An efficient alternative approach for home furniture detection and localization by an autonomous mobile robot

Overview & Synopsis

In order for service robots to help humans in daily home tasks, they need to have a better understanding of their environment. Therefore, detection and localization of home furniture becomes a very important capability for them. This paper presents an approach to detect typical home furniture by an RGB-D sensor mounted on a mobile robot. The approach is based on the analysis of discriminant features extracted from very easy to compute measures distributions. Over an offline learning phase, each piece of furniture is modeled according to their distributions of: height, color (H and S components) and normals. Then a sequence of distributions analysis are applied to the scene for selecting, according to learned models, the pieces of furniture with a high probability of being present. The point cloud is segmented according to model analysis and then segmented points are projected to the floor plane for clustering and noise removal. According to previous analysis and footprints of segmented points clouds, regions are then classified as a possible piece of furniture. Finally, the orientation and localization of the detected furnitures are obtained, using the footprint and their neighbor regions. The proposed approach has been proved to be very efficient in order to be incorporated on mobile robotic platforms.