A Study of a Realistic Camera Model for Global Illumination

• Main Authors: Chen Chien Hao, 陳建豪
• Other Authors: C. M. Wang
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/06558501036545185712

碩士 === 國立中興大學 === 資訊科學研究所 === 89 === Most recent rendering research in computer graphics community has concentrated on two issues: the light reflection problem and the light transport problem. The first issue is to model the reflection of light from materials, while the second one is to calculate the direct and indirect illumination from light sources and other surfaces. Another key component of a rendering system is the camera model. Kolb et. al. investigated realistic cameral models for computer graphics and presented a physically-based camera model. The proposed camera model can accurately compute the irradiance on the film given the incoming radiance from the scene. However, their works also demonstrate some limitations. The camera model they presented is not sufficient for generating images with motion blur or rendering scenes under different media, nor can the final images produce plausible visual effects such as color bleeding and inter-reflection. This is due to that they only employed local illumination model for distributed ray tracing algorithm. In this thesis we try to explore methods and techniques to overcome the previous restrictions. More precisely, we adopt Monte Carlo Path Tracing as the rendering algorithm, together with a global illumination model, the modified Phong Model, as a reflectance model. Similar to Kolb’s work we utilize the thick lens approximation to synthesize images of adopting a physically-based camera model, which is described as a lens system and film backplane. However, we present a further analysis for sampling the lens system through the values of the RMS error so that the radiometry is computed accurately and efficiently. Besides, the visual effect of motion blur is synthesized using a proposed technique, which takes into consideration both the camera shutter and the elapsed time interval. Also present is our approach to rendering scenes under various media given the corresponding indices of refraction. Finally, we employ two filtering techniques, the box filter and the non-uniform filter, to alleviate the aliasing impacts on the synthesized images. We implemented the proposed techniques as a camera module using C programming language. This module was then embedded into an existing rendering system modified by a public domain software, Rayshad. Experimental results demonstrate that taking 400 samples on a single pixel is a time-effective approach to generating images with plausible visual effects using a physically-based camera model. We implemented a full lens simulation and compared the rendered images with those synthesized by the thick lens approximation approach. Experimental results verify the feasibility of adopting this approach since the visual effects are almost negligible when using Monte Carlo Path Tracing for global illumination calculations. We display some rendered images illustrating the focusing effects in the camera model, together with some synthesized images using various lens systems including Double-Gauss, Wide-Angle, Fish-Eye, and Telephoto lens systems. In conclusion, the synthesized images presented in the thesis have demonstrated their plausible visual effects including color bleeding and inter-reflection, while a number of visual effects can now be synthesized using the proposed techniques. We conclude that the major contribution of this study is the generalization of previous approaches proposed by the Kolb et. al., and the extension of the realistic camera model for global illumination.