Rapid point cloud registration method based on low-dimensional feature constraint and median absolute deviation

Automatic point cloud registration is essential for 3D scanning data processing, yet feature-based methods face computational and robustness limitations with dense clouds. This paper presents an efficient framework combining 8D feature histograms for reduced complexity and Median Absolute Deviation (MAD) for robust estimation. The method constructs low-dimensional feature descriptors, computes transformations using the Bursa-Wolf model, and integrates MAD to mitigate outlier effects. Experiments demonstrate significant acceleration through dimensionality reduction while preserving accuracy using MAD-based outlier rejection. The approach effectively balances computational efficiency and registration precision, addressing critical challenges in dense point cloud alignment.