Automated Task-Based Labour Allocation Extraction from Scanned Tables to Estimate Productivity

Labour allocation tables are widely used in construction and manufacturing environments to record workforce assignments and tasks. However, these tables are often maintained as scanned documents with mixed printed and handwritten content, posing significant challenges for automated data extraction due to variations in handwriting, table layouts, and image quality. This paper presents a novel image processing workflow for automated task-based labour allocation extraction from scanned tables. By combining advanced image processing techniques with pre-trained vision-language models, the method accurately detects and interprets handwritten marks, textual entries, and table structures within scanned documents. Furthermore, the developed approach can run locally on a standard computer, ensuring data privacy and minimising cost. It has been applied to historical labour allocation tables from a real construction project. The system efficiently transforms scanned data into structured, machine-readable formats, enabling seamless downstream labour productivity estimation. Experimental results demonstrate robust performance across diverse document layouts, highlighting its potential to enhance digital workflows in workforce management and production monitoring.