Collaborative Intelligent Cross-Camera Video Analytics at Edge: Opportunities and Challenges

TL;DR

This paper explores a collaborative edge-based video analytics approach where cameras share insights to improve accuracy and efficiency, addressing computational and bandwidth challenges in large-scale surveillance systems.

Contribution

It proposes a novel collaborative paradigm for cross-camera analytics at the network edge, enhancing accuracy and reducing latency and bandwidth usage.

Findings

Collaborative approach improves detection accuracy.

Reduces inference latency and bandwidth consumption.

Addresses challenges in resource-constrained edge environments.

Abstract

Nowadays, video cameras are deployed in large scale for spatial monitoring of physical places (e.g., surveillance systems in the context of smart cities). The massive camera deployment, however, presents new challenges for analyzing the enormous data, as the cost of high computational overhead of sophisticated deep learning techniques imposes a prohibitive overhead, in terms of energy consumption and processing throughput, on such resource-constrained edge devices. To address these limitations, this paper envisions a collaborative intelligent cross-camera video analytics paradigm at the network edge in which camera nodes adjust their pipelines (e.g., inference) to incorporate correlated observations and shared knowledge from other nodes' contents. By harassing redundant spatio-temporal to reduce the size of the inference search space in one hand, and intelligent collaboration between video nodes on the other, we discuss how such collaborative paradigm can considerably improve accuracy, reduce latency and decrease communication bandwidth compared to non-collaborative baselines. This paper also describes major opportunities and challenges in realizing such a paradigm.