Covers construction phases: foundation, main structure, exterior/enclosure (walls, doors & windows, roof), and interior finishing & MEP installation. Many construction steps look visually similar in images but correspond to different progress stages.
Traditional construction progress tracking relies on manual inspection and record-keeping, which has several limitations: Delayed updates: Hard to detect construction delays in real time. Low efficiency: Manual inspections are time-consuming for high-rise or large-scale projects. Subjective records: Human error and inconsistent standards. Poor traceability: Paper or Excel records are difficult to analyze historically or visualize.
Stage 1 – Coarse-grained recognition High-frequency, distinctive construction steps are identified via high-accuracy image matching to provide a rough progress estimate. Stage 2 – Fine-grained recognition vLLM fine-tunes the progress estimation for precise output, which is automatically uploaded to the project management platform.
Image & annotation storage → MinIO Vector storage & retrieval → Milvus + ResNet Progress optimization → vLLM