3D Shape Features for Colorless Camera Pose Estimation

The patent introduces shape feature values as an intermediary representation that bridges the gap between 3D mesh data and image matching. Instead of directly using color information from the mesh, the system extracts shape features (such as surface normal vectors, curvature, and geometric descriptors) that can be projected onto the image plane and used for similarity calculation with photographed images. This intermediary approach enables accurate estimation without requiring color information in the 3D data.

The patent replaces the traditional color-based matching mechanism with a shape feature-based mechanism. Instead of relying on texture/color information from the 3D mesh to match with the photographed image, the system substitutes this with geometric shape features that are extracted from the mesh structure and projected onto the image plane. This substitution allows the system to function effectively with 3D data that lacks color information while maintaining estimation accuracy.

The patent extracts only the necessary shape feature information from the 3D mesh data, separating this from the color information. By extracting shape features such as surface normals, curvature values, and geometric descriptors, the system creates a simplified representation that can be projected onto the image plane. This extraction approach reduces the complexity of processing while maintaining the essential information needed for accurate position and posture estimation.

The patent segments the 3D mesh data into distinct shape feature components that can be independently processed and projected. Instead of processing the complete mesh data with color information, the system divides the data into geometric features (vertices, edges, faces, normals) that can be selectively extracted and transformed. This segmentation allows for more efficient processing and reduces the computational burden while preserving estimation accuracy.