Automation in Construction, 181, 106589
Environment|구충완
Automated reliability-based multi-camera strategy for excavator tracking under dynamic occlusion using deep learning with instance segmentation
2026International Journals
Research AreaEnvironment
Professor구충완
AuthorsAhn, S., Seo, S., Shin, Y., and Koo, C. (Corresponding Author)
Publication Year2026
Journal (Volume, Issue, Pages)Automation in Construction, 181, 106589
AbstractOcclusion in vision-based excavator tracking is a major issue in dynamic construction environments where frequent obstructions significantly degrade object tracking performance. To overcome this, this paper proposes an automated reliability-based multi-camera strategy for robust excavator tracking under simultaneous occlusion in dynamic construction environments by integrating deep learning-based segmentation. The paper was conducted in three phases: (i) performance validation using authentic and synthetic videos; (ii) reliability modeling with occlusion ratio and viewpoint analysis; and (iii) strategy validation in real-world scenarios. The key findings are as follows. The reliability of support vector classifier reached a weighted F1-score of 0.904 in classifying reliable tracking zones. The mask-based method achieved a multi-object tracking accuracy of 84.41 % under real-world scenarios for empirical validation. These results demonstrate the effectiveness of the proposed approach in mitigating occlusion-induced degradation, laying the foundation for automation in carbon emission and productivity analysis on construction sites.
Excavator TrackingOcclusionVision-Based MonitoringDeep Learning Instance SegmentationTracking ReliabilityMulti-Camera Strategy