Study on object recognition by active stereo camera for clean-up robot

Abstract

This paper describes a clean-up robot that is composed of a mobile base and a manipulator. An object recognition algorithm that is based on an active stereo camera system is also proposed for the clean-up robot. In order for the robot to clear a dining table, the stereo camera system must identify the objects on the dining table and evaluate their positions. For localizing and detecting objects, a convolutional neural network (CNN) that requires a large amount of image data and numerous computations is generally employed. In this study, we use a transfer learning method that is capable of omitting the huge data and the sliding window method that can extract the object area. The SUEF (speeded up robust features) feature points that are extracted by the right camera are compared with those obtained by the left camera, and the area where the number of matched points is the largest is outputted as the object that is to be identified by the left and right cameras. The effectiveness of the object recognition algorithm is verified through clean-up experiments.