Dylan Seychell
University of Malta, Artificial Intelligence, Faculty Member
The Selfie project was not only inspired by the long history of the self-portrait, but also intended to create a genealogy between the self-portraits of masters from the Modern art era and the selfie. The project, designed as a... more
The Selfie project was not only inspired by the long history of the self-portrait, but also intended to create a genealogy between the self-portraits of masters from the Modern art era and the selfie. The project, designed as a walkthrough experience, consisted of three major engagement areas. On entering the space, children were directed into a ‘transformation' area – a typical theatrical wardrobe, where they could dress up in a variety of costumes, including hats and wigs. Once garbed, children were given smart phones and led to the area where they could take a selfie with a celebrity such as Gauguin, Cézanne, Monet, Van Gogh, Modigliani and Munch. Finally, they could manipulate the selfie using gesture-based technology and post it online. The attraction proved to be extremely popular and the children who participated were extremely satisfied with the experience.
Research Interests:
Research Interests:
The success of social networking sites has led people to require the use of multiple accounts on different platforms which effectively increases the risks in managing them. Following and finding information about friends and family has... more
The success of social networking sites has led people to require the use of multiple accounts on different platforms which effectively increases the risks in managing them. Following and finding information about friends and family has become an issue too. Guided by these observations and by careful research of existing adaptive web technologies, the authors’ team worked on the development of SNAP - an adaptive social network integrator which aimed to amalgamate various social networks (Facebook, Twitter, and Flickr) in one adaptive environment, which unobtrusively sorts the users’ feed according to his/her preference. To achieve data transfer and authorisation, SNAP uses the newest version of the OAuth protocol. Adaptivity was achieved through statistical filtering. The initial field tests show that the system works, however there is definitely room for improvement in terms of Social Network Integration, and testers generally expressed an interest in the idea of using an adaptive s...
Research Interests:
The first generation of Digital Natives (DNs) is now growing up. However, these digital natives were rather late starters since; their exposure to computers started when they could master the mouse and the penetration of computers in... more
The first generation of Digital Natives (DNs) is now growing up. However, these digital natives were rather late starters since; their exposure to computers started when they could master the mouse and the penetration of computers in educational institutions was still very low. Today, a new breed of digital natives is emerging. This new breed includes those individuals who are being introduced from their first instances to the world of wireless devices. One year olds manage to master the intuitive touch interfaces of their tablets whilst sitting comfortably in their baby bouncers. The controller-less interfaces allow these children to interact with a machine in a way which was unconceivable below. Thus, our research investigated the paradigm shift between the different generations of digital natives. We analysed the way in which these two generations differ from each other and we explored how the world needs to change in order to harness the potential of these new digital natives.
Research Interests:
Research Interests:
Europe faces a considerable challenge in providing good quality health care in the forthcoming future as the aging population increases. The phenomenon also results in a considerable cost on society due to the dependency on the public... more
Europe faces a considerable challenge in providing good quality health care in the forthcoming future as the aging population increases. The phenomenon also results in a considerable cost on society due to the dependency on the public health sector particularly because such individuals would not be able to contribute to the economy. On the other hand, younger persons would need to make alternative arrangements to assist their elderly parents or relatives, potentially affecting productivity. The project PervasIve Nursing And docToral Assistant (PINATA) seeks to tackle this matter through the merging of Ambient Intelligence (AmI) and semantic web technologies. PINATA utilises pervasive devices to aid doctors and nurses to focus on the patient and thus improve the quality of healthcare services. This project proves the significant importance of using wireless technology in healthcare. This paper focuses on the use of Wi-Fi and RFID in an effort to enable continuous and intelligent moni...
Research Interests:
ABSTRACT One of the challenges when predicting human behavior is the variation of the same behavior in relation to the culture of the individual in question. This paper demonstrates a technique that can be employed in order to predict... more
ABSTRACT One of the challenges when predicting human behavior is the variation of the same behavior in relation to the culture of the individual in question. This paper demonstrates a technique that can be employed in order to predict certain individual cultural attributes based on a regression model that considers age, gender and nationality. The technique makes use of the data set collected through the World Values Survey. An equation based on the Multinomial Logistic Regression was derived in order to extract the values of the designated cultural attributes with respect to the given parameters. This paper briefly explores the World Values Survey together with the theory behind this concept that lead to the composition of a set of equations. A possible application of this model is also presented.
