Classification at the Edge: implementation and performance evaluation

Abstract

Modern mobile computational hardware poses a threat to classic Cloud Computing technology. Each year mobile devices become more affordable and at the same time more powerful. Thus Edge Computing ─ a new paradigm to utilize those advantages was developed. Studies show that the most demanded nowadays is an intelligent image analysis. Edge Computing allows this task to be performed at the Edge, ensuring security, effective energy consumption, and realtime execution. This article discusses the classifier based camera resolution selection technique for adaptive distributed image stitching. Implementation on cloud and seamless migration to the edge using Docker containers are described. Performed laboratory experiments and their results confirm performance improvement when image stitching is implemented on the dedicated mobile hardware.