Scalable Edge-Based Architecture for Real-Time Video Analytics in Smart Transportation Systems

Abstract

The expanding pressure of urban activility and congestion has led to the evolution of intelligent transportation systems (ITS) that are taking recourse to real time video ways to either be more efficient in their operations and maintain public safety. In this research, a scalable edge-based design that could facilitate distributed and low latency video processing is presented to power smart transport networks. The new system will implement the containerized modules of analytics which will be deployed on edge nodes placed at any intersection points or roadside units to provide on-site detection of objects, and mult-object tracking. The hybrid video processing pipeline is used such that it incorporates convolutional neural networks (CNNs) and lightweight tracking algorithms (e.g., DeepSORT) in order to provide high performance but be efficiently computable on resources-limited devices. In order to assess the performance of the system, the task of testing latency, throughput, and scalability was realized using real-world traffic video sets. As experimental results demonstrate, there is a 45 percent end-to-end latency reduction with a 60 percent reduction in a bandwidth used in the cloud as opposed to centralized cloud processing models. The architecture was also proved to be invariant to object detection and frame processing rate when faced with greater camera loads. The study validates the possibility of implementing edge-oriented intelligence in intelligent transportation systems to allow incidents to be detected and identified quicker, rely less on cloud systems and scale better. The suggested framework has given the future edge-to-cloud integrated ITS deployment an initial framework, which requires real-time response and resource optimization efficiency.