Computer vision for surface analysis of stretch wrapped pallets

Abstract

The majority of the handling of goods in a distribution center is performed automatically, except removing of stretch film from arriving pallets. The use of cameras and artificial intelligence to analyze the pallets, decide which of those that are stretch-film-wrapped and then calculate how a robot can remove the stretch film would solve this last stage to make a distribution center fully automated. The goal was to develop a deep learning model that can identify which pallets are suitable for automatic unwrapping. To make a dataset, cameras and a photocell were mounted at two strategic locations where pallets are passing. The result is more than 50000 images of pallets, divided into six classes. Reaching the target, automatic removal of the stretch film, must be regarded as future work due to the limited amount of time of this master thesis. Nevertheless, a model that is capable of dividing pallets into six different classes on the fly by using computer vision is ready. It is by the use of the model possible to sort pallets wrapped with stretch film in the direction of an unwrapping station, whilst other pallets are directed elsewhere. The result may be useful for further development after finishing this thesis.