GFMS: An automatic system for gap and flush measurement in automobile assembly seams based on 3D vision

The gap and flush (G&F) of automobile assembly seams are important indicators that directly determine the exterior quality and performance. Traditional manual inspection methods fail to meet modern demands for efficiency and accuracy. Although point cloud-based G&F detection technology demonstrates excellent accuracy characteristics, it faces challenges in complex assembly line environments caused by noise, redundant data, and inter-seam interference. To address these problems, we design an automatic G&F measurement system (GFMS) that integrates point cloud data acquisition and analysis. Then, we propose an adaptive optimization method based on the spatial distribution of adjacent frame line point clouds, achieving adaptive segmentation of seam regions through spatial density and local geometric constraints, multi-level filtering for outlier removal and surface smoothing, and Gaussian-weighted filling for missing regions. Finally, we propose a G&F analysis method based on interference suppression. A weighted voting mechanism with growing clustering is introduced to eliminate interference points between seams. A prior constrained circle fitting is adopted to reduce errors caused by missing fillet profiles. The final G&F are obtained by averaging the calculation results of all frame point clouds. Experimental validation on standard blocks demonstrated maximum deviations of 0.006 mm (gap) and 0.004 mm (flush). The GFMS exhibited a maximum deviation of 0.02 mm in both G&F measurements when compared with the ground truth (CMM) standards during automobile seams G&F validation. Compared with traditional measurement methods, the proposed method greatly improves accuracy and efficiency. • An automatic measuring system for gap and flush of automobiles is designed. • An adaptive optimization method is employed to eliminate the seam irrelevant regions. • An accurate gap and flush analysis method based on interference suppression is proposed.