Beyond The Kiss-Cam: Measuring The Fan With Computer Vision Based Analytics

What are fans really doing at the game? When are they watching the action on the field/court, and when are they buried in their phone or tablet? Which ads do they notice on the JumboTron, and which ones do they ignore? Who’s joining in the wave and who’s not? Are they having fun? If you are a club trying to retain seat-buying fans, fill empty chairs, or effectively use an advertising budget, then these are important questions to answer. Video surveillance tools can help, but sifting through crowd footage does not scale. In this paper, we describe a computer vision-based software method that goes beyond just capturing videos. We propose a machine learning-based framework that can automatically recognize certain kinds of fan activity. Our technique is based on convolution neural networks and it leverages crowd-sourcing to inexpensively and quickly build a training database. We conducted experiments on crowd footage of a college basketball game and describe our results. Importantly, we show how its possible to preserve the anonymity of the fans.

Back to Videos