Rebar counting detection using object detection based on deep learning

A high-precision DJI Phantom 4 Pro drone was commissioned at five unique construction sites in South Korea during peak productivity hours. Construction supervisors manually controlled the drone, thus ensuring the capture of clearly visible images of the rebars. Representative samples from the original dataset are displayed in Figure 3. The drone's path was methodically guided in a vertical trajectory above each column, positioning it directly above the rebars at an estimated altitude of 1 to 2 meters. At each position, still images were captured with the columns nearly centred in each frame. To underscore the pragmatic viability of the proposed method, rebar images were sourced under authentic operational conditions, thus encapsulating the complexities and challenges that typify bustling construction sites. This dataset contained a diverse array of variations in factors such as illumination, scale, and perspective. Moreover, other construction equipment such as scaffolding, timber and moulding were also observed in the images. In total, 728 images of rebars with a resolution of 1,500 × 900 pixels were captured.本研究于韩国五处各具特色的建筑工地的作业高峰时段，启用了一台高精度大疆精灵4 Pro（DJI Phantom 4 Pro）无人机开展数据采集工作。由施工监理手动操控该无人机，从而确保能够清晰捕捉到钢筋（rebars）的图像。原始数据集的代表性样本示于图3中。无人机沿垂直轨迹有条不紊地在每根柱体上方飞行，将其精准悬停于钢筋正上方，预估飞行高度为1至2米。在每个悬停点位，均采集静态图像，且每帧画面中柱体基本处于画面中心位置。为凸显本文所提方法的实用可行性，本数据集的钢筋图像均采集自真实作业场景，涵盖了繁忙建筑工地典型存在的各类复杂情况与实际挑战。该数据集涵盖了光照、尺度、视角等多种因素下的多样化样本。此外，图像中还包含了脚手架、木料、建筑模板等其他施工设备与物料。最终共采集得到728张钢筋图像，图像分辨率为1500×900像素。