Poster Abstract: On-the-Fly Autoencoders as Subscriptions for Face Detection on a Publish/Subscribe System

Abstract

This paper presents a publish/subscribe system that accepts sample face images from the subscribers to autoencode a person in emergent pursuit or tracking. Autoencoders trained on the fly based on a convolutional neural network are deployed to publish/subscribe brokers in the form of subscriptions. Face captures sent from the surveillance cameras are injected into the autoencoders to detect the person in search. We discuss the matching accuracy and the overhead of the training process to obtain the autoencoders. © 2022 Owner/Author.