End-to-end single-shot composite color FPP network for multiple separated objects reconstruction

Single-shot fringe projection profilometry (FPP) has been widely used in 3D reconstruction of dynamic scenes. However, 3D reconstruction of multiple separated objects using a single image is inherently an ill-posed problem, as it is not possible to accurately unwrap the phase of separated objects based on a single image. To address this problem, this paper adopts the method of projecting composite color fringe, which involves projecting fringes with different frequencies through three color channels. Also, this paper designs a tailored network structure that can output depth information end-to-end based on a composite color fringe, referred to as WLKDCA-Net. Inspired by wavelet transform profilometry (WTP), WLKDCA-Net introduces the wavelet transform domain information to enhance its phase retrieval ability for single fringe image. To fully utilize different frequency fringe patterns between color channels, WLKDCA-Net incorporates an implicit Frequency-domain Channel Attention (FCA) module. In addition, we construct the first composite color FPP dataset for multiple separated objects using virtual software. Experimental results on both virtual and real data show that our network outperforms traditional and other deep learning methods. The code is available at: https://github.com/LianpoWang/WLKDCA-Net .