https://www.um.edu.mt/library/oar/handle/123456789/91669 Fri, 10 Apr 2026 21:21:04 GMT 2026-04-10T21:21:04Z https://www.um.edu.mt/library/oar/handle/123456789/95681 Title: Deep sketching : vectorization of sketched drawings using deep learning Abstract: Sketch vectorization is an essential step to bridge the gap from hand drawn rough sketches to images that can be interpreted by computer based systems. Through this dissertation, a Fully Convolutional Neural Network is designed to automatically clean raster rough sketches into their line drawing counterpart. A custom loss function was created inspired from traditional feature based methods of analysing the quality of line drawings. The loss function quantifies the quality of the lines and gaps being extracted. It also promotes that the lines extracted are clearly separate from the background. A dataset was curated from images found ‘in the wild’. The ground truth images were drawn by hand and method to align hand drawn ground truths to the found sketches, irrespective of printing and scanning resolution, was created. The curated dataset was used along with another dataset created in similar conditions. The design proposed and the dataset used allowed sketches from different sources to be simplified. Over strokes were converted to their clean line counterparts, shading and hatching, which are artifacts which hinder the vectorization process which follows, are removed whilst detail from the sketches was retained. A vectorization algorithm is then applied to the output from the cleaned line drawing. The time taken to convert the sketched image to its cleaned vector counter parts is quick, hence the method can be used in down stream applications. Finally, metrics measured show an improvement of the proposed loss function when compared to the Mean Squared Error (MSE) loss function as well as an improvement upon the more simplified version of proposed loss functions. For validation the output is also compared to other varieties of methods proposed which are deep learning based or feature based. Description: B.Eng. (Hons)(Melit.) Fri, 01 Jan 2021 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/95681 2021-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/95581 Title: A robotic training partner for track runners Abstract: Typical training sessions of long-distance runners, both casual and professional, often include running at very specific paces for a certain distance or time intervals, in order to focus the training session on improving endurance or running power. Currently, runners keep track of their pace using equipment like a stopwatch or a GPS-enabled smartwatch, which provide the runner with a pace readout in real-time. However, it is well-known that a ‘physical’ pacer, also known as a rabbit, can be highly beneficial by providing a visual cue for the runner to follow at the desired pace. This can allow the runner to focus more on the running technique, or even act as an artificial competitor which can highly motivate the runner in demanding training sessions or time trials. A mobile robot can act as a physical pacer for a runner training on a running track. More specifically the mobile robot can employ a line-following system to follow one of the lines on a running track, and a speed control system to regulate its driving speed according to a custom running workout set by a user. Ultimately, the robot will act as a precise programmable ‘physical pacer’. In a previous project, an off-the-shelf electric remote-controlled model car was converted into a line-following robot using an infrared sensor array and a digital PID controller for closed loop steering control. This prototype robot was tested in controlled environments. The main objective of this dissertation is to build on the previous work to develop this robotic pacer by designing and adding a speed control system. This system makes use of an incremental encoder coupled to the drive shaft of the car to measure its translational speed, which is used as the feedback signal in a PID control loop to regulate the motor’s driving speed. Additionally, the robotic pacer was fitted with a Bluetooth module to facilitate wireless communication with a smartphone. The results obtained from tests performed on the speed control system contributed towards the understanding of the car’s speed response which can be used to improve the control systems and develop the robotic pacer prototype further in future works. Description: B.Eng. (Hons)(Melit.) Fri, 01 Jan 2021 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/95581 2021-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/95579 Title: Multi-robot coverage control Abstract: Robots have become an essential part of society. Their uses range from domestic to industrial and with emerging applications like the rise in self-driving vehicles, it is becoming more evident that robots can undertake more responsibilities that are typically associated with humans. The applications for multiple robot systems are also becoming more common, especially within the field of swarm robotics. Every job that robots can do are due to multiple and complex algorithms that make each application possible. Coverage control algorithms provide the tools necessary for multiple robots to be able to organise themselves within an area, and effectively cover all the area in the most optimal way. Such algorithms can give rise to new applications for robots in autonomous surveillance or search and rescue for example. The primary objectives of this dissertation are to review relevant theory and current mathematical approaches taken to solve the coverage control problem, implement one or more of the approaches found in literature and validate it on a group of mobile robots. Further complexities could be introduced with the application of a single peak, multiple peak or time varying density function that represents the importance of each region in the environment, by taking the robot dynamics into account, or by assuming a fleet of robots with different energy levels and sensing capabilities. The dissertation starts by looking at the mathematical approaches for tackling the coverage control problem. A coverage control algorithm is then explained and broken into its main components, which are further explained in more detail. Then, the approach taken to create and simulate the coverage control algorithm along with each one of its components is described. The added complexities are then introduced and explained. Finally, the results obtained from the modelling of the coverage control algorithm are evaluated and discussed. The results validated the algorithms successfully and the resulting simulation could be used to demonstrate the effectiveness of a real coverage control algorithm for a fleet of mobile robots. Description: B.Eng. (Hons)(Melit.) Fri, 01 Jan 2021 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/95579 2021-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/95576 Title: Head pose estimation using deep learning Abstract: Convolutional Neural Networks (CNNs) perform well on the head pose estimation problem, however, their generalisation ability depends on the training data provided to the CNN, in order to extract sufficient features to obtain an efficient head pose result. A method for estimating head pose using a CNN trained on real head images is proposed, however, real data can be sparse and laborious to collect. Thus, a CNN trained on synthetic head images is also investigated in this dissertation because it is easier to create synthetic data, which may be used to produce rare head poses in large enough quantities. The estimation of head pose by the CNN is formulated as a regression problem. An image pre-processing stage incorporates facial landmarks information into the face shape normalisation by the task simplifier, normalises the image array values, and generates facial landmark heatmaps. This is established prior to the feed-forward neural network, thus, this information is used to aid feature extraction from head images. Datasets which render head images that take gender, race, age, and expression into account are used, namely: 300W-LP, AFLW2000-3D, BIWI, and NVIDIA Synthetic Head. Six methods are being presented in this dissertation that use real data, synthetic data, and a combination of real and synthetic data. The results reveal that when the feed-forward neural network is trained on 300W-LP, fine tuned by classification on NVIDIA Synthetic Head, and further fine tuned end-to-end on a portion of BIWI, the Standard Deviation (SD) for each of the head pose angles is improved. Moreover, the average mean absolute error decreases from 4.67° to 2.93°on AFLW2000-3D, and from 6.08° to 2.59°on BIWI. Furthermore, when a model is trained on NVIDIA Synthetic Head and is fine tuned end-to-end on BIWI and 300W-LP, the average Mean Absolute Error (MAE) obtained is 2.96° when tested on BIWI, and 3.98° when tested on AFLW2000-3D. This dissertation shows that the CNN can extract features which can reflect head pose accurately even when the model is trained on synthetic data, significantly enhancing the possibility to train head pose models by only using computer generated images. Description: B.Eng. (Hons)(Melit.) Fri, 01 Jan 2021 00:00:00 GMT https://www.um.edu.mt/library/oar/handle/123456789/95576 2021-01-01T00:00:00Z