Construction work sampling is a well-established method for productivity measurement but remains heavily dependent on manual site observation. Recent advances in computer vision enable automated activity detection; however, directly interpreting frame-level detections as productivity indicators may violate work sampling principles. This study proposes an AI-driven framework that transforms vision-based activity detection outputs into work-sampling-consistent productivity measurements. The framework integrates frame-based tracking and snapshot-level aggregation to ensure consistent worker counting and unbiased man-hour estimation. Results demonstrate that the proposed approach enables automated generation of work sampling tables and productivity visualizations while preserving the theoretical foundation of classical work sampling.

Keywords:Work Sampling, Construction Productivity, Computer Vision.