Scanning Path Planning Method for Three-Dimensional Reconstruction of Large-Scale Structural Components

To address the problems of extensive coverage blind spots, high path costs, and the susceptibility of traditional algorithms to local optima in the scanning of large structural component surfaces, a scanning path planning method for 3D reconstruction of large-scale structural components was proposed. First, a visibility cone measurement model for a line laser scanner was constructed, defining the scanner's physical constraints (measurement inclination, field of view, depth of field) and collision constraints, providing fundamental constraints for path planning. Second, based on the dynamic curvature features of the component's CAD model, free-form surface discretization was achieved. Initial sampling points were generated by dense sampling in high-curvature areas and sparse sampling in low-curvature areas. An initial viewpoint network was constructed by offsetting along the normal vectors of these sampling points by the optimal working distance. This network was then refined using a regional growth clustering method constrained by the visibility cone, and simultaneously a collaborative cost model integrating positional distance and orientation differences was established. Finally, an improved marine predators algorithm (MPA) was proposed for path optimization: a hybrid initialization strategy was adopted to enhance initial solution quality, a Weibull-based three-stage motion was introduced to strengthen global exploration capability, and a non-linear adaptive step size was designed to balance exploration and exploitation. This was combined with a two-level collision detection scheme (rapid screening based on the separating axis theorem—precise verification based on KD-tree nearest neighbor search) and an improved 2-Opt algorithm (incorporating posture continuity) for path smoothing. Using a car hood measuring 1660mm×1070mm×170mm as experimental object, the improved MPA was compared with simulated annealing, genetic algorithm, and ant colony optimization. The results show that the improved MPA generated 49 viewpoints, achieving 100 coverage with no collision risk and the lowest comprehensive path cost; the computation time was 514.73s; compared to simulated annealing (1121.44s) and the genetic algorithm (574.98s), it demonstrated more stable convergence and offered higher solution quality than the ant colony optimization algorithm. The proposed method can efficiently meet the requirements for high-precision, full-coverage scanning of large-scale structural component surfaces.