A Long-Term Operational Study of Edge-Based Video Analytics in Transportation Facilities

Abstract

Edge computing has been widely adopted for real-time video analysis in transportation facilities. Nevertheless, long-term operational data on such deployments remains limited. A video monitoring system was deployed across two railway stations and one bus terminal, consisting of 58 cameras and 14 edge servers. The system ran continuously for 11 months, performing passenger flow analysis and security screening. Hardware failures, network interruptions, and workload fluctuations were recorded. System logs show that peak processing loads occurred during national holidays, with inference queues increasing by up to 240%. Network congestion caused intermittent frame loss during evening rush hours. After introducing adaptive buffering and local caching, average system availability improved from 96.4% to 99.1%. Operational reliability depends strongly on environmental and organizational factors beyond algorithmic performance.