Indoor home furniture detection with RGB-D data for service robots

Home furniture detection is a very important topic enabling a robot to provide useful services at home. This paper presents an algorithm to identify and detect home furnitures by an autonomous service robot. The furniture considered in this paper includes large objects (e.g. beds, sofas, etc.) that can be moved by humans or by the robot on common tasks. 3D data acquired from an RGB-D camera mounted on the robot are analyzed to find discriminant features that characterize the pieces of furniture to be detected. The proposed methodology avoids the processing of the complete frame by the use of a small set of random points. These points are learned and classified in function of several attributes: color, 3D position and 3D normals. A function of random region growing and partial 3D modeling is then applied to validate the detection of a specific piece of furniture regarding the set of known furniture models. The process runs in real-time and can be easily incorporated to service robots.