Multi-object geometric tracking of boxes in packing operations: Effects of prior knowledge and camera velocity

Accurate tracking of multiple boxes in 3D space is essential for implementing augmented reality (AR)-assisted manual packing tasks. This work presents a vision-based tracking system that estimates the size and position of boxes using egocentric RGB-D input, promoting human-centric automation aligned with Industry 5.0 principles. The system reconstructs boxes by clustering plane segments from point clouds, enabling simultaneous size and position tracking. Unlike prior studies focused on assisting cargo placement during manual packing procedures, this work addresses an earlier operation: identifying and selecting the correct item to be packed from a consolidation zone. The study examines the influence of prior knowledge— predefined box dimensions—on tracking performance under varying camera velocities. A tolerance-based error function is proposed to assess accuracy in alignment with industrial requirements. A two-factor experimental design evaluates the effects of algorithm configuration (with or without prior knowledge) and camera velocity (low or high). Analysis using the 6D-ViCuT dataset reveals that both prior knowledge ( p = 0 . 000 ) and camera velocity ( p = 0 . 038 ) significantly impact precision and F1-score. Incorporating prior knowledge leads to an 8.56% increase in precision and a 5.00% increase in F1-score, primarily by reducing false positives. Recall remains statistically unaffected. Lower camera velocities consistently improve tracking performance, with maximum observed precision and recall values of 74.83% and 74.71%, respectively. These findings highlight the system’s potential to improve usability and efficiency in AR-assisted logistics. Future work will explore real-time deployment, analyze the effect of box proximity, and extend evaluations over work cycles to assess impacts on logistic performance indicators and operator workload. • Geometric method to estimate size and position of multiple boxes from plane segments. • Improved tracking performance through integration of prior box size information. • Explore AR-based assistance for item picking in consolidation zone of packing units. • Novel validation of Box Tracking system under realistic egocentric packing conditions. • New error function aligned with industrial tolerances for logistic applications.