How ROI Similarity Solves Point Cloud Noise: A New Standard for Change Detection

Global, Source:

Laser point cloud change detection has faced persistent hurdles due to spatial and structural uncertainties. Residual registration errors, uneven point cloud density, viewpoint shifts, and object occlusion frequently lead to false alarms, making accurate monitoring difficult. To address these challenges, researchers have developed a novel Region of Interest (ROI)-level 3D point cloud change detection method based on keypoint similarity.

The proposed method begins by filtering potential change points using point-to-point distances. Regions of Interest are then generated through RBNN clustering, converting global point cloud comparison into local comparison among ROI units. Keypoints within these ROIs are extracted for surface reconstruction, allowing for the identification of effective overlapping regions between paired ROIs. Regional similarity is constructed by fusing surface normal consistency and dynamic-evolution features of complex networks for change discrimination.

Experiments conducted on simulated mine-tunnel data and field-measured underground-garage data demonstrate the method's effectiveness. Under unseen background disturbances, including residual registration error, occlusion, and uneven point cloud density, the proposed method achieved the highest F1-score of 0.9004 for the simulated tunnel scenario compared with competing methods. It delivers effective false-alarm suppression without degrading detection recall for real-change regions.

This advancement is particularly significant for industries relying on precise 3D scanning, such as mining, construction, and autonomous navigation. By focusing on local similarities rather than global comparisons, the method reduces noise and improves accuracy, offering a robust solution for real-world applications where environmental conditions are often unpredictable.