Master thesis presentation Elina Sahlberg, Björn Krook Willén, MPCAS

Abstract: Svenska Sjöräddningssällskapet have partnered with Infotiv to develop a prototype
drone and launcher system for search-and-rescue missions at sea.In this thesis project
we investigated the possibility of automating parts of the launch sequence of a
seaborn surveillance drone. The main goal was to train a neural network-model to
use a camera-feed to determine if it was safe or not to launch the drone in a given
direction, and then integrate this solution with the GUI through an API. Monocular
depth estimation using transfer learning and the Kitti data set was evaluated. The
kitti data set does not contain maritime scenery leading to a unsatisfactory monocu-
lar depth estimation model. U-net and CNN models were trained on the MaSTr1325
dataset, the data set contains semantic segmented maritime imagery. We collected
additional data for the semantic segmentation models, along with a post processing
step that evaluated if it was safe to launch or not. These models yielded satisfactory
results that will be used by the drone operator as an extra safety measure during
launch.and the backend endpoint was the final result of this project.

Supervisors: Victor Nilsson, Hamid Ebad
Examiner: Giovanni Volpe
Opponent: Marcus Degerman