Drone-Based Identification of Containers and Semi-Trailers in Inland Ports

EasyChair Preprint 14025

Abstract

This paper introduces a novel application utilizing drones and deep learning to identify containers and semi-trailers, enhancing inland port operations. With this drone-based image and text recognition system, the basic condition of the yard/storage area can be determined at any time without using (human) labor, eliminating the need for manual inspections. To our knowledge, this is the first instance of identifying containers and semi-trailers in a deep learning application through drone imagery. Automating identification through drone flights is one of the main goals of our InteGreatDrones (IGD) project. This paper lays the foundation and provides a first building block by addressing the challenges posed by the real-world data and the different drone perspectives, including the various altitudes, scenes, and viewpoints captured in this project, with the goal of cargo identification. We use a two-step recognition process, first localizing the text ID and then reading/identifying it. We take established methods such as EAST for scene text detection and TrOCR for optical character recognition and fine-tune them to enable accurate identification from drone imagery. Despite the challenging real-world images, we achieve an F1 of 0.5 for text detection and a CER of 0.16%.

Keyphrases: Automated ID Recognition, Intermodal Handling, Loading Unit Identification, Port Logistics, UAV-Based Monitoring

@booklet{EasyChair:14025,
  author    = {Jana Teegen and André Kelm and Ole Grasse and Maris Hillemann and Emre Gülsoylu and Simone Frintrop},
  title     = {Drone-Based Identification of Containers and Semi-Trailers in Inland Ports},
  howpublished = {EasyChair Preprint 14025},
  year      = {EasyChair, 2024}}