DescriptionMaritime vessel detection from inshore and offshore imagery can be crucial in
regulating maritime traffic, its environmental impact, sea, and coastal security, etc. Since the IMO (International Maritime Organization) has focused more and more on autonomous navigation in the last decade, it came into spotlight in commercial applications too. However, the accuracy of algorithms trained on general object detection datasets is tremendously affected by various factors (occlusion, scale variation, environmental conditions) when tested in complex marine environments. The availability of domain-specific datasets is limited. Datasets of waterborne maritime vessels are scarce, a reduced number of studies addressing ship detection. We present in the poster our experiences regarding the collection of maritime datasets, with a special focus on the Finnish Archipelago. We will discuss the challenges that arise when creating annotations using different approaches. Moreover, we evaluate various deep learning approaches for object detection and how different factors influence object detection in the marine environment.
|20 Jun 2022 → 21 Jun 2022
|Annual Symposium for Computer Science 2022