Development on Fovea-Lens Camera System for Human Recognition in Robot System

Abstract

This paper describes the development on the fovea-lens camera system for a robotic system which recognize and track human object. The fovea lens has the characteristics imitating human eye: wide range of view and high resolution at the center of image. The wide range of view enables to find human earlier and the high resolution enables to recognize human well. We realized the realtime system to recognize human by using AdaBoost machine learning. An experiment shows that our fovea-lens camera is suitable for the robotic camera. The human recognition rate with the fovea lens is more than 91 % even though the rate with the wide-range lens is less than 80 % with 5 m distance between the human and the camera. The detection range in horizontal plane of the fovea lens is almost same as the wide-range lens.