Journal of Construction Automation and Robotics ISSN:2800-0552(Print) 2951-116X(Online) 건설자동화·로보틱스 논문집

Construction productivity is an important index that enables to plan and manage construction project. Therefore, accurate productivity measurement is required, and productivity measurement has been made by collecting field information in various ways. Recently, many studies have attempted to measure productivity by applying a deep learning-based computer vision algorithm to identify the location or condition of construction equipment or workers from site video. However, most studies are limited to testing the performance and applicability of individual algorithms, and most of the studies that comprehensively analyze the productivity of construction work are still based on data such as daily construction report. In this aspect, we present a methodology to analyze the work productivity of structural work by using computer vision algorithm to comprehensively analyze the productivity of construction workers. The methodology presented in this study analyzes productivity in the steps of (1) classification of the type of work, (2) detection and tracking of the number of workers, and (3) action recognition of the construction worker. Lastly, the proposed methodology was tested on construction site video of concrete work for productivity analysis and validation purposes.