Camera Marker Networks for Pose Estimation and Scene Understanding in Construction Automation and Robotics.

Abstract

The construction industry faces challenges that include high workplace injuries and fatalities, stagnant productivity, and skill shortage. Automation and Robotics in Construction (ARC) has been proposed in the literature as a potential solution that makes machinery easier to collaborate with, facilitates better decision-making, or enables autonomous behavior. However, there are two primary technical challenges in ARC: 1) unstructured and featureless environments; and 2) differences between the as-designed and the as-built. It is therefore impossible to directly replicate conventional automation methods adopted in industries such as manufacturing on construction sites. In particular, two fundamental problems, pose estimation and scene understanding, must be addressed to realize the full potential of ARC.

This dissertation proposes a pose estimation and scene understanding framework that addresses the identified research gaps by exploiting cameras, markers, and planar structures to mitigate the identified technical challenges. A fast plane extraction algorithm is developed for efficient modeling and understanding of built environments. A marker registration algorithm is designed for robust, accurate, cost-efficient, and rapidly reconfigurable pose estimation in unstructured and featureless environments. Camera marker networks are then established for unified and systematic design, estimation, and uncertainty analysis in larger scale applications.

The proposed algorithms' efficiency has been validated through comprehensive experiments. Specifically, the speed, accuracy and robustness of the fast plane extraction and the marker registration have been demonstrated to be superior to existing state-of-the-art algorithms. These algorithms have also been implemented in two groups of ARC applications to demonstrate the proposed framework's effectiveness, wherein the applications themselves have significant social and economic value. The first group is related to in-situ robotic machinery, including an autonomous manipulator for assembling digital architecture designs on construction sites to help improve productivity and quality; and an intelligent guidance and monitoring system for articulated machinery such as excavators to help improve safety. The second group emphasizes human-machine interaction to make ARC more effective, including a mobile Building Information Modeling and way-finding platform with discrete location recognition to increase indoor facility management efficiency; and a 3D scanning and modeling solution for rapid and cost-efficient dimension checking and concise as-built modeling.