omni-directional image - PDF document

FEV 1 FEV2 omni-directional image fisheye An acquisition directional images points with extraction algorithm [9]. in two orthogonal directions, e.g., feature point. real scene direct light sources such energy comes from indirect illumina- tion such t,he direct light illumination distribution a real scene. Fortunately, direct light appear as bright points direct light feature points using a stereo algorithm (41. are approximated This part indirect light do not appear distinct point in entire shape real scene, e.g., indirect light In our [a], imaging plane and the related as 2: radiance distribution the diameter fisheye lens and cr and the ray from object surface lens center. Therefore, by color values Equation(l), radiance distribution limited dy- taken with one scene accurately. taken with produce each omni-directional a virtually extended dynamic Superimposing Virtual Objects Measured Illumination radiance distribution scene is mea- can render virtual object onto the by using radiance distribution. each pixel the input scene, a ray extending image coordinate system4 real object in the 3CCD color camera (Victor KYF-57) and fisheye view of ' calibration algorithm proposed by Ill] camera parameters images synthesized our method this paper, proposed a virtual objects onto an scene by taking into account radiance distribu- scene. For matching illumination, diance distribution real scene is two omni-directional scene. Un- previously proposed light sources a real can measure an entire radiance distribution method can superimpose objects with correct ra- can be used for generating shadows cast (b 1 calibration image (c) FEV l FI:V~ . - [I] R. T. Azuma, "A survey of augmented reality," Presence: Te[eoperators and Virtual Environments, vol. no. 4, pp. 355-385, .August 1997. 12) Debevec, "Rendering Synthetic Objects with Global Illumination Range Pho- Pho- G. Drettakis, L. Robert, S. Bougnoux, "Interactive Com- Eurographics Workshop on Rendering, pp. 45-57, 1997. [4] 0. Faugeras Three-Dimensional Computer Vision: A Figure 5: Geometric Viewpoint, MIT Press, 1993. omni-directional A. Fournier, A. Gunawan, and C. Romanzin, "Com- between Real and Computer '99, pp.254- 262, 1993. 16) B. K. K. S. K. Nayar, K. Ikeuchi, and T. Kanade, "Surface re- flection: physical geometrical perspectives." PAMI, no. 7, 611-634, 1991. [8] objects into Report, Computer Univer- 4 'la sity of Tokyo, 11s-CVL-98-101, January 1998. [9] C.Tomasi Kanade, "Shape Motion from Factorization Method," International Journal vol. 137-154, 1992. [lo] K. E. Torrance and E. M. Sparrow, "Theory for off- from roughened Journal Society of America, vo1.57, pp.1105-1114, 1967. [ll] Accuracy Machine images synthesized with our IEEE Trans. R El A, vol. 3, no. 4, 323-344, 1987. accoi~nt: time. More virtual object has location in real scene, object must correct location in geometry). the virtual has to match that of other objects in tlie scene, and tlie virtual object must a correct i.e., a shadow those of shactows real scene virtual object In the past, consistency of 'l'okyo lOf~-XT,SX Japan. E-mail: in~arik, ysato, kiQcv1.iis.11-Lokyo.ac..jp ' Ibr a good survey of augment4 reality techno log it?^, see ['I virtual ol~jects. This is mainly 1)ecansc. real scenes 110th direct and indirect illumination distril~i~tcd a complex Therefore, it is not easy to obtain correct illi~mination modrls to be used for ai~gmrnted reality systems. srgmcrits. propose a a real scene using a radiance distribution distribntion in the real scene by using two omni-directional im- ages of the scene. Then the measured radiance dis- tribution is used for rendering virtual ohjects scene image. previously proposed our method can from all by using a by a fisheye This paper is organized radiance distribution