Overlapped object recognition using range and image data for a service robot

Abstract

We propose an algorithm for object recognition in indoor service robots. The problem of object recognition is one of the key challenges in the creation of realistic robotic services. Despite great advancements in the past, sufficiently accurate object recognition for service robots in real-world environments remains problematic. Our algorithm uses image and range data information that is available on a service robotic platform to execute the segmentation and classification steps. The segmentation decision rule is applied to correctly segment objects even in overlapped placements. In the classification step, the bag of words is employed with feature descriptors that are constructed from image and range information of segmented regions. In experiments, a working service robotic platform recognizes objects of similar shapes and colors. In addition, we test the recognition capability of overlapped objects. The results demonstrate the feasibility of the proposed algorithm. © 2013 Springer-Verlag.