Characterizing the Computational Feasibility of AI Video Analytics Across Edge, Fog, and MEC in Urban Intelligent Transportation Systems
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Urban Intelligent Transportation Systems increasingly rely on AI-based video analytics to support cooperative and safety-related services. In urban living labs, many applications are built on a shared AI perception pipeline—comprising object detection, multi-camera aggregation, and trajectory prediction—fed by distributed roadside cameras. In this context, a fundamental question arises: where should computation be placed to meet strict end-to-end latency requirements, typically below 100ms.This paper addresses this question through an experimental study conducted in the Modena Automotive Smart Area (MASA), a real-world urban living lab featuring Edge, Fog, and Multi-access Edge Computing (MEC) nodes. Since network latency in MASA is negligible, the analysis focuses exclusively on computational latency and scalability.We evaluate two representative applications based on the same AI video pipeline: AI-CAM, targeting cooperative perception at moderate frame rates, and Urban ADAS (U-ADAS), which requires higher frame rates and stricter latency constraints. Experimental results show that neither fully Edge-based processing nor naïve centralization on MEC nodes can consistently meet latency targets when multiple camera streams must be aggregated. Edge devices lack sufficient computational capacity, while MEC-centric solutions face scalability limits under concurrent workloads.We demonstrate that Fog nodes enable an effective trade-off, supporting aggregation from 4–6 cameras while maintaining sub-100ms processing latency. These results highlight the importance of hybrid Edge–Fog–MEC architectures for scalable, low-latency AI services in urban ITS deployments.
Fields of Study & Subjects
IoT and Edge/Fog Computing
Advanced Neural Network Applications
Autonomous Vehicle Technology and Safety
Publication details
- Journal / source
- IEEE Vehicular Technology Conference
- Pages
- 1–6
- Publication date
- 1 June 2026
- Language
- EN