Haptic rendering of highly detailed point-based virtual models

Abstract

In this work, we present a comprehensive approach to haptically render highly-detailed point clouds without first creating their corresponding polygonal mesh. We say that our approach is comprehensive because it addresses: kinesthetic and tactile rendering; static and dynamic models; collision detection and force response; force shading; deformation and stiffness; and friction. These features compromise the majority of the haptic interactions possible while using a 3-degrees-of-freedom haptic device, which is the target device for our algorithms. Furthermore, we look in this work at height fields and redefine them as a special case of point clouds for which we present a specialized haptic rendering approach that includes all the features already mentioned for our general purpose approach. Our work relies on redefinitions of what a point cloud's surface is; for the purposes of collision detection, we look at it as a collection of touching, if not slightly overlapping, axes-aligned bounding cubes; and for the purposes of force response and haptic effects rendering, we look at the surface as a neighborhood of points where each point knows its immediate neighbors. Our collision detection algorithms are novel, and our force response algorithms are loose adaptations to point clouds of standard constraint-based approaches. Our work is motivated by, and largely designed for, models that are the result of scanning real-life objects in 3D. These models are of great practical use in a large number of fields ranging from arts to manufacturing, and from entertainment to medicine. The scanning technology (laser or contact) is of no consequence to our work, but what is relevant is the high-density point cloud that typically results from 3D scans. When possible, we also make use of our a priori knowledge of the path the scanner takes to sample the real-life object in a novel approach to compile neighborhood information. Finally, this work will demonstrate, through experimental results, its effectiveness in conveying haptic information, its speed even when all haptic algorithms run in a single thread on a single processor, and its insensitivity to the size of the input point clouds; factors that make the case for our approach's adaptation in place of mesh reconstruction techniques.