Color-Enhanced Local Feature Fusion for FPFH-Based Point Cloud Registration

To address the limitations of the traditional Fast Point Feature Histogram (FPFH) in heterogeneous building point cloud registration-over-reliance on geometric information, neglect of color cues, and susceptibility to lighting variations-this paper proposes a color-enhanced local feature fusion framework. First, the RGB color space is transformed to HSV, with histogram equalization applied to the Hue channel to improve illumination robustness and color discriminability. Then, normalized FPFH geometric features and PCFH color features are fused into a 44-dimensional joint vector. Experimental validation on real shopping mall indoor point clouds shows that the proposed method performs well in complex scenarios: RRE and RMSE are reduced by up to 74.9% and 74.6% under different initial poses; robust registration is achieved in asymmetric point-count scenarios where FPFH fails; and RMSE remains 27.73% lower than FPFH even with only 500 sampling points. An outdoor campus road experiment further verifies generalizability, with RRE, RTE, and RMSE reduced by 66.2%, 63.4%, and 59.8%, respectively. Overall, integrating compact color features introduces negligible computational overhead while improving registration accuracy and robustness.