Research Interests:
Research Interests:
Vision and language tasks such as Visual Relation Detection and Visual Question Answering benefit from semantic features that afford proper grounding of language. The 3D depth of objects depicted in 2D images is one such feature. However... more
Vision and language tasks such as Visual Relation Detection and Visual Question Answering benefit from semantic features that afford proper grounding of language. The 3D depth of objects depicted in 2D images is one such feature. However it is very difficult to obtain accurate depth information without learning the appropriate features, which are scene dependent. The state of the art in this area are complex Neural Network models trained on stereo image data to predict depth per pixel. Fortunately, in some tasks, its only the relative depth between objects that is required. In this paper the extent to which semantic features can predict course relative depth is investigated. The problem is casted as a classification one and geometrical features based on object bounding boxes, object labels and scene attributes are computed and used as inputs to pattern recognition models to predict relative depth. i.e behind, in-front and neutral. The results are compared to those obtained from aver...
Research Interests:
With the emergence of deep learning methods for image segmentation, the potential of approaches for automatic brain tumour delineation has increased substantially. This paper presents a model which is inspired by U-Net++ for this task... more
With the emergence of deep learning methods for image segmentation, the potential of approaches for automatic brain tumour delineation has increased substantially. This paper presents a model which is inspired by U-Net++ for this task which makes training more efficient whilst also returning better accuracy. Our approach obtained Dice Scores of 0.90, 0.85, and 0.68 on the whole tumour, tumour core, and enhanced tumour core classes. These results were obtained on a holdout set of 68 scans from the BraTS 2019 training dataset. Our model also uses half the parameters of a popular U-Net adaptation which makes use of residual blocks, resulting in faster training. On average, our model performed 8.44% better than the latter for Dice scores for all three classes within our setup.
Research Interests:
Bib number recognition (BNR) from unstructured marathon images can be a challenging task. This is because the images captured at these events are very inconsistent since, they are often captured by multiple photographers, at various... more
Bib number recognition (BNR) from unstructured marathon images can be a challenging task. This is because the images captured at these events are very inconsistent since, they are often captured by multiple photographers, at various locations and times. This results in images containing different backgrounds, angles and illumination. The images often contain multiple participants in various poses, where the bib numbers can be obstructed by the participants themselves. The bib numbers are often printed on flexible paper and can easily be deformed which distorts the printed numbers. In this work we present a BNR system based on deep learning which is able to locate bib numbers in unstructured, complex marathon images. Using the segmented bib numbers the system then, recognizes the digits and finally outputs the bib numbers that it was able to detect in the image. The first stage consists of a fully Convolutional Neural Network (CNN) to segment the bib numbers while the second stage consists of a Convolutional Recurrent Neural Network (CRNN) used to recognize the detected numbers. The proposed method obtained an F1 score of 0.69 which outperformed existing methods.
Effective immersion takes place when the user can relate to the 3D environment presented and interact with key objects. Efficiently predicting which objects in a scene are in the user’s attention, without using additional hardware, such... more
Effective immersion takes place when the user can relate to the 3D environment presented and interact with key objects. Efficiently predicting which objects in a scene are in the user’s attention, without using additional hardware, such as eye tracking solutions, provides an opportunity for creating more immersive scenes in real time and at lower costs. This is nonetheless algorithmically challenging. In this paper, we are proposing a technique that efficiently and effectively identifies the most salient objects in a scene. We show how it accurately matches user selection within 0.04s and is over 95% faster than other saliency algorithms while also providing a ranking of the most salient segments in a scene.
Research Interests:
Image Inpainting techniques are generally challenging to evaluate objectively due to the lack of comparative data, as a reference image of the new scene, does not exist.. This paper presents an approach that uses our newly released... more
Image Inpainting techniques are generally challenging to evaluate objectively due to the lack of comparative data, as a reference image of the new scene, does not exist.. This paper presents an approach that uses our newly released dataset specifically designed to allow objective evaluation of inpainting techniques. In this work we demonstrate how traditional in-painting techniques can be objectively evaluated and compared together with modern deep learning and adversarial approaches. We further demonstrate how an unsupervised technique compares better than deep learning approaches.
