Human-assisted fitting and matching of objects to sparse point clouds for rapid workspace modeling in construction automation
Most current site modeling methods in the construction industry use large, expensive
laser-scanning systems that produce dense range point clouds of a scene from
different perspectives. While useful for many purposes, this approach is not feasible
for real-time modeling which would enable automated obstacle avoidance and
improved semi-automated equipment control. The dynamic nature of the construction
environment requires that a real-time local area modeling system be capable of
handling a rapidly changing and uncertain work environment. In practice, large,
simple, and reasonably accurate object primitives are adequate feedback to an
operator who is attempting to place target materials in the midst of obstacles with an
This dissertation presents human-assisted rapid environmental modeling methods for construction. These methods exploit the human operator’s ability to quickly evaluate and associate objects in a scene and only require a limited number of scanned range data (sparse point clouds). These sparse clouds are then used to create geometric primitives for visualization and modeling purposes Five fitting and matching methods were developed that make use of sparse (fewer than fifty) point clouds per object: (1) workspace partitioning (planar least squares fit), (2) cuboids, (3) solid cylinders, (4) hollow cylinders and (5) spheres.
Experiments have been conducted to determine how rapidly and accurately fitting and matching methods can model the objects in a scene, by comparing location, orientation, and size of objects between modeled and actual objects. Method development and revisions were also based on lab experiments. The experimental results indicated that these models can be created rapidly and with sufficient accuracy for automated obstacle avoidance and equipment control functions for safety applications.