Image Acquisition Planning for Image-based 3D Reconstruction Using a Robotic Arm

As-built modeling using advanced visual sensing technologies (i.e., photogrammetry and laser scanning) provides an opportunity for a rapid assessment of construction performance, identifying deviations from 3D CAD/Building Information Modeling (BIM) models (i.e., as-planned model). For reliable decision-making, the accuracy and quality of the as-built model are critical. In particular, data collection techniques for image-based 3D reconstruction influence the quality of as-built modeling. In addition, manual data capture is time-consuming, error-prone, and infrequent, which impedes the assessment process. To overcome the challenges, this paper proposes a model-based planning system to automate image data collection for 3D reconstruction. The data collection system uses a camera-installed robotic arm. The locations of images to be captured are planned based on a given 3D CAD/BIM model using the camera parameters, distance away from the target object, and overlap ratio. The end-effector of the robotic arm, where a camera is installed, moves along the planned locations and captures images at each location. The complete image data set uses computer vision algorithms (i.e., Structure from Motion and multiple view stereo) to create the as-built model. The preliminary experiment produces an as-built model of the object successfully. Moreover, the quality of the as-built model can be improved by the presented method on the leverage of a camera-mounted robotic arm.