A Novel Hybrid Method for Registering Partially OverlappedPoint Clouds

Point cloud registration plays a crucial role in ۳D computer visionapplications by aligning multiple views of an object into a commoncoordinate system. While traditional methods like the Iterative ClosestPoint (ICP) algorithm offer precise local registration, they suffer fromsensitivity to noise and outliers. Global methods such as the ۴-PointCongruent Sets (۴PCS) algorithm are more robust but computationallyexpensive. This paper presents a novel hybrid algorithm that combinesRestricted ۴PCS (R-۴PCS) and Accelerated ۴PCS techniques to achieveboth computational efficiency and registration accuracy. The proposedmethod refines candidate point selection, limits the search space forfaster matching, and incorporates angular filtering to eliminate falsecorrespondences. Through experiments on several datasets with varyingoverlap ratios, the hybrid approach demonstrates superior performancein terms of speed and accuracy, particularly in challenging low-overlapscenarios. This algorithm offers a promising solution for real-time ۳Dpoint cloud registration in diverse applications, including autonomousnavigation, object recognition, and industrial inspection.

کلیدواژه ها:

۳D point cloud ، registration ، low-overlap ، ۴PCS