VDR-AM: View-dependent representation of articulated models

Abstract

We present a novel, View-Dependent Representation of Articulated Models (VDR-AM), and show its main benefits in the context of view-dependent rendering integrated with occlusion culling for large-scale crowd scenes. In order to provide varying resolutions on each animated, articulated model, we propose to use a cluster hierarchy in the VDR-AM for an articulated model. The cluster hierarchy serves as a dual representation for both view-dependent rendering and occlusion culling. For a high-performance view-dependent rendering and occlusion culling, we construct each cluster of the cluster hierarchy to contain a spatially coherent portion of the mesh that also has similar simplification errors. To achieve our goal, we present an error-aware clustering method for articulated models. We also identify a subset of animation poses that well represents the original pose data and perform the well-known quadrics-based simplification to efficiently compute our representation, while achieving a high quality simplification. At runtime, we choose a LOD cut from the cluster hierarchy given a user specified error bound in the screen space and render all the visible clusters in the LOD cut. We implement our method in GPU and achieve interactive performance (e.g., 40 frames per second) for large-scale crowd scenes that consist up to thousands of articulated models and 242 M triangles, without noticeable visual artifacts.