3D face recognition using inception networks for service robots

Abstract

The field of face recognition has significantly advanced as deep learning methods, such as those using CNNs, continuously show improvements. However, despite face recognition's promising potential, there are still many concerns regarding privacy and safety. Moreover, the first 2D algorithms, besides having good performance, turned out to be influenced by several factors like the environment's lighting conditions, pose, and facial expression of the subjects, compromising the model's accuracy. This work describes the development of a computer vision system using Deep Learning methods to detect and recognise human faces in 3D in real-time. The RGB images and depth maps from several subjects were captured using an Intel RealSense D455, processed, and consequently provided into two independent CNNs, an Inception-Resnet V1 to deal with the RGB images and an Inception V3 to deal with depth maps. The final algorithm was implemented on the anthropomorphic domestic and healthcare service robot CHARMIE (Collaborative Home Assistant Robot by Minho Industrial Electronics) to perform its tasks according to the recognised user.