Development of Human Pose Analyzing Algorithms for the Determination of Construction Productivity in Real-Time

Abstract

To enhance the capability of rapid construction, an automated on-site productivity measurement system is developed. Employing the concepts of Computer Vision and Artificial Intelligence, the developed system wirelessly acquires a sequence of images of construction activities. The system first processes these images in real-time to generate human poses associated with the construction workers at a project site. The poses are first manually classified into three categories as effective work, ineffective work, and contributory work. Then, a built-in neural network trained on these classifications, determines the worker's status by comparing the in-coming images to the developed human poses. The labor productivity is determined from these comparison statistics. This system has been tested for accuracy on a bridge construction project. The analysis results were accurate as compared to the results of the manual method. This research project made several major contributions to the advancement in construction industry. First, it applied advanced image processing techniques for analyzing construction operations. Second, the results of this research project made possible the automatic determination of construction productivity in real-time. Thus, an instant feedback to the construction crew was possible. As a result, the capability of rapid construction was improved using the developed technology.