Short term re-identification of Automatic Teller Machine (ATM) users via face and body appearance features

Abstract

This paper describes a novel biometric scenario, where a person is authenticated at an ATM, and has to be re-identified from a camera within a very short time period, under very challenging illumination and pose conditions, and using data from a single session. The application scenario is the automatic retraction of forgotten card or cash at an ATM, which happens frequently, and causes inconvenience to users and loss of profits for the banks. We propose a multimodal authentication system that operates under the constraints imposed by this applications scenario, and implement face recognition and color based body appearance recognition to create a system that improves ATM behavior in case of forgotten card or cash by re-identifying the user from an embedded ATM camera. We focus on the scenario and the platform, and report tests with the proposed system under challenging conditions, obtained from ATMs placed in the field.