Networked Cameras Are the New Big Data Clusters

The increasing complexity of deep learning and massive deployment of cameras at the edge have drastically increased the resource demand of edge data analytics. Compared to traditional Internet web applications, such resource demand (in computing, storage and networking) is not limited by millions of human users, but rather the continuous activities of billions of sensors. This paper presents the abstraction of camera cluster as an attempt to address this challenge in the context of video analytics. We envision a novel analytics stack that orchestrates the computing resource of massive networked cameras to enable efficient edge video analytics.