Systematic Review of Motion Error Compensation Methods in Phase-Shifting Profilometry

Phase-shifting profilometry （PSP） is a high-precision three-dimensional （3D） imaging technique capable of acquiring industrial-grade point clouds. However， its application to dynamic measurements is significantly affected by motion-induced errors. These errors are primarily classified into two categories： ghosting artifacts caused by in-plane motion and ripple-like distortions caused by motion along the line of sight. This paper presents a systematic review of motion error compensation methods developed for dynamic PSP in recent years. First， the mechanisms responsible for these two types of errors are explained. Existing compensation methods are then classified into three categories. The first category addresses ghosting artifacts through motion estimation and image registration based on feature matching， optical flow， or frequency-domain analysis. The second category focuses on ripple-like distortions by correcting motion-induced deviations in phase shifts； these methods are further divided into non-pixel-wise approaches， which rely on global or neighborhood information， and pixel-wise approaches， which preserve pixel independence. The third category comprises unified motion models that aim to simultaneously compensate for both ghosting artifacts and ripple-like distortions. The review also compares the applicability of these methods to different motion patterns and identifies two major limitations of existing approaches. First， most methods compensate for only one type of motion-induced error， which restricts their applicability. Second， many methods rely on assumptions of globally uniform motion， making them unsuitable for complex dynamic scenes involving multiple moving objects or nonrigid deformation. Finally， future research directions are discussed， including the development of integrated compensation frameworks that address both types of motion-induced errors， the replacement of global motion assumptions with pixel-wise or locally adaptive motion estimation， and the use of deep learning and virtual simulation data to enable end-to-end motion error compensation.