Task-Oriented Multi-View Acquisition Planning for Visual Inspection of Panel Furniture

Automated multi-view acquisition is widely used in industrial production but remains underexplored for furniture inspection. This study proposes a task-oriented multi-view acquisition planning method for panel furniture. The method constructs a hierarchical inspection-surface model from furniture geometry and inspection requirements, evaluates target visibility by one-sided triangle-mesh ray casting, and iteratively refines the candidate pool based on coverage residuals. Viewpoints for must-inspect surfaces are selected using a quality-aware marginal score, with their number determined by comparing local pose replacement with continued viewpoint addition. Supplementary viewpoints for optional surfaces are ranked by coverage gain, and their number is determined using the L-method. Visit sequences are optimized using ant colony optimization and 2-opt under a hard keep-out-zone constraint. For four furniture models, 19, 23, 30, and 72 viewpoints achieved must-inspect-surface coverage rates of 100.00%, 100.00%, 99.86%, and 98.16%, respectively. For matched numbers of must-inspect-surface viewpoints, 25° effective coverage exceeded the 20-run means of genetic algorithm and binary particle swarm optimization baselines by 1.19–13.16 and 20.09–36.72 percentage points, respectively. Under fixed total viewpoint counts, local pose replacement increased must-inspect-surface coverage by 0.10–1.72 percentage points. Across 20 paired trials per model, keep-out-zone-constrained sequencing produced zero intersecting edges while increasing path length by 1.33–6.91%. These results demonstrate the effectiveness of the proposed method for front-end multi-view acquisition planning in panel furniture inspection.