An Integrated Lifecycle-Management Pipeline for UAV-Augmented Smart Parking Surveillance Across the Cloud–Edge–End Continuum
Lopullinen julkaistu versio - 2.55 MB
Kadouma, A., Meliani, A., Li, Q., Shukla, A. K., & Elmusrati, M. (2026). An Integrated Lifecycle-Management Pipeline for UAV-Augmented Smart Parking Surveillance Across the Cloud–Edge–End Continuum. IEEE Access, 14, 131277-131296. https://doi.org/10.1109/ACCESS.2026.3725528
©2026 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Lataukset16
Pysyvä osoite
Kuvaus
The proliferation of unmanned aerial vehicles (UAVs) and Internet of Things (IoT) devices in urban monitoring creates a need for automated management of heterogeneous distributed computing environments. This paper presents the design, implementation, and experimental validation of a UAV-augmented smart parking surveillance system deployed across a cloud, edge, and end/far-edge computing continuum. The contribution is an integrated lifecycle-management pipeline—rather than a new standalone algorithm—in which an Ontology Semantic Reasoner (OSR) captures application intent and deployment constraints, a lifecycle manager (LiSO) deploys the microservices across cloud, edge, and end/far-edge clusters, a software-defined wide-area network (SD-WAN) controller manages runtime connectivity, and a forecasting-assisted policy maintains service continuity during UAV-triggered service hand-offs. The pipeline is evaluated on a real multi-site testbed spanning IONOS cloud in Germany and EURECOM edge/far-edge infrastructure in France. Deploying deep-learning inference at the far edge reduces total network traffic by at least 82% relative to cloud-centric processing. SD-WAN-based dynamic path selection increases video-analytics throughput from 53.7 to 114.9 frames per second (FPS), a gain of 61.2 FPS, relative to a single static path. A checkpointer operator integrated with LiSO characterizes stateful-migration saving and restoration overhead across state sizes of 10–500 MB and up to five concurrent instances, and the downtime and total migration time are reported across seven heterogeneous migration scenarios. To assess the energy effect of onboard inference, a Jetson-based power-consumption experiment was conducted using a 10-minute idle window and a 20-minute active DeepStream detection window. The active workload increased board input power from 5.307 W to 6.137 W on average, an increment of 0.830 W or 15.6%, indicating that migrating the workload away from the UAV-class node reduces compute-side energy demand. A fine-tuned time-series forecasting model with a tunable sensitivity parameter α drives proactive migration. The results provide experimental evidence for forecasting-assisted, descriptor-driven orchestration in multi-site IoT deployments.
Emojulkaisu
ISBN
ISSN
2169-3536
Aihealue
Kausijulkaisu
IEEE access|14
OKM-julkaisutyyppi
A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä (vertaisarvioitu)