点光源による半影表現のためのレイトレーシング拡張アルゴリズム

Abstract

Penumbra is an important factor for three dimensional Computer Graphics in order to represent realistic images especially at interior views. Point light represents images with sharp-edge shadow that is far from reality. On the other hand, area light casts penumbra at the situation where occluders prevent some part of light area from reaching calculation points. Although path tracing and radiosity are technologies that handle area light, both of them take so enormous times for processing that they are not suitable to use at the stage of reviewing designs which are weighed with turnaround time. Therefore, this paper proposes a new high-speed algorithm that represents high resolution penumbra casted from psuedo area light by applying ray tracing that handles point light only. It is inevitable that the previous technologies, namely percentage closer filtering (PCF) and variance shadow maps (VSM), generate low quality penumbra because they are algorithms based on shadow maps that include aliasing problem. We name this new algorithm “light obstacle depth”. Light obstacle depth calculates intensity of shadow by point light at each calculation point at first and subsequently blends shadow around the calculation point according to the distance between the nearest occluder and the point. Ray tracing produces shadow with shading at a time which contains textures, reflection, transparency, and refraction, etc. However, light obstacle depth separates shadow calculation process from other shading process, calculates shadow intensity independently, stores them at each pixel to make shadow buffer, and at last combines the shadow buffer and other shading together after blending shadow buffer. Blending problems around occluders occur when we apply the basic blending technique, so we propose an enhanced blending technique to solve the problem. In addition, we show the equation that estimates blending area using size, direction, and distance of an area light simulated. We also propose a technique to estimate blending area on image with perspective transformation. To validate quality and speed of the new algorithm, we used two different types of models, namely a model with scattered polygons and a model crowded in small area. Compared with images generated accurately by area light, images by our new algorithm show that light obstacle depth is able to simulate penumbra of area light with practically sufficient quality. In terms of the calculation time, our new algorithm is capable to generate penumbra at almost the same time as point light calculation and several tens of times faster than normal area light calculation. Furthermore, we show that calculation speed of our new algorithm becomes more efficient for more complex data. We aim that light obstacle depth is processed using GPGPU (general purpose GPU) technology. As GPGPU processing is expected to be several tens of times acceleration, real time penumbra generation will come true after GPGPU application to light obstacle depth is realized.