YOLOv12 - Powered Smart Retail Automation: A Comparative Performance Analysis of Model Variants for Checkout Systems

Javed Khan; Anushka Nagpal

doi:10.69968/ijisem.2026v5i3131-135

Authors

Javed Khan Department of Computer Science and Engineering, Haryana Engineering College, Jagadhari, Kurukshetra University, Kurukshetra, Haryana, India
Anushka Nagpal Department of Computer Science and Engineering, Haryana Engineering College, Jagadhari, Kurukshetra University, Kurukshetra, Haryana, India

DOI:

https://doi.org/10.69968/ijisem.2026v5i3131-135

Keywords:

YOLOv12, smart retail, automated checkout, object detection, comparative performance analysis, real-time invoicing, grocery verification

Abstract

Automated checkout is increasingly viewed as a transfor-mative solution for improving operational efficiency, reduc-ing revenue loss, and enhancing customer experience in re-tail. While deep-learning-based object detectors offer strong potential for retail automation, deploying them on resource-constrained edge hardware requires a careful evaluation of ac-curacy, computational cost, and storage footprint. This paper presents a YOLOv12-powered smart retail checkout system and reports a comparative evaluation of its Nano (YOLOv12n) and Small (YOLOv12s) variants. A custom dataset of 7,637 annotated images (6,000 train / 909 validation / 728 test) span-ning five grocery categories was used to train and evaluate both models. Results show that YOLOv12n, at only 5.27 MB, matches YOLOv12s (18.06 MB) on mAP@0.5 (99.5%) and achieves higher precision (99.7% vs. 99.3%), while training 24% faster and requiring roughly 71% less storage; YOLOv12s edges ahead only on mAP@50–95 (0.9123 vs. 0.9057), a 0.66% margin. Based on this trade-off, YOLOv12n was selected as the deployment model. The selected model was integrated into a real-time invoicing pipeline that uses a virtual trigger-line (Temporal Crossing Detection) mechanism over a conveyor-belt video feed to prevent duplicate/missed billing, retrieve product information, and automatically gen-erate itemized PDF invoices. The results demonstrate that a lightweight YOLOv12 model can deliver near state-of-the-art detection accuracy while remaining practical for low-cost, edge-based autonomous checkout deployment.

References

[1] M. A. Anusuya et al., “BillSmart: An Automated Gro-cery Billing System Using YOLO,” Int. J. Sci. Res. Eng. Manag. (IJSREM), vol. 9, no. 5, 2025.

[2] L. Tan et al., “Enhanced Self-Checkout System for Retail Based on Improved YOLOv10,” arXiv preprint, 2024.

[3] S. M. Shelke et al., “Handsfree Checkout Using YOLO Object Detection,” Int. J. Sci. Res. Eng. Manag. (IJS-REM), 2025.

[4] S. K. Oishi et al., “Enhancing Retail Checkout Efficiency Through a Hybrid YOLOv8-Based Grocery Detection and Billing System,” Front. Comput. Sci. Artif. Intell., 2026.

[5] R. Sapkota et al., “Comprehensive Performance Evalua-tion of YOLOv12, YOLO11, YOLOv10, YOLOv9 and YOLOv8,” AI Open, 2025.

[6] Z. Jiang, “YOLOv5-based Intelligent Detection Method for Retail Goods,” Inf. Technol. Control, vol. 54, no. 2, pp. 504–519, 2025.

[7] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” in Proc. IEEE CVPR, Las Vegas, NV, USA, 2016, pp. 779–788.

[8] M. Zarkoosh, “Efficient YOLOv12 Variants: Pruned Models Optimized for Object-Size Distribution,” 2025.

[9] J. L. Satore, M. Fernandes, R. Costa, and A. Silva, “Com-parative Study of YOLOv10, YOLOv11 and YOLOv12 Lightweight Models for Multi-Class Maritime Search and Rescue Using UAV Imagery,” 2025.

[10] O. D. Kurniawan and E. H. Rachmawanto, “YOLOv12 Based on Stationary Vehicle for License Plate Detec-tion,” J. Appl. Inform. Comput., vol. 9, no. 5, 2025.

[11] R. Sapkota, Z. Meng, M. Churuvija, X. Du, Z. Ma, and M. Karkee, “Comprehensive Performance Evalua-tion of YOLOv12, YOLO11, YOLOv10, YOLOv9 and YOLOv8 on Detecting and Counting Fruitlet in Complex Orchard Environments,” 2025.

[12] M. A. R. Alif and M. Hussain, “YOLOv12: A Break-down of the Key Architectural Features,” 2025.

[13] Y. Tian, Q. Ye, and D. Doermann, “YOLOv12: Attention-Centric Real-Time Object Detectors,” 2025