A photographer robot with multiview face detector

Abstract

Most photographer robots can only capture frontal-like faces, posing limitations on broader applications, especially when it is needed to capture non-frontal faces. We built a photographer robot with a multiview face detector embedded so that it can capture faces with a large range of poses. The core part of the multiview face detector is a state-of-the-art facial landmark model which can track facial features and contours and is robust to illumination variation and partial occlusion. Our photographer robot is built on the low-cost Turtlebot platform, which solves the issues of SLAM and auto-navigation. To reveal its application scope, it has been tested in various scenarios, including single and multiple persons, faces of different poses and under various lighting conditions.