Amazon patents describe biometric user authentication and video tracking system
A pair of patent applications from Amazon apply the company’s facial recognition technology to user authentication and extend the tracking of identified individuals based on metadata and other observable features.
A patent for “Image analysis for user authentication” reported by EcommerceBytes describes a system for applying facial recognition with anti-spoofing measures. Reports of apparent tests of the described technology, or a similar one, emerged earlier this month, as a seller verification mechanism.
Amazon acknowledged its use of facial recognition for sellers to EcommerceBytes at the time.
“We always experiment with new ways to verify the information sellers provide us in order to protect our store from bad actors,” a representative said. “Seller identification information is securely stored and used only for identify verification.”
The company claims in the patent that authentication with facial biometrics is user friendly compared to legacy methods such as lengthy passwords, and that the difficulty of reproducing a user’s face in three dimensions makes it more secure.
Another patent, for “Selective identity recognition utilizing object tracking” details a method of identifying a person with facial recognition, and then maintaining that identity even if the person’s face becomes occluded or leaves the frame, based on other characteristics.
“Multimedia content such as video content is used in multiple settings and presents many difficult challenges in the context of video processing, such as recognizing and identifying objects moving through the multimedia content,” according to the patent background. “Multimedia content often includes large amount of information that make it challenging to process the video to determine, for example, when a particular individual is present in the video. Determining when a particular individual or other object is present in the video may require performing an identity recognition process and it may be computationally infeasible or inefficient to perform repeated identity recognition processes.”
As pointed out by Buzzfeed News, the patent specifically mentions associating people with other personal information, such as their job, by analyzing their clothes. This could be used to disambiguate an individual from others with the same name, but also seems to require a rapid build-up of data about the individual, which is bound to cause concern among privacy advocates.
AWS has faced an ongoing controversy about the use of Rekogniton by law enforcement.