Enhanced MEP construction progress tracking using panoramic mobile positioning and optimized pipeline segmentation

Efficient progress management is pivotal for the successful delivery of MEP construction projects. While non-intrusive methods, such as image recognition, hold promise in enhancing progress management efficiency, several challenges in MEP scenarios, particularly irregular and sparse features, may constrain the progress recognition accuracy. This paper proposes a framework for automated MEP construction progress tracking, which integrates panoramic mobile positioning, instance segmentation, and Unreal Engine to compare virtual (as-planned) and real (as-built) MEP construction scenes. Moreover, quantitative progress can be estimated by identifying and mapping unfinished construction components. Remarkably, dynamic snake convolution is introduced to enhance the component segmentation accuracy, specifically for irregular and sparse features. The framework is successfully applied to an MEP construction project in China, achieving commendable component segmentation (mAP₅₀ = 85 %) and progress recognition accuracy. This paper provides theoretical references for vision-based MEP construction progress tracking, offering practical insights for intelligent construction inspection in continuous spaces.