The analysis of imagery from outdoor remote sensing is a technique widely used for surveying and data gathering. This paper studies techniques to be deployed in small object localisation using Convolutional Neural Networks (CNN), with the... more
The analysis of imagery from outdoor remote sensing is a technique widely used for surveying and data gathering. This paper studies techniques to be deployed in small object localisation using Convolutional Neural Networks (CNN), with the aim to detect litter in outdoor non-urban imagery. The detection of small objects requires distinguishing features between foreground and background. A litter detection application has to counter high variability in the foreground, as litter is defined as a super-class of common objects, and the high variability found in a rural or coastal backgrounds. Remote sensing imagery of nonurban scenery does not offer high contrasting features, reducing the effect of normal object localisation techniques.
Research Interests:
Vision and language tasks such as Visual Relation Detection and Visual Question Answering benefit from semantic features that afford proper grounding of language. The 3D depth of objects depicted in 2D images is one such feature. However... more
Vision and language tasks such as Visual Relation Detection and Visual Question Answering benefit from semantic features that afford proper grounding of language. The 3D depth of objects depicted in 2D images is one such feature. However it is very difficult to obtain accurate depth information without learning the appropriate features, which are scene dependent. The state of the art in this area are complex Neural Network models trained on stereo image data to predict depth per pixel. Fortunately, in some tasks, its only the relative depth between objects that is required. In this paper the extent to which semantic features can predict course relative depth is investigated. The problem is casted as a classification one and geometrical features based on object bounding boxes, object labels and scene attributes are computed and used as inputs to pattern recognition models to predict relative depth. i.e behind, in-front and neutral. The results are compared to those obtained from aver...
Research Interests:
Analysing Handwritten Documents is a challenging task. This particular area cannot always come up with general solutions, given that most handwritten manuscripts contain unique characteristics that describe how the document was written,... more
Analysing Handwritten Documents is a challenging task. This particular area cannot always come up with general solutions, given that most handwritten manuscripts contain unique characteristics that describe how the document was written, which include different handwritings. These challenges in transcribing different handwriting styles are due to various scribes contributing to the transcription of the text and degradation of the script. In this chapter, an overview of different techniques used in handwritten text recognition systems is presented. The approaches and algorithms can be adopted for different document types irrespective of the state of the scanned documents. Moreover, two different general approaches to handwritten character recognition are shown. The first approach goes through a fairly standard process to normalise, segment and recognise characters. The other approach is a segmentation free approach that uses neural networks for both segmentation and recognition.
Research Interests:
With the emergence of deep learning methods for image segmentation, the potential of approaches for automatic brain tumour delineation has increased substantially. This paper presents a model which is inspired by U-Net++ for this task... more
With the emergence of deep learning methods for image segmentation, the potential of approaches for automatic brain tumour delineation has increased substantially. This paper presents a model which is inspired by U-Net++ for this task which makes training more efficient whilst also returning better accuracy. Our approach obtained Dice Scores of 0.90, 0.85, and 0.68 on the whole tumour, tumour core, and enhanced tumour core classes. These results were obtained on a holdout set of 68 scans from the BraTS 2019 training dataset. Our model also uses half the parameters of a popular U-Net adaptation which makes use of residual blocks, resulting in faster training. On average, our model performed 8.44% better than the latter for Dice scores for all three classes within our setup.
Research Interests:
Research Interests:
Touring around a City can sometimes be frustrating rather than an enjoyable experience. The scope of the Virtual Mobile City Guide (VMCG) is to create a mobile application which aims to provide the user with tools normally used by... more
Touring around a City can sometimes be frustrating rather than an enjoyable experience. The scope of the Virtual Mobile City Guide (VMCG) is to create a mobile application which aims to provide the user with tools normally used by tourists while travelling and provides them with factual information about the city. The VMCG is a mash up of different APIs implemented in the Android platform which together with an information infrastructure provides the user with information about different attractions and guidance around the city in question. While providing the user with the traditional map view by making use of the Google maps API, the VMCG also employs the Wikitude® API to provide the user with an innovative approach to navigating through cities. This view uses augmented reality to indicate the location of attractions and displays information in the same augmented reality. The VMCG also has a built in recommendation engine which suggests attractions to the user depending on the att...
