Rapid Human-Assisted Creation of Bounding Models for Obstacle Avoidance in Construction

Abstract

State-of-the-art construction equipment control technology creates the opportunity to implement automated and semi-automated object avoidance for improved safety and efficiency during operation; however, methods for constructing models of local objects or volumes in real-time are required. A practical, interactive method for doing so is described here. The method: (1) exploits a human operator's ability to quickly recognize significant objects or clusters of objects in a scene, (2) exploits the operator's ability to acquire sparse range point clouds of the objects quickly, and then (3) renders models, such as planes, boxes, and generalized convex hulls, to be displayed graphically as visual feedback during equipment operation and/or for making proximity calculations in an obstacle detection system. Experimental results indicate that bounding models can be created rapidly and with sufficient accuracy for obstacle avoidance with the aid of human intelligence and that humanassisted modeling can be beneficial for real-time construction equipment control.

Human-Assisted 3D Spatial Modeling of Construction Sites Using Sparse Range-Point Clouds

“…Moreover, humans are adept at recognizing objects, especially in cluttered scenes such as construction sites [7], so by incorporating human perception into the overall modeling enterprise, an objective-driven, sparse point cloud approach has the potential to reduce not only the data-acquisition time but also the need for computationally intensive and/or expensive processing [3,12].…”

“…(1) such volumes are inherently conservative, because of their convex nature, (2) any number of points can be picked, anywhere, and (3) the resulting hull is bounded by planar faces, thereby enabling rapid computation of distances [13].…”

“…Safety-driven applications can easily be envisaged, such as a backhoe being operated next to a busy city street; in this scenario, a wall would act as a partition between the operator's workspace and the forbidden zone of the street. A detailed description of the workspace-partitioning algorithm is given in [12]. An example of the workspace-partioning algorithm is shown in figure 4.…”

Autonomous Modeling of Pipes within Point Clouds

“…Building on foundational research in robotics and machine vision, research on automated as-built generation goes back well over ten years (Kwon et al 2003, McLaughlin et al 2004, Rabbani et al 2005, Brilakis et al 2010, Ahmed et al 2012. Some of the knowledge thereby created has influenced or been adopted in practice.…”

Primitives Merging For Rapid 3D Modeling

“…Its main characteristics are the use of sparse point clouds, the utilization of the operator's scene recognition skills, and the division of the environment into target and peripheral objects. While peripheral objects can be roughly modeled by using convex hulls and bounding boxes (McLaughlin, 2003), target objects must be more accurately modeled. For target objects, geometric primitives are used: cuboids, cylinders, spheres, planes, and lines (Feddema, 1997;Kwon, 2003).…”