基于图像的室内虚拟环境的研究

基于图像的建模和绘制技术，提出了一个构造室内虚拟环境的完整方案，用户只需要输入少数照片，即可重建室内场景的全景图像，方案主要包括以下几点：首先由用户交互确定图像中的匹配象素，通过运动分析算法恢复整个场景的几何结构，然后，将原始图像变换至平面的参数坐标系，抽取纹理图像，并在参数空间对纹理图像进行拼接；最后生成场景的全景图像，算法对拍摄条件和设备没有苛刻要求，运算量较小，有较强的稳定性。

AN IMAGE-BASED APPROACH FOR THE RECONSTRUCTION OF VIRTUAL INDOOR ENVIRONMENT

Traditional approach proposed for walking through indoor scenes is generally geometry based, which uses 3D computer graphics to model and render virtual environments. Main disadvantages of this approach are that it usually requires laborious modeling, and the rendering images are not realistic. In recent years, in order to generate real time photorealistic graphics, image based method have been emerged for the development of virtual reality systems by modeling existing environment. This paper presents a new approach for generating the panoramic image of real indoor scene from a sparse set of photographs. The approach includes several steps: At first, a set of corresponding pixels in the input photographs are determined interactively by users. Secondly, with feature based structure from motion algorithm, geometrical model of the scene is reconstructed to form both these corresponding pixels and the geometrical constraints in the object space. Thirdly, the input photographs are mapped to 2D parameter planes, from which the scene's textures are extracted. At the next step, the textures that locate on the same plane are stitched together. Finally, a panoramic image can be generated from these textures. The experimental results show that these algorithms are robust and need small capacity to store the panoramic image. Additionally, no strait condition is imposed on the original photographs.