This paper presents a volumetric method for integrating range images to reconstruct complex 3D surfaces with high detail and accuracy. The method addresses key challenges in surface reconstruction, including handling directional uncertainty, filling gaps, and robustness to outliers. The algorithm uses a cumulative weighted signed distance function to represent the surface, combining range images incrementally and efficiently. It employs run-length encoding for space efficiency and resampling to align with the voxel grid for time efficiency. The final surface is extracted as an isosurface from the volumetric grid, which is optimal in the least squares sense under certain assumptions. The method is capable of integrating up to 70 range images, producing high-resolution models with up to 2.6 million triangles. It handles gaps in the reconstruction by identifying empty regions and filling them with plausible surfaces. The algorithm is robust to sensor errors and can handle complex shapes, including those with sharp corners and thin surfaces. It also supports efficient processing through run-length encoding and synchronized scanline traversal. The paper discusses previous work in surface reconstruction, including methods based on unorganized points, parametric surfaces, and implicit functions. It highlights the advantages of the proposed volumetric approach, including its ability to represent directional uncertainty, incremental updating, and efficient space and time usage. The method is implemented with optimizations for real-time processing and is tested on various objects, demonstrating its effectiveness in generating watertight, high-detail models suitable for rendering and rapid prototyping. The algorithm is also extended to fill holes in the reconstructed surface by classifying voxels as unseen, empty, or near the surface. This allows for intelligent hole filling based on the spatial relationships between observed and unobserved regions. The method is evaluated on a range of objects, showing its robustness and ability to handle challenging cases such as thin surfaces and sharp corners. The results demonstrate that the method achieves high accuracy, with RMS distances between original and reconstructed surfaces of approximately 0.1 mm, indicating nearly optimal surface reconstruction. The paper concludes with a discussion of the algorithm's limitations and future directions, including potential applications to other scanning technologies and large-scale objects.This paper presents a volumetric method for integrating range images to reconstruct complex 3D surfaces with high detail and accuracy. The method addresses key challenges in surface reconstruction, including handling directional uncertainty, filling gaps, and robustness to outliers. The algorithm uses a cumulative weighted signed distance function to represent the surface, combining range images incrementally and efficiently. It employs run-length encoding for space efficiency and resampling to align with the voxel grid for time efficiency. The final surface is extracted as an isosurface from the volumetric grid, which is optimal in the least squares sense under certain assumptions. The method is capable of integrating up to 70 range images, producing high-resolution models with up to 2.6 million triangles. It handles gaps in the reconstruction by identifying empty regions and filling them with plausible surfaces. The algorithm is robust to sensor errors and can handle complex shapes, including those with sharp corners and thin surfaces. It also supports efficient processing through run-length encoding and synchronized scanline traversal. The paper discusses previous work in surface reconstruction, including methods based on unorganized points, parametric surfaces, and implicit functions. It highlights the advantages of the proposed volumetric approach, including its ability to represent directional uncertainty, incremental updating, and efficient space and time usage. The method is implemented with optimizations for real-time processing and is tested on various objects, demonstrating its effectiveness in generating watertight, high-detail models suitable for rendering and rapid prototyping. The algorithm is also extended to fill holes in the reconstructed surface by classifying voxels as unseen, empty, or near the surface. This allows for intelligent hole filling based on the spatial relationships between observed and unobserved regions. The method is evaluated on a range of objects, showing its robustness and ability to handle challenging cases such as thin surfaces and sharp corners. The results demonstrate that the method achieves high accuracy, with RMS distances between original and reconstructed surfaces of approximately 0.1 mm, indicating nearly optimal surface reconstruction. The paper concludes with a discussion of the algorithm's limitations and future directions, including potential applications to other scanning technologies and large-scale objects.