DNN-based scene understanding is the dominant approach for accurate 2D–3D pose estimation in the construction industry. This technology improves safety and productivity by providing unparalleled insights into the dynamics of construction equipment, and their interaction on job sites. However, the full potential of these supervised learning methods remains untapped due to a critical shortage of construction equipment-specific training data. In-lab data generation from miniature-scale construction equipment has a great potential in addressing the shortage of image data. However, data quality is greatly influenced by camera-related hyperparameters, including, the number and placement of cameras, and their viewpoint. This paper proposes a pipeline as a vision-aided decision-making support tool used in the design and pre-construction phase of in-lab data generation. The pipeline first optimizes camera configurations to meet in-lab constraints. Subsequently, it simulates the camera setups, the target subject (in this article, we used a miniature-scale excavator as a case study), and the detailed lab environment within Blender, enabling the rendering of images that showcase the cameras’ viewpoints for dataset generation. This pipeline accommodates user-defined constraints, ensuring its adaptability, while offering significant potential for generating high-quality images for DNN training across diverse scenario.