Improved structured light system based on generative adversarial networks for highly-reflective surface measurement

Gray code pattern structured light projection technology is widely used in industrial inspection due to its good robustness and anti-noise performance. Gray code pattern technology projects a sequence of encoded fringe patterns with black and white strips onto the scanned object in order to measure its height distribution. However, if the scanned object has strong specular reflection properties, the acquired encoded fringe images tend to miss significant amounts of local area information. As a result, the measured three-dimensional point clouds contain many missing points, and hence the measurement accuracy is severely degraded. To address this problem, the present study proposes a novel fringe-inpainting system based on a generative adversarial network framework, to repair the fringe features in the regions of the scanned surface in which the local information is lost. The performance of the proposed fringe-inpainting system is compared with that of several other advanced highly-reflective surface measurement technologies reported in the literature. The experimental results show that the proposed method significantly outperforms these techniques and yields an excellent encoded fringe inpainting for all of the considered objects.