Robust structured light system against subsurface scattering effects achieved by CNN-based pattern detection and decoding algorithm

Ryo Furukawa, Daisuke Miyazaki, Masashi Baba, Shinsaku Hiura, Hiroshi Kawasaki; Proceedings of the European Conference on Computer Vision (ECCV) Workshops, 2018, pp. 0-0

To reconstruct 3D shapes of real objects, a structured-light technique has been commonly used especially for practical purposes, such as inspection, industrial modeling, medical diagnosis, etc, because of simplicity, stability and high precision. Among them, oneshot scanning technique, which requires only single image for reconstruction, becomes important for the purpose of capturing moving objects. One open problem of oneshot scanning technique is its instability, when captured pattern is degraded by some reasons, such as strong specularity, subsurface scattering, inter-reflection and so on. One of important targets for oneshot scan is live animal, which includes human body or tissue of organ, and has subsurface scattering. In this paper, we propose a learning-based approach to solve pattern degradation caused by subsurface scattering for oneshot scan. Since patterns are significantly blurred by subsurface scattering, robust decoding technique is required, which is effectively achieved by separating the decoding process into two parts, such as pattern detection and ID recognition in our technique; both parts are implemented by CNN. To efficiently achieve robust pattern detection, we convert a line detection into segmentation problem. For robust ID recognition, we segment all the region into each ID using U-Net. In the experiments, it is shown that our technique is robust against strong subsurface scattering compared to state of the art technique.