Evaluation of Face Recognition Systems for RoboCup@Home Service Robots

• Federico Peña Universidad de la República
• Manuel Eirea Universidad de la República
• Mercedes Marzoa Tanco Universidad de la República
• Federico Andrade Universidad de la República

Resumo

Face recognition technology is becoming increasingly important in a variety of applications. In the context of service robotics, accurate and efficient face recognition has big potential to improve overall performance and capabilities, including the ability to create more natural human-robot interactions. This paper presents an in-depth study of state-of-the-art face detection and recognition algorithms, focusing on their use in the RoboCup@Home competition. Several face recognition architectures were evaluated, including DeepFace, ArcFace and FaceNet, considering two alternative implementations of the latter. The methods were first tested on two state-of-the-art datasets, and then, to test them under competition-like conditions, they were tested on images captured in real time, taking into account different situations, such as lighting, use of accessories, face position, and multiple faces in the same scene. Results show that Esler's implementation of FaceNet would perform better on RoboCup@Home challenges.

Palavras-chave: Face Recognition, FaceNet, Service Robots, RoboCup@Home