Monitoring concrete pouring progress using knowledge graph-enhanced computer vision

Abstract

Accurate progress measurement in concrete pouring is essential to prevent project delays and material waste. This paper introduces a knowledge graph (KG)-enhanced computer vision (CV) method to improve the accuracy and generalizability of traditional methods used in concrete pouring monitoring, which often struggle to integrate contextual data. By combining object detection and extracting information from BIM models, the method creates a KG to represent spatial–temporal relationships among building components and pouring-related resources (e.g., concrete mixer, bucket, hose, workers). Rule-based interpretation and Graph Neural Networks (GNN) classify pouring states and cycles, achieving 80.3% accuracy with the rule-based system and 89.2% with GNN in conducted experiments on ten samples across two construction sites. These findings demonstrate that the KG-enhanced CV method provides generalizability, offering data-driven support for site managers to efficiently coordinate processes. This approach lays the foundation for detailed process-oriented digital twinning of construction projects, enabling deeper insights and better decision-making.