Research Interests:
Object detection has progressed rapidly during the last few years and has become a highly significant area in computer vision. Amidst the rise of autonomous vehicles and smart traffic management systems, accurate vehicle detection under... more
Object detection has progressed rapidly during the last few years and has become a highly significant area in computer vision. Amidst the rise of autonomous vehicles and smart traffic management systems, accurate vehicle detection under various lighting conditions has become paramount. This paper compares four state-of-the-art models Faster R-CNN, RetinaNet, YOLOv3 and YOLOv4 on how precise they detect vehicles under day and night-time scenarios. Experiments measure the 50% and 75% threshold average precision differences of multiple vehicle classes across two datasets. The results reveal a worse accuracy average of 15-20% and a maximum difference of 33% at night when compared to more illuminated day images.
The acquisition of any goal happens only with the correct dose of motivation instilled in the individual pursuing it. Mobile technology is at the same time providing us with different sensors and technology which allow us to measure... more
The acquisition of any goal happens only with the correct dose of motivation instilled in the individual pursuing it. Mobile technology is at the same time providing us with different sensors and technology which allow us to measure valuable attributes around a person who is engaged in a learning experience. In this paper we will be studying what motivates an individual while finding methods on the mobile device which will reach this motivation. The socio-cultural background of the individual undergoing learning will also be brought into context by acting as one of the driving forces of the presented recommendation technique.
Research Interests:
Detecting human abnormal activities is the process of observing rare events that deviate from normality. In this study, an automated camera-based system that is able to detect irregular human behaviour is proposed. PoseNet and OpenPose,... more
Detecting human abnormal activities is the process of observing rare events that deviate from normality. In this study, an automated camera-based system that is able to detect irregular human behaviour is proposed. PoseNet and OpenPose, which are pre-trained pose estimation models are used to detect the person in the frame and extract the body keypoints. Such data is used to train two types of AutoEncoders based on LSTM and CNN units in a semi-supervised approach where the goal is to learn a general representation of the normal behaviour. Evaluated on a challenging realistic video dataset, the results show that both types of models were able to correctly distinguish between normal and abnormal data sequences, with an average F-score of 0.93. The results also show that the proposed method outperformed similar work done on the same dataset. Furthermore, it was also determined that pose estimated data compares very well with sensor data. This shows that pose estimated data can be infor...
Research Interests:
Image Inpainting techniques are generally challenging to evaluate objectively due to the lack of comparative data, as a reference image of the new scene, does not exist.. This paper presents an approach that uses our newly released... more
Image Inpainting techniques are generally challenging to evaluate objectively due to the lack of comparative data, as a reference image of the new scene, does not exist.. This paper presents an approach that uses our newly released dataset specifically designed to allow objective evaluation of inpainting techniques. In this work we demonstrate how traditional in-painting techniques can be objectively evaluated and compared together with modern deep learning and adversarial approaches. We further demonstrate how an unsupervised technique compares better than deep learning approaches.
One of the emerging challenges in 3D technologies, such as augmented reality, is the ability to thoroughly select an object in a given scene. Most approaches deal with the selection of a 2D object making it difficult for the computing... more
One of the emerging challenges in 3D technologies, such as augmented reality, is the ability to thoroughly select an object in a given scene. Most approaches deal with the selection of a 2D object making it difficult for the computing device to efficiently process the end result and to render a newly blended object into the scene without perceivable artefacts in the effected region. This paper presents a solution that takes advantage of the texture and depth information in the process of representing and therefore accurately selecting an object in a scene. Moreover, the technique allows for further segmentation of the selected object, referred to as ‘intra-object segmentation’, based on the depth information. The result is an object that is split in layers which facilitates the subsequent editing of the scene. Results show the efficacy of the proposed solution. The intra-object segmented output can be used as input to blending methods, such as inpainting, where these can be applied ...
