Outlier removal based on boundary order and shade information in structured light 3D camera

Structured light 3D camera is considered as one of the most reliable techniques for 3D reconstruction by producing high accuracy 3D point cloud. However, in case that the camera is slant with object surface, the reflected light into camera may be weakened, resulting the occurrence of outliers (i.e. wrong decoded pixels) in 3D point cloud. State-of-the-art methods to deal with this problem have made remarkable improvements, but still depended on the assumption that the surface to be captured is smooth. In this paper, instead of operating on the point coordinators only, we propose a new method to identify outliers based on the shade information and camera-projector correspondence. To enhance our solution, we also introduce a new shade detection algorithm that using the variation of gray colors from captured patterns. Compared with other methods, the experimental results have proved that the proposed method is more effective in removing outliers, even when the surface is not smooth. Besides, our method is also able to detect clusters of density outliers, which is one of limitation of existing approaches.