Humanoid Vision Design for Object Detection, Localization and Mapping in Indoor Environments

Abstract

While most humanoid robots have a depth camera on their head, a single depth camera may not necessarily be the best form of humanoid robot vision. This paper presents humanoid vision design consisting of two depth cameras for localization and mapping. Unlike most localization and mapping papers, this paper investigates the effect of the humanoid vision design on localization and mapping in addition to object detection. The resulting robot trajectories have shown that the localization and mapping is most accurate when the robot with large head angle (head inclination) is moved in a wide corridor. The mathematical analysis shows that the area of Field-of-View (FOV) is large when the head angle is less than 29°. These results conclude that the optimal humanoid head angle is between 20° and 29°.