Marine pollution is a critical issue impacting the global community, with underwater waste a particularly daunting challenge. While autonomous detection and collection of underwater waste is highly desirable, these are extremely difficult tasks. This difficulty arises from the intrinsic complexities of the aquatic environment, including variable lighting conditions, reduced visibility, and the complex nature of water currents. This paper focuses on novel approaches for autonomous detection of underwater waste and proposes to incorporate domain knowledge to refine deep-learning-based underwater object detection techniques. More specifically, the domain knowledge is represented with models in the state space form that describe the motion of the target objects to assist the classification of objects. Moreover, optical flow is combined with a k-means clustering algorithm to extract the trajectory of the target objects from videos. These trajectories are subsequently fed into a neural network that is trained using knowledge distillation, enhanced with domain knowledge. For our experiments, a simulator is devised to facilitate the creation of a dataset for developing and testing the proposed architecture. The results of the experiments demonstrate that including domain knowledge within the object detection approach with neural networks provides numerous substantial advantages, including enhanced robustness against noisy and poorly labelled data, facilitation of semi-supervised learning, and consistent superiority in accuracy over the baseline scenario. Additionally, combining the domain knowledge with a neural network significantly increases the computational speed of object detection compared to using the standalone domain knowledge module.

While various tracking algorithms have demonstrated effectiveness in terrestrial and aerial contexts, their performance in underwater settings remains unexplored. Object tracking in underwater videos presents unique challenges due to variable lighting, water turbidity, and unpredictable camera movement, all of which are likely to hinder the performance of traditional detection and tracking methods. Addressing this gap is crucial for applications such as marine biology research, underwater surveillance, and autonomous underwater vehicles.

This thesis first evaluates existing tracking algorithms, SORT, ByteTrack, and Bag-of-Tricks SORT (BoT-SORT), each incorporating motion estimation and linear data assignment methods on a novel underwater video dataset with a moving camera. The thesis then improves on the SORT algorithm by adopting velocity estimation techniques, a formula-based and an optical flow-based, giving rise to two new algorithms, SORT-V and SORT-OF. Furthermore, the thesis proposes a novel tracker that utilises an Interacting Multiple Model filter to estimate the location of the target object. The evaluation focuses on finding a balance between specific metrics, such as tracking accuracy, identity switches, and the number of tracked and lost trajectories. The results indicate that using velocity estimation techniques improves the tracking accuracy by 11% and tracks more objects by nearly halving the number of lost objects. Incorporating a constant acceleration model in the IMM filter gives the best result, with the highest tracking accuracy and with the least number of identity switches, all in real-time computational speed.

This thesis reports the results of research into the stability of the all-to-all coupled discrete time Kuramoto model under constant, matched input disturbances. The discrete time Kuramoto model can be used as a dynamic, decentralized multi-agent orientation coordination system: once initialized, the agents will communicate their orientations to all other agents and calculate their own step-update based on the received data. A properly controlled and undisturbed Kuramoto model can direct agents to two final, stable sets of orientations: either all agents align to the same orientation or they form a balanced set of orientations with the characteristic that the centre of gravity of all orientations on the unit circle is at the origin.

These final states will not be reached when at least one of the agents is influenced by a disturbance. Not only will this agent be affected, but because of the networked system, the disturbance in one agent will influence other agents as well. Understanding the Kuramoto model enables the design of controllers that can attenuate the influence of disturbances. All controllers are designed with the assumption of constant matched input disturbances.
The first controller is an error feedback controller. For this strategy, the original Kuramoto model had to be modified. The resulting controller can attenuate the effects of matched input disturbances in a system of agents, but individual agents with matched input disturbances will not reach a steady state.

The second controller is based on predictor-error feedback and the Kuramoto characteristic that the average orientation in a Kuramoto model is constant. The controller is augmented with an algorithm that generates a one-step ahead prediction based on the known states and inputs. Since the disturbance is assumed to be constant, its effects can be calculated and attenuated in the next time step. This controller succeeds in directing the system to the same aligned set as the undisturbed system, although via a different trajectory. The controller also succeeds in directing the system to a balanced set, but for systems with N ≥ 4 agents that balanced set is different from the undisturbed set.

Since the second controller showed that a deviation from the undisturbed trajectory leads to a different balanced set, the third controller is designed with reference trajectories that do not use the actual states and inputs, but are generated fully autonomously. The difference between reference and actual state is processed by a proportional-integral algorithm to ensure zero steady state error. This controller however has the possibility of destabilizing the system, when not properly tuned.

All controllers have their merits: the first controller decouples the agents, thereby preventing that a disturbed agent affects others. Under constant disturbance, the second controller guarantees stability, but will let the agents follow different trajectories than the undisturbed system, leading to different balanced sets. The third controller can direct all agents to their undisturbed trajectory, but can negatively impact the stability properties of the Kuramoto controller when improperly tuned.