An Identify-Then-Eliminate-Based Pixel Correspondence Establishment Method for Fringe Projection Profilometry Against Interreflection

Widely applied fringe projection profilometry faces significant challenges when measuring objects with strong inter-reflections. The existence of inter-reflections results in the overlap of same-frequency fringes, which severely affects the establishment of the pixel correspondence. Existing methods typically quantify inter-reflection errors pixel by pixel, which is unnecessary for pixels free from inter-reflections. In this paper, an identify-then-eliminate based method to establish pixel correspondence is proposed against surface inter-reflections. Different from current methods, it first identifies severely affected pixels and then eliminates inter-reflection errors in a targeted manner. Specifically, in pixel identification stage, dual-frequency phase-shifting method is utilized and an error model is constructed to accurately identify pixels with notable inter-reflection errors. Then in error elimination stage, reverse-sequence frequency-phase joint encoding patterns are designed to facilitate the separation of inter-reflections. A single-pixel encoding neural network is proposed to obtain accurate pixel correspondence with captured two-stage data. Experiments on industrial parts with strong surface inter-reflections illustrates that compared to existing methods, our method obtains high-precision, continuous, and realistic point cloud data. Additionally, our method also maintains the efficiency of measuring objects with strong inter-reflections.