Research Interests:
Bib number recognition (BNR) from unstructured marathon images can be a challenging task. This is because the images captured at these events are very inconsistent since, they are often captured by multiple photographers, at various... more
Bib number recognition (BNR) from unstructured marathon images can be a challenging task. This is because the images captured at these events are very inconsistent since, they are often captured by multiple photographers, at various locations and times. This results in images containing different backgrounds, angles and illumination. The images often contain multiple participants in various poses, where the bib numbers can be obstructed by the participants themselves. The bib numbers are often printed on flexible paper and can easily be deformed which distorts the printed numbers. In this work we present a BNR system based on deep learning which is able to locate bib numbers in unstructured, complex marathon images. Using the segmented bib numbers the system then, recognizes the digits and finally outputs the bib numbers that it was able to detect in the image. The first stage consists of a fully Convolutional Neural Network (CNN) to segment the bib numbers while the second stage co...
In this paper we are proposing an Android mobile application. The main idea behind this system is to make use of localisation techniques together with information extraction techniques in order to develop a localised mobile price... more
In this paper we are proposing an Android mobile application. The main idea behind this system is to make use of localisation techniques together with information extraction techniques in order to develop a localised mobile price comparison application. This system extracts information from the web given a particular location and a source and provides price-comparison information to the end user on the mobile device. This paper briefly shows the background, methodology and evaluation of this decision support system for shoppers.
Research Interests:
Analysing video data requires the use of different models trained to retrieve or process data for a particular task. In this paper, we introduce an approach to represent the visual context within a video as queryable information. Through... more
Analysing video data requires the use of different models trained to retrieve or process data for a particular task. In this paper, we introduce an approach to represent the visual context within a video as queryable information. Through the use of computer vision techniques, we can detect and classify objects. Our system processes these classifications in order to construct a queryable data-set referred to as the real world model. The advantage of this approach is that through the formalisation of the information, we can create generic queries to retrieve information. This approach allows for processing to be done on edge devices such as embedded cameras while only transmitting detected information reducing the transmission bandwidth as well as infrastructural costs. The final recognition data is processed on the cloud. The analysed case study works on video traffic representation - an experiment around the transport domain. We evaluate and validate our approach by posing several q...
Several historic sites in the Maltese Islands are not easy to explore; their construction typology or fragile state of conservation can impede accessibility to all. The Saint Paul’s Catacombs in Rabat is one such historic site. With this... more
Several historic sites in the Maltese Islands are not easy to explore; their construction typology or fragile state of conservation can impede accessibility to all. The Saint Paul’s Catacombs in Rabat is one such historic site. With this in mind, can Virtual Reality (VR) be used to add value to the user experience (UX) of someone exploring these heritage sites? And if so, how can this value be added? This chapter attempts to answer these questions through a qualitative study of a VR system created for the St Paul’s Catacombs in Malta. An assessment of similar applications created for other cultural heritage sites around the world is conducted. This chapter also describes how this VR experience was created through the use of Game Engines, Laser Scanning and other technologies.
This study exploits the selfie phenomenon in order to develop an application that motivates city exploration and tourism to encourage greater mobility. The application uses a reward based system that rewards the user at specific target... more
This study exploits the selfie phenomenon in order to develop an application that motivates city exploration and tourism to encourage greater mobility. The application uses a reward based system that rewards the user at specific target locations by allowing the user to take pictures and selfies at these landmarks only. This in turn encourages the user the visit more sites and unlock further content. Initially the research looks at evaluating whether taking a ‘selfie’ is sufficiently motivating for a user to be regarded as a reward. The survey results obtained show a strong correlation between photography, ‘selfie’ and travel. A number of users were then given a prototype of the application to use in Valletta the capital city of Malta. The users were then interviewed, the interviews established that users find such an approach for city exploration to be meaningful. Ultimately this means that through the application tourists are encouraged to explore more of the captivating sites foun...
