Survey of Techniques for Rendering Real-Time Virtual Humans

AbstractReal-time rendering of photo-realistic humans is considerably outside the scope of current consumer level com-puter hardware. There are many techniques which attempt to bridge the gap between what is desired and what ispossible. This paper aims to give an overview of the techniques designed to alter the complexity of the model’sgeometry (Level of Detail), or replace it with a flat image (Visual Impostor) and also how to improve the lightingmodel for character (Lighting and Shadows). Recent years have shown a boom in the power and availability ofconsumer level programmable graphics processors, thus techniques that make use of these features are coming tothe forefront.Categories and Subject Descriptors (according to ACM CCS) : I.3.7 [Computer Graphics]: Three-DimensionalGraphics and Realism I.3.3 [Computer Graphics]: Picture/Image Generation1. Introduction1: The desire to render the world around us is common tomany applications. As part of this "virtual world" it is nat-ural to include representations of humans. A realistic visualapproximation of a human requires a polygon mesh of sig-nificant complexity, especially as it has to be sufficiently tes-sellated to allow smooth deformations of the model.Visualising one virtual human in real-time can consumea high proportion of processing time. In many situations itis not one, but a crowd of humans, which meed to be ren-dered. The common aim of all of the papers discussed here isto reduce the computational expense of a particular elementof the scene, allowing interactive frame rates. In addition tothis, the aim is to free the CPU to handle other elements ofthe simulation, e.g. A.I., I/O or physics.The current growth in power of consumer level graphicsprocessors is significant and with this many features are pro-vided which are designed to facilitate efficient rendering ofscene. As the growth in power continues, so does the desireto render even more complex scenes with realistic lighting.Visualising virtual environments is a huge area of research,with many inter-related fields. However this paper aims tocover only recent advances in the field applicable to the vi-sualisation of virtual humans.What particular problems need to be overcome in visual-ising virtual humans? The human body needs to be modelledas a deformable structure, thus many of the traditional tech-niques that are designed for rigid body techniques are notapplicable. One partial solution is to split the human bodyinto a series of rigid elements, considering each element asan individual body on which to apply the techniques. Asmany comment, including Aubel et al [ABT99], this causesproblems at the joints (the boundaries between the rigid ele-ments) that have to be carefully maintained to avoid cracksappearing. As with all deformable body work the mesh maybe sufficiently tessellated in its rest pose. However as thecharacter moves and bends the model may be too coarse togive realistic deformation of joints.Virtual humans also have further complications over stan-dard deformable bodies. The human visual system puts alarge emphasis on the human face and thus any interactionwith a virtual human should be echoed in facial expressionas well as the actions. Realistic lighting and self shadowingcontribute heavily to how we see facial expressions. This isvery computationally expensive and has only recently beencontemplated for real-time circumstances.In this paper, Level of Detail (LOD) methods are first ex-amined, looking at mesh decimation and mesh refinement