Measurement-constrained feature clustering for automated structured-light viewpoint planning

Automated structured-light 3D measurement of complex industrial parts requires efficient viewpoint planning for accurate feature acquisition. However, existing methods mainly focus on surface coverage and lack feature-oriented region partitioning, resulting in redundant viewpoints and insufficient observation of geometric features. This paper proposes a measurement-constrained feature clustering-based viewpoint planning framework for automated structured-light measurement. A two-stage clustering strategy is developed to partition feature sampling points into measurement-compatible regions by jointly considering normal consistency and sensor field-of-view constraints. Spectral clustering based on normal-vector cosine similarity is first applied to group points with consistent observation directions, followed by agglomerative hierarchical clustering with spatial distance constraints to generate compact regions suitable for viewpoint planning. Based on the clustering results, candidate viewpoint generation, visibility evaluation, viewpoint selection, and robot path optimization are performed to obtain the final measurement trajectory. Experiments on an automotive side panel with 55 geometric features show that the proposed method achieves reliable feature reconstruction and dimensional measurement using only 76 optimized viewpoints.