Detection of Household Furniture Storage Space in Depth Images

Autonomous service robots assisting in homes and institutions should be able to store and retrieve items in household furniture. This paper presents a neural network-based computer vision method for detection of storage space within storage furniture. The method consists of automatic storage volume detection and annotation within 3D models of furniture, and automatic generation of a large number of depth images of storage furniture with assigned bounding boxes representing the storage space above the furniture shelves. These scenes are used for the training of a neural network. The proposed method enables storage space detection in depth images acquired by a real 3D camera. Depth images with annotations of storage space bounding boxes are also a contribution of this paper and are available for further research. The proposed approach represents a novel research topic, and the results show that it is possible to facilitate a network originally developed for object detection to detect empty or cluttered storage volumes.