Towards User Identification for Autonomous Vehicles by an ANN-Based Face Recognition Approach

Abstract:
Accurate and robust user identification is relevant to various automated driving tasks. One is stopping of autonomous vehicles at the exact position of the user, even when other people are standing nearby. Other tasks are more related to access control, e.g. to ensure that only the user who booked the vehicle can enter it. In this publication, the authors develop and evaluate a neural network-based face recognition approach. It can be used as one of the main components in implementing user identification. By employing transfer learning techniques and the latest neural network architectures, it is possible to achieve good performance, even if only a few training samples are used. The paper also describes how the approach can be employed in an automotive environment, facing challenges such as varying backgrounds, different object scales, etc. Finally, the authors give an outlook on how to run the approach on an ECU.