Options
Development and analysis of physics-based models for autonomous underwater vehicle navigation and the reconstruction of underwater images
Publikationstyp
Doctoral Thesis
Publikationsdatum
2016-05-31
Sprache
English
Author
Advisor
Referee
Title Granting Institution
Universität Duisburg-Essen
Place of Title Granting Institution
Duisburg
Examination Date
2016-05-02
TORE-URI
Autonomous underwater vehicles (AUVs) have changed the way marine environment is surveyed, monitored and mapped. They have a wide range of applications in research, military, and commercial settings. AUVs should not only perform a given task but also adapt to changes in the environment. Typical effects are sudden side currents, downdrafts, and other effects which are extremely unpredictable. Simultaneous localisation and mapping (SLAM) is a well-known and well-understood problem in robotics. For land-based robots in 2-D environments this problem is generally considered to be solved. SLAM algorithms for these tend to rely on optical recognition in combination with dead reckoning and inertial measurement units. The optical properties of water and especially seawater prevent the use of established optical recognition algorithms. High quality images with correct colouring simplify the detection of underwater objects and may allow the use of visual SLAM algorithms developed for land-based robots underwater. Hence, appropriate image processing is required especially in deep water. In this thesis physics-based models for autonomous underwater vehicle navigation are developed with an emphasis on fast exploratory AUVs with cruising speeds in the range of 5 kn to 20 kn. The system should be capable of detecting disturbances in the water flow and be able to use a camera for object detection, ground survey, and especially for navigational purposes. Furthermore, it should be possible to integrate the system into existing autonomous underwater vehicles. Therefore, the system must be small and lightweight such that the payload of the AUV is not reduced significantly. The required computational power and the power consumption must also be small such that the duration of the vehicle does not decrease strongly. The algorithms should also be fast to allow SLAM application. In the first part of the thesis the applicability of different learning methods for determining flow parameters of a surrounding fluid from pressure on an AUV body are tested based on numerous computational fluid dynamical (CFD) simulations and using pressure data from specified points on the surface of the AUV. It is shown that a combination of support vector machines (SVM) is an excellent choice to perform this task. With the findings from the simulations the position of pressure measurement points is then optimised such that the most significant pressure changes due to changing flow velocities can be captured. This also reduces the number of measurement points. It is then shown that also for the optimised setup support vector machines are the best choices for the given task. However, fewer machines are required in this case. In the second part of the thesis different learning methods are applied for the reconstruction of underwater images. First laboratory tests are performed using a special light source imitating underwater lighting conditions. It is shown that a combination of the k-nearest neighbour method and support vector machines yields excellent results. Based on these results an experimental verification is performed under severe conditions in murky water of a diving basin. It is shown that the k-nearest neighbour method gives very good results for small distances between the object and the camera and for small water depths in the red channel. For higher distances, water depths, and for the other colour channels a combination of support vector machines is the best choice for the reconstruction of the colour as seen under white light from the underwater images. Thus, a novel approach to autonomous underwater vehicle navigation and the reconstruction of underwater images is proposed and developed in this thesis.
DDC Class
600: Technik
620: Ingenieurwissenschaften