In our diploma thesis we issue a navigation problem of autonomous surface vehicles.
Presume we have a vessel appointed with sensors, such as GPS, IMU, compass,
a stereo sistem of two cameras and a standalone processing unit that combines
sensor data and performs segmentation and planning in real time. We begin by
introducing the algorithm of segmentation, that along with help of advanced com-
puter vision methods extracts useful visual information, therefore avoids obstacles
in its path (boats, swimmers, buoys). Then, we focus on horizon estimation by taking into account data of innertial measurement unit,
with whom we improve the estimation of a sea border. To improve localization
of obstacles in front of the vessel, we calculate depth of every pixel in the image. Image pairs are first rectified as we simplify the correspondence search to single dimension. Furthermore, we present, implement and evaluate our methods. We conclude by discussing further work that can be done.