A conceptual study of infrared and visible-light image fusion methods for three-dimensional object reconstruction
A conceptual study of infrared and visible-light image fusion methods for three-dimensional object reconstruction
复制标题
用于三维物体重建的红外和可见光图像融合方法的概念研究
DOI：
10.1117/12.2527428
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Borges
中科院分区：
文献类型：
--
作者：
Marcellino;de Oliveira;Borges
在工业4.0蓬勃发展的同时，与工业产品三维评估相关的需求也在增加。通常，单个传感器无法在不显著增加程序成本的情况下实现这种复杂性，这对许多应用来说可能是令人望而却步的。摄影测量是一种基于图像的三维测量技术，可以提供良好的测量结果，但在使用对象纹理进行点匹配时存在局限性。使用红外相机可以在测量表面特征或半透明零件数量较少的部件时具有优势，但通常它们没有足够的分辨率来使用典型的摄影测量技术正确重建对象。这种情况适合使用两个或多个传感器的数据融合，从而产生更多信息的点云。因此，本文提出了利用可见光摄像机和红外摄像机采集的数据对透明零件进行三维测量的方法。在该方法中，基于二维小波分解的像素级图像融合技术被用作配准过程的一个步骤来组合来自这些设备的图像。将该过程与仅使用红外图像和仅使用可见光图像重建对象进行了比较。结果表明，与仅使用可见光或红外图像相比，使用数据融合可以获得更完整的点云。
While industry 4.0 thrives, the needs related to three-dimensional assessment of industrial products growths in complexity. Frequently, this complexity cannot be achieved by one single sensor without outstandingly escalating procedural expenses, which can be prohibitive for many applications. Photogrammetry is an imagebased three-dimensional measurement technique which can offer good metrological results, although presenting limitations specially when using the object texture for point correspondence. Employing an infrared camera can imply in advantages when measuring components with low number of surface features or translucent workpieces, but generally they do not have resolution enough to properly reconstruct the object with typical photogrammetry techniques. This scenario is proper for using data fusion of two or more sensors, resulting in a more informative point cloud. Therefore, it is proposed in this work the three-dimensional measurement of a transparent workpiece using data seized from a visible-light camera and an infrared camera. In the proposed approach, a pixel-level image fusion technique based on two-dimensional wavelet decomposition as a step of the registration process is used to combine images from these devices. This procedure is compared with the reconstruction of the object using only the infrared images and only the visible-light images. The results show a more complete point cloud using data fusion compared to use only visible-light or infrared images.