Exploiting Light Directionality for Image‐Based 3D Reconstruction of Non‐Collaborative Surfaces

Abstract

Three-dimensional (3D) measurement of non-collaborative surfaces is still an open research topic. This paper investigates and quantifies for the first time the effect of light directionality and fusion of multiple images as a method to improve the quality of photogrammetric 3D reconstruction. For this aim, an image acquisition system that employs multiple light sources was developed to highlight the roughness and microstructures of the object under investigation. Images were captured at various grazing angles to highlight the local surface roughness and microstructures. Individual point clouds, created using images taken at different grazing angles, were produced using dense image-matching techniques. These point clouds were then compared against different 3D photogrammetric reconstructions obtained from a pre-processing of the acquired images based on diffuse lighting, median and average images. Experiments showed that exploiting light directionality significantly improves image-matching quality. Furthermore, depending on the light direction, the root mean square (RMS) error of the 3D surfaces obtained using the proposed system were up to 50% less than those created by traditional diffuse lighting.

Combining Photogrammetry and Photometric Stereo to Achieve Precise and Complete 3D Reconstruction

“…The first step is to provide an automatic image acquisition system to capture images under different illuminations and from different camera stations (camera positions) to satisfy the input requirements of the integrated method. A 3D point cloud is then generated with a photogrammetry pipeline [ 11 ]. The photogrammetric 3D shape measurements and the calibrated light positions are utilized to compute light direction and intensity attenuation (radial and angular) at each surface point.…”

“…Therefore, this impact must be addressed for accurate measurement of the normal surface as any change in light position negatively affects the normal and consequently the resulting 3D reconstruction. For these reasons, we explored a geometric model with punctiform light sources and light divergence as expressed in [ 11 , 25 ].…”

“…In the case of reflective objects, cross polarisation [ 35 , 36 ] and image pre-processing approaches [ 37 , 38 , 39 ] have also been employed to decrease specular reflections. However, these procedures may smooth off surface roughness or vary the texture from one view to the next, negatively affecting the results [ 11 ]. Another common approach to spray the surface with a thin layer of white or colored powder [ 1 , 2 , 25 , 40 , 41 , 42 ] can also be used as a common solution.…”

“…These methods are commonly used in various applications [ 4 , 5 , 6 ]. In particular, in the field of industrial metrology, they can be used for quality control [ 7 , 8 , 9 , 10 ], inspection [ 11 , 12 , 13 , 14 ], reverse engineering [ 10 , 15 , 16 , 17 ] or 3D micro-measurement [ 18 , 19 , 20 ].…”

“…Finally, we co-register and fuse all obtained 3D measurements to generate a precise and complete 3D reconstruction of non-collaborative objects. The paper builds upon [ 11 , 25 ], in which they only present the main concept and high-level architecture of the prototype system developed in the FBK-3DOM laboratory to synergistically combine photogrammetry and photometric stereo methods. They discuss one of its main components, for instance, the concept of utilizing light directionality to increase texture quality by exploiting shade and shadow phenomena in detail and only addressed parallel light direction without taking intensity attenuation into account.…”