An automated system for semantic object labeling with soft object recognition and dynamic programming segmentation

Abstract

This paper presents an automated system for generating a semantic map of inventory in a retail environment. Developing this map involves assigning a department label to each discrete section of shelving. We use a priori information to boost data from laser and camera sensors for object recognition and semantic labeling. We introduce a soft object map and a dynamic programming algorithm for point cloud segmentation. The primary contribution of this work is the integration of multiple systems including an automated path planning and navigation subsystem and a semantic mapping object recognition system. This work also represents an important contribution to robots working reliably in human environments. To our knowledge this is the first actual implementation of a fully automated robot inventory labeling system for a retail environment. The framework presented in this paper is easily scalable to other retail environments and is also relevant in any indoor environment with organized shelves, such as business storage facilities and hospital pharmacies.