Christopher Burtt
Global Signage Product Manager
Thomson Reuters
September 2009 Question:
Facial recognition software is definitely a new and interesting topic. What do we need to know to better understand its application and when to consider incorporating this new technology in an existing DS installation?
Before implementing this technology, we would need to know what capabilities it offers. It has been touted as a way to more accurately capture demographics, and we would need to know if it can determine gender, ethnicity, and age. At a minimum it should be able to determine the number of screen viewers, but I would want to know how it deals with large numbers of simultaneous viewers. What is the upper limit to the number of faces it can count, and can it accurately identify demographics of multiple viewers? I would also want to know the margin of error for each category of demographics to know which I could rely on. That is, how often does it get the age, gender or ethnicity wrong?
If the technology works in a cost effective manner, it could fill a large gap in signage measurement today. Outside of retail environments, most statistics are based on sampling, typically at lengthy intervals. While this gives a good general picture, it does not account for special events, changes in traffic patterns or seasonal differences. Even retail, which can count the number of transactions, does not give a detailed account of specific store areas, including browsers who didn't purchase. Facial recognition could turn digital signage into one of the most accurately measured media, which would greatly assist in proving the worth of any individual sign.
Facial recognition is also promising as a way of changing content based on audience demographics, for example changing screen language based on ethnicity. However, this should be undertaken very carefully. While margins of error would probably be cancelled out over time, they could lead to embarassment if targetting individuals. I picture a teenage boy with a girlish face being pitched a lipstick ad. A better use would be to change future programming based on past patterns that have emerged.