A Novel Point Cloud Registration Method with Error-Guided Threshold Adjustment Mechanism

In point cloud registration, traditional Iterative Closest Point (ICP) algorithms often fail in low-overlap scenarios and under noise interference. This paper proposes an improved point cloud registration method based on Error-Guided Threshold Adjustment (EGTA) mechanism, aiming to enhance the accuracy and robustness of registration. In the coarse registration stage, the Fast Point Feature Histograms (FPFH) and the Random Sample Consensus (RANSAC) algorithm are combined to select distinctive corresponding points through random sampling and the application of a triangle similarity constraint, providing an initial estimation of the pose transformation between point clouds. In the fine registration stage, the proposed EGTA mechanism dynamically refines the distance threshold based on matching error in each iteration, ensuring that each point in the source point cloud is matched only with the nearest point in the target point cloud that lies within the threshold, filters out invalid correspondences. The method was validated on several public point cloud datasets, including complex geometric models and large-scale scenes. Experimental results demonstrate that the proposed method significantly improves registration accuracy and maintains robust performance in low-overlap and noisy scenarios.