Efficient Picking through Atomic Operations

Abstract

Picking is the process through which a single entity or a list of entities is selected from a scene. The subject of picking is both a rendering and a collision detection problem, with the majority of research being on optimizing the ray-scene intersection problem. Several algorithms that solve the picking problem exist in the context of rasterization, but all of them lack several of the features of our proposed solution while, with one exception, all being much more expensive in terms of computational time. We propose a novel single frame method which is able to correctly select not only primitives but also any type of objects that may appear on the screen at a fragment level including hardware instanced, alpha culled, hardware tessellated, hardware animated and particle systems. The proposed technique has optimal memory requirements and offers the opportunity to pick at micro polygon level and is not limited to the first contact, offering the full intersection list per ray if required to do so. The proposed technique offers further unique opportunities such as flexible fuzzy object selection and is the only technique that selects without disregarding opacity accumulation from transparent objects.