摘要： 在基于神经隐式表面学习的多视图三维重建过程中，复杂物体的几何形状和外观表示存在潜在的模糊性。因此，物体的几何细节信息在纹理稀疏区域、边界区域与较大光滑区域中容易丢失，难以精确恢复。为解决这个问题，提出了一种基于多视图混合一致性约束的神经隐式表面重建方法。该方法采用多视图立体（MVS）、多视图光度一致性与特征一致性、体渲染技术来优化隐式表面表示，从而重建具有精细几何细节的复杂物体模型。首先，提出了一个基于多视图立体的稠密点生成模块，通过MVS生成稠密点，来补充物体表面纹理稀疏区域与边界区域的细节信息，从而实现物体表面的多视图几何优化。其次，提出了多视图混合一致性约束模块，通过符号距离函数（SDF）定位零水平集，利用多视图光度一致性约束来对物体光滑区域进行几何约束，监督所提取的隐式表面，并对经过线性插值的SDF过零处的表面点应用多视图特征一致性约束，弥补纹理稀疏区域或结构复杂区域像素匹配的误差，从而优化物体重建模型。最后，通过应用体渲染技术，利用隐式的SDF得出高质量的图像渲染，以实现复杂物体的精确表面重建。实验结果表明，在DTU数据集中，相比于Colmap等方法，所提方法峰值信噪比（PSNR）提升了40.3%以上，实现了物体表面的精确重建。

Abstract: In the process of multi-view 3D reconstruction based on neural implicit surface learning, there are inherent ambiguities in the representation of the geometric shape and appearance of complex object. Therefore, fine geometric details of the object are prone to being lost in sparse texture areas, boundary, and large smooth surfaces, making accurate recovery difficult. To address this issue, a novel neural implicit surface reconstruction method based on multi-view mixed consistency constraints is proposed. This method uses multi-view stereo (MVS), multi-view photometric consistency, feature consistency, and volume rendering techniques to optimize the implicit surface representation, enabling the reconstruction of object models with fine geometric details. Firstly, a dense point generation module based on multi-view stereo is proposed, which generates dense points through MVS to supplement detail information in sparse texture and boundary of the object surface, achieving multi-view geometric optimization of the object surface. Secondly, a multi-view mixed consistency constraints module is introduced, which uses the signed distance function (SDF) to locate the zero-level set. It applies multi-view photometric consistency constraints to impose geometric constraints on the smooth regions of the object, supervising the extracted implicit surface. Additionally, multi-view feature consistency constraints are applied to surface points at the zero-crossing of the linearly interpolated SDF, compensating for pixel matching errors in texture-sparse or structurally complex regions, thereby optimizing the object reconstruction model. Finally, volume rendering technology is applied to produce high-quality image renderings from the implicit SDF, enabling precise surface reconstruction of objects. Experimental results show that, compared to methods like Colmap, the proposed method achieves a improvement in peak signal-to-noise ratio (PSNR), increasing by over 40.3% on the DTU dataset, and successfully enables accurate surface reconstruction of the objects.