基于OpenGL驱动的三维场景重构

3D scene reconstruction based on OpenGL

基于Open GL环境,提出利用场景渲染结果即帧缓存数据(主要包括颜色缓存及深度缓存)进行局部三维场景重构的方法。根据投影变换原理和深度缓存特点研究了逆投影变换,得到了三维点元的重构方法。在此基础上,利用帧缓存数据重构每像素对应的三维点元,从而构成相机坐标系下的三维点云模型,即原始三维场景在当前相机视角下的离散采样,被定义为虚拟视模型。根据透视投影和帧缓存的特点,分析了虚拟视模型的点位精度,实验表明虚拟视模型具有较高的相对精度。提出的场景重构方法具有天然的多分辨特性,支持场景简化与流式传输以及独立于场景的具体表示形式,可采用拦截方式从一般三维软件中实时重构虚拟视模型,从而提供了一种获取三维数据的新方法。

Based on Open GL,this paper presented a local 3D scene reconstruction from the frame buffers（ including color buffer and depth buffer） after 3D scene rendered. According to projection principles and depth buffer characteristics,it discussed an inverse projection processing and obtained theoretically a 3D vertex reconstruction method. As a result,it transformed all valid pixels within frame buffers and generated a 3D vertex set which was a point cloudy model in the camera space in the conducted experiments. Obviously,this cloudy model was a kind of a sample of the original 3D scene in current camera system and named as virtual vision model（ VVM）. Secondly,it analyzed the point-wise precision of VVM detailed. The results of experiments show that VVM appears very high relative accuracy to the original 3D scene. Above all,the reconstruction method appears natural multi-resolution feature with respect to the resolution of frame buffers and supports scene simplification and streaming transmission. On the other hand,this method is capable of abstracting VVM real-time from the common 3D software easily without any details of 3D scene known previously,which provides a new approach acquiring 3D data efficiently.