Integrated computer vision for autonomous marine operations

Abstract

In this graduation thesis the feasibility of a camera as sensor for nautical operations is investigated. In
aerospace and automotive industries autonomous vehicles are already widely tested and used. However
the marine industry has fallen behind in this development. Therefore this research is focussing on
automation in the maritime sector. First a system was developed which measures distance and heading
from a video feed. Thereafter simulink models were made to develop a system that autonomously can
perform nautical operations. For this research multiple operations have been modeled. First of all
the platform approach and the replenishment approach. In these operations the system should be able
to approach a stationary or moving target. Also station keeping after the approach is required. The
complexity of the operations was increased with the choice for follow the leader from the maneuvering
operations. Finally the avoidance of upcoming collisions was modeled. The models for all of these
operations were rst developed and tested with an oine model. This enabled a quick and accurate
controller tuning. The rst real test for these models was done in a ship handling simulator. Here the
models were tested and the in
uence of environmental conditions was measured. In the simulator all
models performed quite well. However the performance is largely limited by the choice for the camera
and environmental conditions. The most interfering conditions are lateral incoming waves, fog and light.
Models for small scale testing were developed, but the communication between dierent components and
the control system proved to be a large issue. Therefore these models could not be tested. All in all
this research shows that after improving system component, a camera system can very well be used as a
sensor for autonomous operations at sea.