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Object selection is a challenge in computer vision since it is generally a trade-off between accuracy and performance. A popular approach is the use of bounding boxes around objects that are to be selected. Other common techniques provide... more
Object selection is a challenge in computer vision since it is generally a trade-off between accuracy and performance. A popular approach is the use of bounding boxes around objects that are to be selected. Other common techniques provide a set of objects from which the user can then choose. The method presented in this paper is designed around the priority of performance and granular selection of objects in the scene. Experiments performed on a non-parallel implementation of the proposed solution return results in an average time of 0.043s. The technique also returned very good results in the processing of objects that are partially occluded, hence enabling future work in improved identification and recognition of such objects.
Cheap depth sensors that can be integrated in consumer cameras provide additional data that can be used for improved post-processing results of the captured images. Removal of objects in a scene is one such editing procedure that demands... more
Cheap depth sensors that can be integrated in consumer cameras provide additional data that can be used for improved post-processing results of the captured images. Removal of objects in a scene is one such editing procedure that demands inpainting techniques that limit noticeable artifacts generated in the process. In this paper, a monoscopic inpainting technique that uses depth information to process results is presented. It allows users to select an object from the foreground that needs to be removed and then inpaints this region from the neighborhood. This approach uses texture and depth information and is pipelined in a way that allows for parallelization. Results are returned in 0.034 seconds on average. A mean opinion score evaluation was carried out and the current technique scored an average of 3.24 from a scale of 5 on the quality of inpainted regions. This exercise was held to identify the attributes that need to be improved in future implementations.
This study1 sought to understand the concept of Gamification, the science that drives it, how to successfully implement Gamification and the issues that may arise when Gamification is introduced in a corporate environment. This was... more
This study1 sought to understand the concept of Gamification, the science that drives it, how to successfully implement Gamification and the issues that may arise when Gamification is introduced in a corporate environment. This was achieved through the development of 2 artifacts that were used in an experimental exercise to test the relevance and acceptance of Gamification in Project Management. Due to the limited secondary data available related to Gamification for project management within a corporate environment, this research project conducts exploratory, qualitative research to provide further insight in relation to this topic. Encouraging data in favour of the introduction of Gamification was extracted with a number of discussion points which require further evaluation.
Research Interests:
[47] claims that education is there to guarantee that all students, irrespective of their background can benefit from the learning experience to the full. This is achieved by ensuring an active participation in the community. To do so,... more
[47] claims that education is there to guarantee that all students, irrespective of their background can benefit from the learning experience to the full. This is achieved by ensuring an active participation in the community. To do so, when considering our fast changing world, educators must move away from old methodologies and experiment with new ones in order to allow their students to reach their goals. This does not mean that the introduction of technology in the classroom is a guarantee of success.
Research Interests:
We are increasingly realising that the exposure of children and young people to technology is strongly affecting the way in which society develops. Individuals, who do not find the complexity of the digital era and constant updates in the... more
We are increasingly realising that the exposure of children and young people to technology is strongly affecting the way in which society develops. Individuals, who do not find the complexity of the digital era and constant updates in the field of technology problematic, are generally referred to as ‘Digital Natives (DNs). This notion was introduced by Marc Prensky [254] when he defined the gap in the way these two generations deal with computers and the Internet. Prensky’s essays [256] [254] discuss the concept of Digital Natives and Immigrants in the education context but this book takes it even further.
The incidence of football injuries may be double the amount of injuries happening in sports such as basketball, and is estimated to range from 10 to 35 injuries per 1000 game hours. This high risk of injuries in football is evident in... more
The incidence of football injuries may be double the amount of injuries happening in sports such as basketball, and is estimated to range from 10 to 35 injuries per 1000 game hours. This high risk of injuries in football is evident in professional, amateur and recreational levels. Despite the significant increase in female participation as well as the well-known injury risks, research on women football players is very limited as most studies are still focusing on injuries experienced by the male gender . Considering the gap existing in the field of applied research, this study has explored the rate of injuries in senior Women football players in Malta. It has also looked at the underlying causes of these injuries during the 2018/2019 season. Ethical clearance was obtained from the Ethics Board within the Malta College of Arts Science and Technology, and permission was granted by the Malta Football Association. The sample was made up of 100 participants from 111 players over 16 years...