We address the problem of normal re

我们解决了透明物体的法线重建问题，该透明物体的整体表面是目标物体的整体形状，如果在其外表面上存在，则可以保留显着特征和精细结构。这种保留细节的外表面表示法对于视觉和机器人应用程序是足够的，在视觉和机器人应用程序中，透明物体要由机械手抓住，或在混乱的场景中由导航机器人避免。在我们的论文中，我们的目标不同于透明物体的真实感渲染或高精度重建，在透明物体的重建中，通常认为定制设备，校准和机械捕获对于实现高达如示踪物所表现出的复杂折射光传输路径的精度是必要的。目标对象。另一方面，<br>使用较小的预算。<br>无需昂贵或复杂的设置，同时仍然可以移植足够的重建，可以利用哪些关于透明物体的视觉线索？尽管我们当中有些人经历过不愉快地撞入玻璃窗的不愉快经历，但尽管它们具有明显的透明度，我们仍然可以看到各种各样的透明物体，因为它们中的大多数会折射并反射入射的光。使用校准的设置和捕获来跟踪折射光的传输路径，有助于将瞄准技术用于高精度重建。另一方面，本文利用直接反射自透明物体的镜面反射来产生适当的重建。但是，由于我们廉价的摄像机的动态范围较低，因此，复杂的光传输（例如，焦散和全内反射）也会产生强度很高的高光，其强度与直接镜面高光一样强。因此，要解决的主要问题是在每个像素处进行识别<br>由直接镜面反射引起的高光的子集。

We address the problem of normal reconstruction for a transparent object, where the integrated surface is an overall shape of the target object that preserves salient features and fine structures if present on its exterior surface. Such detail-preserving exterior surface representation is adequate for vision and robotics applications where transparent objects are to be grabbed by a robotic arm, or avoided by a navigating robot in a cluttered scene. In our paper, our goal is different from photorealistic rendering or high-accuracy reconstruction of transparent objects, where custom equipment, calibrated and mechanical capture are often deemed necessary to achieve precision as high as to trace the complex refractive light-transport paths exhibited by the target object. On the other hand, when an adequate shape without this level of precision is sufficient, it is possible to propose a reconstruction approach that uses a simpler setup realizableusing a smaller budget.Without expensive or complicated setup while still supporting an adequate reconstruction, what visual cues concerning a transparent object can be utilized? Although some of us have had the unpleasant experience of smacking into a glass window without seeing it, we can still see a wide range of transparent objects despite their apparent transparency, because most of them refract and reflect incoming light. Tracing refractive light-transport paths using calibrated setup and capture had contributed to the success of techniques aiming at high-precision reconstruction. This paper on the other hand makes use of specularities directly reflected off an transparent object to produce an adequate reconstruction. Due to the low dynamic range of our inexpensive video camera, however, indirect reflection caused by complex light transport (e.g., caustics and total internal reflection) also produces strong highlights with intensity that appears as strong as direct specular highlights. Thus,the main problem to be solved is to identify at each pixelthe subset of collected highlights that are caused by direct specular reflection.<br>