Enhancing detailed haptic relief for real-time interaction

  1. Theoktisto Costa, Victor Simon
Dirigida por:
  1. Isabel Navazo Álvaro Director/a
  2. Marta Fairen González Director/a

Universidad de defensa: Universitat Politècnica de Catalunya (UPC)

Fecha de defensa: 25 de noviembre de 2015

Tribunal:
  1. Juan Carlos Torres Cantero Presidente
  2. Antoni Susín Sánchez Secretario/a
  3. Riccardo Scateni Vocal

Tipo: Tesis

Teseo: 413464 DIALNET lock_openTDX editor

Resumen

The present document exposes a different approach for haptic rendering, defined as the simulation of force interactions to reproduce the sensation of surface relief in dense models. Current research shows open issues in timely haptic interaction involving large meshes, with several problems affecting performance and fidelity, and without a dominant technique to treat these issues properly. Relying in pure geometric collisions when rendering highly dense mesh models (hundreds of thousands of triangles) sensibly degrades haptic rates due to the sheer number of collisions that must be tracked between the mesh's faces and a haptic probe. Several bottlenecks were identified in order to enhance haptic performance: software architecture and data structures, collision detection, and accurate rendering of surface relief. To account for overall software architecture and data structures, it was derived a complete component framework for transforming standalone VR applications into full-fledged multi-threaded Collaborative Virtual Reality Environments (CVREs), after characterizing existing implementations into a feature-rich superset. Enhancements include: a scalable arbitrated peer-to-peer topology for scene sharing; multi-threaded components for graphics rendering, user interaction and network communications; a collaborative user interface model for session handling; and interchangeable user roles with multi-camera perspectives, avatar awareness and shared annotations. We validate the framework by converting the existing ALICE VR Navigator into a complete CVRE, showing good performance in collaborative manipulation of complex models. To specifically address collision detection computation, we derive a conformal algebra treatment for collisions among points, segments, areas, and volumes, based on collision detection in conformal R{4,1} (5D) space, and implemented in GPU for faster parallel queries. Results show orders of magnitude time reductions in collisions computations, allowing interactive rates. Finally, the main core of the research is the haptic rendering of surface mesostructure in large meshes. Initially, a method for surface haptic rendering was proposed, using image-based Hybrid Rugosity Mesostructures (HRMs) of per-face heightfield displacements and normalmaps layered on top of a simpler mesh, adding greater surface detail than actually present. Haptic perception is achieved modulating the haptic probe's force response using the HRM coat. A usability testbed framework was built to measure experimental performance with a common set tests, meshes and HRMs. Trial results show the goodness of the proposed technique, rendering accurate 3D surface detail at high sampling rates. This local per-face method is extended into a fast global approach for haptic rendering, building a mesostructure-based atlas of depth/normal textures (HyRMA), computed out of surface differences of the same mesh object at two different resolutions: original and simplified. For each triangle in the simplified mesh, an irregular prism is considered defined by the triangle's vertices and their normals. This prism completely covers the original mesh relief over the triangle. Depth distances and surfaces normals within each prism are warped from object volume space to orthogonal tangent space, by means of a novel and fast method for computing barycentric coordinates at the prism, and storing normals and relief in a sorted atlas. Haptic rendering is effected by colliding the probe against the atlas, and effecting a modulated force response at the haptic probe. The method is validated numerically, statistically and perceptually in user testing controlled trials, achieving accurate haptic sensation of large meshes' fine features at interactive rendering rates, with some minute loss of mesostructure detail.