In-situ 3D reconstruction of worn surface topography via optimized photometric stereo

Since worn surfaces contain rich information of the wear mechanisms, in-situ measurements of surface topography can characterize ongoing wear degradation in machines. With the help of photometric stereo vision, threedimensional (3D) topography of worn surfaces is obtained with a monocular microscope. However, the accuracy of the reconstructed surfaces remains low due to the non-Lambertian reflections of worn surfaces and noise in the image acquisition equipment. To address this issue, an optimized photometric stereo approach is proposed for the improvement of worn surface reconstruction. To accommodate the non-Lambertian reflections, a multi branch network is constructed to estimate normal vectors from both the photometric images and the incident illumination directions. The estimated normal vectors are adopted to reconstruct worn surface topography by embedding prior knowledge. With this design, the overall distortion caused by image noise is effectively suppressed. The proposed method is verified by comparing with the Laser Scanning Confocal Microscopy (LSCM). As the main result, over 88% similarity on the worn surface roughness can be obtained.

Automated summary

This study proposes an optimized photometric stereo approach to improve the reconstruction of worn surfaces. By constructing a multi-branch network and embedding prior knowledge, the influence of image noise on the reconstruction results is effectively suppressed. Compared with Laser Scanning Confocal Microscopy, this method can achieve over 88% similarity in worn surface roughness.