The automation of design processes in the Building Information Modeling (BIM) environment is becoming increasingly feasible due to artificial intelligence tools, including Convolutional Neural Networks (CNNs) and alternative object recognition methods. Their application in analyzing architectural drawings, such as floor plans and building sections, can significantly accelerate the identification of structural elements (e.g., walls, windows, doors). A dataset of 77 raster images of floor plans was compiled, labeled, and used to train object detection models (Roboflow 3.0, YOLOv11, and YOLOv12). In this task, three classes of objects are defined: walls, doors and windows. The performance evaluation, based on Mean Average Precision (mAP), Precision, and Recall, demonstrated the models’ capability to detect building components. Furthermore, the extracted object coordinates from detected text labels are used in Revit to generate floor models using Python scripts. The results indicate the potential application of these methods in automating 3D model generation, aiding tasks such as inventory assessments, planning renovations, and energy simulations.