Generalized vision-based framework for construction productivity analysis using a standard classification system

Abstract

Enhancing construction productivity is paramount, and numerous researchers have utilized computer vision techniques to perform productivity analysis. However, previous approaches are often limited in their scalability and practical implementation as they can only be applied to specific construction works. Additionally, comprehensive training image datasets featuring varied scene compositions are essential for developing high-performance models. To address limitations, this study proposes a vision-based framework that can be applied to various types of work, covering the end-to-end process from constructing training image datasets to conducting productivity analysis. The framework consists of four main processes: (i) construction baseline dataset development, (ii) field optimization, (iii) standard classification system establishment, and (iv) productivity analysis. The experimental results showed satisfactory performance, with an average accuracy of 86.2% for activity analysis and 85.3% for productivity analysis. It suggests its potential application to common construction work types and enables practitioners to enhance productivity analysis in construction projects.