Interfered depth map recovery with texture guidance for multiple structured light depth cameras

In depth acquisition systems with multiple structured light depth cameras (SLDCs), interference between the devices causes interfered regions, which degrade depth quality and impair applications. This paper proposes a novel approach to recover the depth maps. Under the guidance of texture segments, interfered regions are categorized into flat regions and boundary regions. After that, different strategies are applied to the two kinds of regions because of their property differences. For flat regions, a Markov random field (MRF) model is utilized to get the optimal gradient solution. With the gradients, discrete Poisson equation (DPE) is applied to calculate the final depth solution. In boundary regions, another texture-guided MRF is utilized to peruse depth directly. Experiment results demonstrate that the proposed method works well on both synthetic data and captured data. In flat regions, we get smooth-varied depth, and boundaries between interfered and non-interfered regions are seamless. In boundary regions, sharp depth edges between objects are well preserved. Moreover, our method improves the PSNR of synthetic data by 4-14 dB.