Bolstering Maritime Object Detection with Synthetic Data. Issue 31 (2022)

• Record Type: Journal Article
• Authors: Becktor, Jonathan; Schöller, Frederik E.T.; Boukas, Evangelos; Blanke, Mogens; Nalpantidis, Lazaros
• Is Part Of: IFAC-PapersOnLine. Volume 55:Issue 31(2022)
• Page Start: 64
• Page End: 69
• Publication Date: 2022
• Subjects: Autonomous Marine Vessels -- Machine Learning -- Autonomous Navigation -- Simulated data -- Unreal Engine; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2022.10.410
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: For autonomy in the maritime domain, object detection is a very important task, as one needs to perceive the surroundings to take appropriate action decisions. A large issue in maritime object detection and classification is the shortage of thorough datasets. In this work, our aim is to reduce this problem by introducing a pipeline for the generation of simulated data that matches the target domain, thereby achieving a more reliable and robust performance of our object detector. This data generation pipeline is easily modifiable and allows for varying setups that would be hard or dangerous to collect in real life. Furthermore, it enables the introduction of new classes without issue.