Accuracy of sea ice floe size observation from an aerial camera at slant angles

The importance of measuring the size of ice floes in e.g. marine navigation and environmental sciences has made it a frequently performed procedure. When real-time data is required, images from a camera on-board an aerial vehicle or mounted on a marine vessel is commonly preferred over satellite images. Their lower fields of view can be improved by tilting the cameras to capture images of a larger area. However, this introduces a greatly changing ground resolution within the same camera image, which makes size estimation a more complex task. It is nevertheless performed in several methods to estimate the size of ice floes. In this paper, ice floe size estimation is evaluated for different scenarios when using an aerial camera at slant angles. In order to reduce errors caused by automatic image segmentation and attitude estimation algorithms, the methods are aided by human input. The estimates are performed on real world data captured during the Statoil Station Keeping Trials in the Bothnian Bay during March 2017. The results conclude that the major challenge is to guarantee separation between ice floes in the camera images, which is something that requires both enough ground resolution and a suitable image segmentation algorithm.