Flexible 3-D Measurement With Double Homocentric Spherical Feature

3D reconstruction is a key topic in the geometrical measurement studies and industrial applications. To address the limitations in measurement flexibility, a target-based measurement (TBM) with the double homocentric spherical feature is proposed for the 3D vision-based measurement (VBM). A dedicated dataset is constructed for this target, and a single shot multi-box detector (SSD) is employed to automatically extract multi-scale target regions. The Adaptive Density-Based Spatial Clustering of Applications with Noise (ADBSCAN) is employed to cluster the candidate points that belong to the different concentric circles of the double homocentric spherical feature. To enhance the accuracy of feature detection and subsequently optimize measurement precision, the bimodal Gaussian distribution is proposed for the concentric circles to decrease the time required for extracting the centers of the concentric circles. Furthermore, according to sub-pixel double homocentric spherical feature, the objective function and measurement model are established in special Euclidean group (3) (SE(3)). The experiments are conducted to validate the effectiveness and performance of the measurement. The average error of different measurement distances and reference distances of the double spherical measurement is 1.452 mm, which is superior to the three comparison methods and verifies its high accuracy and strong robustness in 3D measurement.