基于结构光相机与隐式表面重建的室内建模

Indoor modeling based on structured-light camera and implicit surface reconstruction

深度相机在室内建模领域中以其快速、精准、鲁棒的特性而备受青睐。然而,在实际应用中,深度相机在获取精准深度信息时面临诸多挑战,如环境光干扰、点云噪声众多以及深度缺值。针对上述问题,提出了一种新型的结构光神经隐式表面(SL-NeuS)重建方法。通过结合广泛的三维重建技术,包括多目立体视觉(MVS)、神经网络、神经辐射场和神经隐式表面(NeuS),克服了现有技术在室内建模中的局限性。特别地,利用基于可微分循环优化启发设计的同步定位和建图(DROID-SLAM)算法精确获取相机外参,并结合单目深度估计和单目法向估计先验信息,有效地实现了精确的室内三维建模。实验结果显示,SL-NeuS重建方法在不同深度相机的性能分析中表现优越,成功降低了环境光影响和光学失真对建模精度的干扰。应用该方法对各种室内场景进行建模,误差可控制在极小范围内,实现了高精度的室内三维重建。本方法不仅提高了室内建模的准确性,还大幅减少了训练时间,为数字建筑和数字导航等领域提供了重要的技术支持。

Depth cameras are highly valued in indoor modeling for their rapidness,precision,and robustness.Yet,they encounter challenges such as environmental light interference,point cloud noise,and incomplete data depth.To overcome these hurdles,an innovative structured light neural implicit surface(SL-NeuS)reconstruction method integrates a broad spectrum of 3D reconstruction technologies,including multi-view stereo(MVS),neural networks,neural radiance fields,and neural implicit surfaces(NeuS).This method leverages the differentiable recurrent optimization-inspired design simultaneous localization and mapping(DROID-SLAM)algorithm to accurately determine camera extrinsics.By incorporating monocular depth estimation and monocular normal estimation of prior information,it achieves precise indoor 3D modeling.Experimental findings demonstrate that the SL-NeuS reconstruction method excels in performance analysis across diverse depth cameras,effectively minimizing the impact of environmental light and optical distortions on modeling accuracy.Applying this method to model various indoor environments keeps errors within a narrow range,ensuring high-precision indoor 3D reconstruction.In addition to enhancing indoor modeling precision,it significantly reduces training time,offering essential technical support for digital architecture and digital navigation fields.