A Robust, Real-Time Capable Framework for Fully Automated Order Picking of Pouch-Parceled Goods

Abstract

The automatic handling of pouch-parceled goods is still challenging because the fillings cause almost random variations of the pouch geometry during conveyance. This paper presents a novel vision-based detection and tracking framework for robustly localizing and tracking pouch-parceled goods in real-time. It is exemplarily integrated in a Cover Belt Conveyor (CBC) system to decide whether and when pouches have to be separated. The core concepts include the estimation of the current pouch velocity using sparse optical flow, the detection and tracking of custom markers by means of structural image analysis and supervised classification, the verification of contextual constraints, and, eventually, the execution of a cutting strategy. The presented concepts tackle the general challenges superimposed by pouch packagings and are thus applicable in any other related automation scenarios as well. An average per-frame processing time of less than 20 ms, an averaged detection accuracy of 99.93% and a cutting accuracy less than 8 mm prove the real-time capabilities, robustness and applicability of the approach.