Efficient and High-Fidelity High Dynamic Range 3-D Imaging via Spatial-Temporal Speckle Projection

The structured light projection profilometry is highly sensitive to the simultaneous overexposure and underexposure caused by high dynamic range (HDR) scene reflectances from a given viewpoint. This leads to a severe loss of fidelity in the measurement results, which is reflected in both completeness and accuracy. The prevailing solutions, such as multiexposure fusion (MEF), adaptive projection, and equipment-based methods, often suffer from low efficiency, particularly in terms of measurement time or system complexity. In this article, we propose a multiframe active 3-D imaging strategy based on spatial-Temporal speckle projection for efficient and high-fidelity HDR reconstruction. First, a gradient-feature-Assisted seed point generation combined with binary-feature-sequence-based spatial-Temporal matching provides robust integer-pixel initial guesses for subpixel refinement, even in low-contrast and saturated regions where conventional speckle matching fails. Then, we introduce an inverse compositional Gauss-Newton (IC-GN) algorithm to HDR 3-D surface reconstruction. By extending the temporal sequence, the matching window size can be reduced without sacrificing accuracy, overcoming the resolution-Accuracy tradeoff inherent to single-pattern speckle methods and enabling pixelwise subpixel-Accuracy reconstruction. The method requires only a single exposure for each speckle pattern to achieve high-fidelity 3-D reconstruction of HDR scenes. The comparative experimental results on various HDR objects demonstrate that the proposed method outperforms the existing techniques in terms of both point cloud completeness and geometric accuracy, while maintaining system simplicity and time efficiency during measurement.