ABSTRACT Recent advances in artificial intelligence have enabled temporal super‐resolution fringe projection profilometry (FPP) to overcome the inherent frame‐rate limitations of image sensors through frequency‐domain multiplexing, thereby achieving high‐speed 3D imaging with low‐frame‐rate cameras. In recent studies, researchers have significantly enhanced frequency‐domain multiplexing efficiency by compressing multiple frequencies of fringe patterns into a grayscale image, enabling 3D imaging with higher temporal resolution. However, this approach relies on frequency‐domain compression, which limits further improvements in compression ratio and restricts the flexibility in selecting fringe pattern frequencies. To address this limitation, we introduce, for the first time, the integration of color‐channel multiplexing with frequency‐domain multiplexing, proposing a triple‐frequency color‐multiplexed FPP (TFCMFPP). In this approach, multi‐directional fringe patterns at three distinct frequencies are encoded into the projector's RGB channels. Under high‐speed projection, a single long‐exposure image can be acquired by a low‐speed color camera to encapsulate the multiplexed fringes, thus enabling one‐shot dense information packing. To decode the multiplexed patterns, we developed a three‐stage deep neural network comprising zero‐order removal, spatial‐frequency decoupling, and direction‐aware expert modules. By combining this architecture with projection distance minimization (PDM) based phase unwrapping, we can robustly recover multi‐directional absolute phases without relying on stereo or temporal priors. This allows 3D reconstruction at up to 24 the native device speed. We validate the proposed method on a variety of transient scenes, including collapsing pillars, rotating rotor‐engine models, and multi‐object rotational scenarios. Experiments demonstrate that TFCMFPP achieves an effective 3D reconstruction rate approaching 5,000 Hz using only a 208 Hz color camera, while maintaining high spatial fidelity, low depth error, and excellent robustness to motion. With its low cost, compactness, and scalability, TFCMFPP integrates color, frequency, direction, and temporal dimensions into a unified coding framework, opening new avenues for high‐throughput 3D imaging.
Liu et al. (Tue,) studied this question.