3 between four rebars of a different size (#4). Figure 4 shows a comparison between a #3 and #4 bar. #3 and #4 bars are 3/8 in. or 0.375 in. (9.525 mm) and 4/8 in., or 0.5 in. (12.7 mm) diameter, respectively. Although the diameter dif- ference is 33.33% between a #3 and #4, the site conditions and the natural distance between the human and the case makes it a difficult task to catch a change in diameter of only 1/8 in. (3.175 mm). Figure 5 shows the different views in a sequence of increasing distances. Researchers chose a longer #3 rebar to illustrate better and track the differences between the two sizes TA B L E 1 Comparison of camera modes Mode Field of view Resolution Functioning range (m) NFOV unbinned 75 × 65 640 × 576 0.5–3.86 NFOV 2 × 2 binned 75 × 65 320 × 288 0.5–5.46 WFOV unbinned 120 × 120 1024 × 1024 0.25–2.21 WFOV 2 × 2 binned 120 × 120 512 × 512 0.25–2.88 1.5 - 2.5 m Figure 3. Inspector with RGBD camera in hand. J A N U A R Y 2 0 2 3 • M AT E R I A L S E V A L U AT I O N 49 2301 ME Jan New.indd 49 12/20/22 8:15 AM