Opportunistic estimation of television audience through smartphones

Abstract

Television audience estimation is an important task for advertisement placement. In this paper we present a system based on a client-server architecture able to recognize a live television show. The clients are implemented throughout smartphones, thus a well known audio fingerprint algorithm has been modified to fit smartphone needs. To reach this aim, the optimization of an ad-hoc cost function has been introduced. Server and client are generally far from being synchronized, due to the variety of broadcasting media (aerial, satellite or streaming over the Internet) leading to different and unpredictable delays. For this reason, we present a new likelihood estimate designed to overcome the lack of synchronization issue.