This study investigates the integration of deep learning-based computer vision for building information modelling (BIM) supported automated construction progress monitoring. A lightweight, explainable image-based system that lever-ages the You Only Look Once v8 Nano (YOLOv8n) object detection model trained on the publicly available site object detection dataset (SODA) is pro-posed. The framework detects construction-related objects in site images and infers corresponding project stages through a rule-based scoring mechanism. Experimental results indicate a competitive trade-off between detection accuracy (average precision: 76.3%, and recall: 65.2%) and inference efficiency (5.4 ms per image), enabling deployment on resource-constrained devices. Validation against 500 expert-labelled images achieved an accuracy of 67% in stage classification, with particularly strong performance in the foundation and super-structure phases. The rule-based inference contributes to interpretability and adaptability, though performance decreases in transitional stages and remains dependent on detection accuracy. The contributions of this work are threefold: it demonstrates the use of YOLOv8n for efficient inference in resource-constrained environments, introduces an interpretable rule-based mechanism for construction stage inference from detected object frequency and category relevance, and positions image-based inference within a BIM-supported monitoring context for structured construction progress assessment.