A unified probabilistic graphical model based approach for the robust decoding of color structured light pattern

Color coding is an important research topic in spatial encoded structured light sensing (SLS). In this study, we propose a novel graphical model based approach for the color pattern decoding task. For efficient color labeling, the color pattern is firstly decomposed into separate binary pattern images. With the labeled pattern elements, a unified probabilistic graphical framework is constructed to represent the pseudorandom pattern as a clique tree structure. The model contains two parts: the Conditional Random Field (CRF) is used to represent the dependences between these local decisions, and the Bayesian network (BN) is applied for the representation of background colors effect. A colorful target is experimented to demonstrate its feasibility. And the 3D reconstructed models based on the decoding results are also provided to show its robustness.