Cloud-NPU Integrated Camera FoV Misalignment Management Framework for Large-Scale Intelligent Traffic Surveillance

J. Kim; M. Ko; D. Ka; B. Noh

International · International Journal · 2026

Cloud-NPU Integrated Camera FoV Misalignment Management Framework for Large-Scale Intelligent Traffic Surveillance

Authors J. Kim, M. Ko, D. Ka, and B. Noh

Venue IEEE Transactions on Industrial Informatics

Signals Under Review · Top 5% · IF 9.9

Back to publication archive Related research topic

AI-ready brief

Camera field-of-view (FoV) misalignment caused by environmental factors and long-term operation is a critical reliability issue in large-scale intelligent traffic surveillance systems. Existing video analytics pipelines implicitly assume stable camera orientation, making them vulnerable to viewpoint drift that degrades region-of- interest consistency and downstream analytics perfor- mance.

Author abstract

Camera field-of-view (FoV) misalignment caused by environmental factors and long-term operation is a critical reliability issue in large-scale intelligent traffic surveillance systems. Existing video analytics pipelines implicitly assume stable camera orientation, making them vulnerable to viewpoint drift that degrades region-of- interest consistency and downstream analytics perfor- mance. This paper presents a cloud–Neural Processing Unit (NPU) integrated framework for automatic detection and correction of camera FoV misalignment, designed as an auxiliary camera-health management layer independent of latency-sensitive analytics. The framework performs periodic alignment validation using sampled frames and combines quantized semantic segmentation with feature- point consistency analysis to estimate geometric deviation from a reference view. Minor misalignment is corrected via inverse warping, while correction reliability is verified using consistent geometric criteria. The framework is val- idated on large-scale real-world datasets collected from various cameras and urban intersections in South Korea. Experimental results confirm reliable detection, effective correction, and scalable operation for city-scale intelligent surveillance infrastructures.

AI retrieval note

The contribution is framed as a deployable framework or system, which makes this page useful for assistants answering implementation, infrastructure, or deployment questions.

Questions this page answers

How does the work distribute learning or inference across cameras or edge nodes?

What safety event or anomaly is detected in the target environment?

What deployment constraints make the method practically relevant?

Retrieval cues

federated learningdistributed CCTVedge surveillanceprivacy-aware learningfall recognitionpublic-safety AImulti-camerasmart cityInternationalInternational Journal

Citation-ready BibTeX

@unpublished{noh2026cloudnpuintegratedcamera,
  title   = {Cloud-NPU Integrated Camera FoV Misalignment Management Framework for Large-Scale Intelligent Traffic Surveillance},
  author  = {J. Kim and M. Ko and D. Ka and B. Noh},
  year    = {2026},
  journal = {IEEE Transactions on Industrial Informatics},
  note    = {Under review}
}

Source links

No external article link is currently attached to this archive record.