Related work

Citations supporting the Phase I paper’s Methods, Validation, and Limitations sections. Extracted from the round-1 Agent D ML/CV bucket report (2026-05-10), with two additions flagged by round-2 Critique G.

Each entry is tagged:

• [verified] — URL or DOI confirmed in a WebSearch result on 2026-05-10.
• [needs-lit-review] — the work plausibly exists (search returned a reference or stub) but the full citation must be fetched and read before paper submission.

Direct prior art (CV-on-factory-floor cycle / activity measurement)

1. “Enhancing Garment Sewing Employee Added Value Monitoring Using a Video-Based Spatio-Temporal Approach Through Computer Vision” — IEEE Conference Publication, document 10652738 (2024). [verified] Directly adjacent: spatio-temporal worker-activity recognition on garment sewing lines; value-added vs non-value-added classification. Methodology for spatio-temporal models is a Phase II reference.

2. “Cycle Time Measurement Using AI-Based Object Detection and Tracking in Industrial Processes” — Springer 2024, link.springer.com. [verified] YOLOv8 + ByteTrack for cycle-time measurement in industrial processes — direct Phase I methodological parallel. Their tracker is heavier than our IoU same-station rule; comparison is paper-worthy (see cycle-event-detection §5).

3. “Automated work cycle classification and performance measurement for manual work stations” — Robotics and Computer-Integrated Manufacturing, 2017, ScienceDirect S0736584516303787. [verified] Multi-camera tracking of operator movement; clustering and classification of work cycles; abnormal-cycle detection. Direct prior art for the cycle_events_rejected taxonomy.

4. “Improving Time Study Methods Using Deep Learning-Based Action Segmentation Models” — Applied Sciences (MDPI), 2024, doi:10.3390/app14031185. [verified] Replaces stopwatch time study with action segmentation. Closest match to our validation problem; pull and align metric definitions with validation-methodology §2.

5. “SEWAbility: Computer-Vision Framework for Sewing-Line Worker Activity Monitoring” — PMC13031723 (2024–25). [verified, round-2 addition] Round-2 Critique G flagged this as missing prior art. Sewing-line-specific worker activity monitoring; closest published peer to Phase I scope. Must read and cite before submission.

6. Springer 2025 — edge-computing sewing-machine state recognition — Springer 2025. [needs-lit-review, round-2 addition] Round-2 Critique G flagged this as missing prior art. Edge-deployed CV for sewing-machine state recognition. Locate the exact citation (Critique G has the lead) and add to the methods comparison.

Lean / IE framing (paper-contribution positioning)

6a. “Computer Vision-Based Work Productivity Optimization with Lean Principles and Simulation Method” — Springer 2025, doi:10.1007/978-3-031-95421-4_7. [needs-lit-review, round-2 addition] Round-2 Critique G’s recommended framing anchor. Positions CV-derived productivity measurement inside the Lean / IE literature rather than the pure CV literature — directly supports the Phase I paper’s “CV pipeline feeding a Lean visibility loop” angle in paper-plan § Claim. Confirms LBZF’s lean operating model (see current-workflow) is the natural framing for the Phase I contribution claim. Pull full text and align terminology before submission.

Action recognition / assembly-line behavior (Phase II relevance)

7. “Supervised and Self-Supervised Learning for Assembly Line Action Recognition” — J. Imaging (MDPI), 2025, PMC11766131. [verified] Self-supervised pretraining for assembly-line action recognition. Phase II reference more than Phase I.

8. “YOLO-HRNet with Attention Mechanism: For Automated Ergonomic Risk Assessment in Garment Manufacturing” — Applied Sciences (MDPI), 2025, doi:10.3390/app152412950. [verified] Apparel-line CV with occlusion-aware attention. Phase II reference.

9. “Hand Gesture Recognition of Methods-Time Measurement-1 Motions in Manual Assembly Tasks Using Graph Convolutional Networks” — Taylor & Francis, 2021, doi:10.1080/08839514.2021.2014191. [verified] Pose-based MTM-1 gesture recognition. Phase II relevance for behavioral monitoring.

Edge-deployment / hardware-anchoring references

10. “Benchmarking YOLOv8 Variants on Jetson Orin NX for Edge Computing” — MDPI Computers 15(2):74. [verified] Anchors the inference-latency targets in cycle-event-detection §G6 and the Phase II compute argument in phase-ii-preview.

11. “YOLO Object Detection for Real-Time Fabric Defect Inspection in the Textile Industry: A Review of YOLOv1 to YOLOv11” — Sensors / PMC11990965 (2025). [verified] Out-of-scope for Phase I (defect detection, not cycle counting) but cited for textile-industry YOLO context.

Agreement and validation methodology

12. Bland & Altman (1986), “Statistical methods for assessing agreement between two methods of clinical measurement” — The Lancet, pubmed.ncbi.nlm.nih.gov/2868172. [verified] Foundational reference for the agreement methodology in validation-methodology §2. Always cited alongside Lin’s CCC.

13. Lin L. I-K. (1989), “A concordance correlation coefficient to evaluate reproducibility” — Biometrics. [needs-lit-review] Original CCC paper. Must be pulled and read before paper submission.

14. Hawthorne-effect literature on observed manual labor — search-term anchor only. [needs-lit-review] Required to defend the “CV outperforms stopwatch” pivot rule in validation-methodology §3.

Legal / consent framework

15. Colombian Law 1581 of 2012 + Decree 1377 of 2013 — Colombian personal-data-protection framework. [verified] Referenced in validation-methodology §5.1; the informed-consent procedure derives from this.

16. Ius Laboris, “An international view on video surveillance of employees” — Ius Laboris. [verified] Grounds the “visible notice on the floor” requirement in validation-methodology §5.1.

Further reading (cited but not yet fetched)

• “A vision-based framework for human behavior understanding in industrial assembly lines” — arXiv 2409.17356 (2024). [needs-lit-review]
• “Graph-based framework for temporal human action recognition and segmentation in industrial context” — Engineering Applications of AI, 2025, S0952197625017129. [needs-lit-review]
• “Monitoring of Assembly Process Using Deep Learning Technology” — Sensors 20(15):4208 (2020). [needs-lit-review]
• CarDA dataset and benchmark for in-factory action recognition — [needs-lit-review] Would anchor an open-dataset comparison paragraph in the paper.

Gaps Agent D flagged for follow-up

• The exact statistical pattern (lognormal / Weibull?) of operator cycle durations in apparel manual work — affects the MAX_CYCLE_DURATION choice in cycle-event-detection. Lives in IE / ergonomics literature outside CV proper.
• Recent (2025–2026) deployment case studies from Form AI / Drishti / Retrocausal / SymphonyAI. Andrew may know which have landed.
• Argentinian Law 25.326 personal-data framework, for the ITBA install. Briefly searched; not yet quoted in validation-methodology §5.