Integrated Aerial-Ground Intelligent Monitoring for Construction Sites Using BIM, LLM, and Computer Vision

Construction site monitoring requires both effective structural inspection and reliable worker safety tracking, yet these two tasks remain disconnected in practice. Drone-based inspections still depend on skilled operators who manually configure flight parameters, while worker identification across ground CCTVs and aerial drones is hindered by extreme viewpoint discrepancies and frequent occlusions from equipment and materials. This research develops an integrated aerial-ground monitoring framework addressing both challenges. For autonomous drone inspection, an LLM-based chatbot interprets natural language commands (e.g., "inspect 3rd floor cracks"), automatically infers implicit technical requirements such as GSD and camera angles using a BIM Knowledge Graph, and generates complete flight missions with optimized waypoints. For worker safety tracking, a graph-based message-passing re-identification module fuses complementary features from TransReID and BPBReID through dual-stage clustering, enabling persistent identity tracking across heterogeneous camera viewpoints. The two modules form a closed-loop system where drone imagery feeds into cross-viewpoint worker tracking, and identified worker locations inform targeted monitoring missions.