Making the link between human emotion and music is challenging. Our aim was to produce an efficient system that emotionally rates songs from multiple genres. To achieve this, we employed a series of online self-report studies, utilising Russell's circumplex model. The first study (n = 44) identified audio features that map to arousal and valence for 20 songs. From this, we constructed a set of linear regressors. The second study (n = 158) measured the efficacy of our system, utilising 40 new songs to create a ground truth. Results show our approach may be effective at emotionally rating music, particularly in the prediction of valence.

Visual information varying over time is typically captured by cameras that acquire data via images (frames) equally spaced in time. Using a different approach, Neuromorphic Vision Sensors (NVSs) are emerging visual capturing devices that only acquire information when changes occur in the scene. This results in major advantages in terms of low power consumption, wide dynamic range, high temporal resolution, and lower data rates than conventional video. Although the acquisition strategy already results in much lower data rates than conventional video, such data can be further compressed. To this end, in this paper we propose a lossless compression strategy based on point cloud compression, inspired by the observation that, by appropriately reporting NVS data in a $(x,y,t)$ tridimensional space, we have a point cloud representation of NVS data. The proposed strategy outperforms the benchmark strategies resulting in a compression ratio up to 30% higher for the considered.

Visual surveillance systems have been improving rapidly over the recent past, becoming more capable and pervasive with incorporation of artificial intelligence. At the same time such surveillance systems are exposing the public to new privacy and security threats. There have been an increasing number of reports of blatant abuse of surveillance technologies. To counteract this, data privacy regulations (e.g. GDPR in Europe) have provided guidelines for data collection and data processing. However, there is still a need for a private and secure method of model training for advanced machine learning and deep learning algorithms. To this end, in this paper we propose a privacy-preserved method for visual surveillance. We first develop a dataset of privacy preserved videos. The data in these videos is masked using Gaussian Mixture Model (GMM) and selective encryption. We then train high-performance object detection models on the generated dataset. The proposed method utilizes state-of-art object detection deep learning models (viz. YOLOv4 and YOLOv5) to perform human/object detection in masked videos. The results are encouraging, and are pointers to the viability of the use of modern day deep learning models for object detection in privacy-preserved videos.

The surveillance videos taken from dynamic cam-eras are susceptible to multiple security threats like replay attacks, man-in-the-middle attacks, pixel correlation attacks etc. Using unsupervised learning, it is a challenge to detect objects in such surveillance videos, as fixed objects may appear to be in motion alongside the actual moving objects. But despite this challenge, the unsupervised learning techniques are efficient as they save object labelling and model training time, which is usually a case with supervised learning models. This paper proposes an effective computer vision-based object identification algorithm that can detect and separate stationary objects from moving objects in such videos. The proposed Advanced Flow Of Motion (AFOM) algorithm takes advantage of motion estimation between two consecutive frames and induces the estimated motion back to the frame to provide an improved detection on the dynamic camera videos. The comparative analysis demonstrates that the proposed AFOM outperforms a traditional dense optical flow (DOF) algorithm with an average increased difference of 56 % in accuracy, 61 % in precision, and 73 % in pixel space ratio (PSR), and with minimal higher object detection timing.

Fire hazards cause huge ecological, social and economical losses in day to day life. Due to the rapid increase in the prevalence of fire accidents, it has become vital to equip the assets with fire prevention systems. There have been numerous researches to build a fire detection model in order to avert such accidents, with recent approaches leveraging the enormous improvements in computer vision deep learning models. However, most deep learning models have to compromise with their performance and accurate detection to maintain a reasonable inference time and parameter count. In this paper, we present a customized lightweight convolution neural network for early detection of fire. By virtue of low parameter count, the proposed model is amenable to embedded applications in real-time fire monitoring equipment, and even upcoming fire monitoring approaches such as unmanned aerial vehicles (drones). The fire detection results show marked improvement over the predecessor low-parameter-count models, while further reducing the number of trainable parameters. The overall accuracy of FireNet-v2, which has only 318,460 parameters, was found to be 98.43% when tested over Foggia's dataset.

Abstract—Control flow hijacking has been a major challenge in software security. Several means of protections have been developed but insecurities persist. This is because existing protections have sometimes been circumvented while some resilient protections do not cover all applications. Studies have revealed that a holistic way of tackling software insecurity could involve watchdog monitoring and detection via Control Flow Integrity (CFI). The CFI concept has shown a good measure of reliability to mitigate control flow hijacking. However, sophisticated attack techniques in the form of Return Oriented Programming (ROP) have persisted. A flexible protection is desirable, which not only covers as many architecture structures as possible but also mitigates known resilient attacks like ROP. The solution proffered here is a hybrid of CFI and watchdog timing via inter-process signaling (IP-CFI). It is a software-based protection that involves recompilation of the target program. The implementation here is on vulnerable RISC-V-based process but is flexible and could be adapted on other architectures. We present a proof of concept in IP-CFI which when applied to a vulnerable program, ROP is mitigated. The target program incurs a run-time overhead of 1.5%. The code is available.

Cyber threats to organisations across all industries are increasing in both volume and complexity, leading to significant, and sometimes severe, conse-quences. The common weakest link in organisations security is the human vulnerability. The sudden popularity of remote-working due to the Covid-19 pandemic opened organisations and their employees up to more risks, partic-ularly as many workers believe that they are more distracted when at home. Existing cyber training using a ‘one-size-fits-all’ approach has been proven inefficient/ineffective and the need for a more fit-for-purpose training is re-quired. When it comes to cyber training, we know that there is no single-training-fits-all solution – people have different technical skills, different prior knowledge and experience, are in different roles, exposed to different security risks, and require knowledge that is relevant to what they do. This study makes a case for tailored role-based cybersecurity training suitable for awareness within organisations across multiple industries. The study ex-plores the strengths and weaknesses of existing cyber training and literature to make recommendations on efficient awareness and training programme strategies. The study carries out knowledge and task analysis of job roles to create profiles of skills and knowledge they require. These are grouped by topic and level to form scenario-based multiple-choice questions which are mapped to create a Cyber Awareness Platform (CAP). A CAP prototype is in-troduced as a flexible web-based system allowing users to assess their prior knowledge and skills personalised to their role. Knowledge gaps and training needs are identified, and recommendations are tailored to the individual. Ini-tial analysis of CAP shows promising results, indicating that such role-sensitive solution would be highly beneficial to users. This offers further de-velopment opportunities in producing an all-in-one cyber assessment and training platform.

We report on an ongoing collaboration between epidemiological modellers and visualization researchers by documenting and reflecting upon knowledge constructs – a series of ideas, approaches and methods taken from existing visualization research and practice – deployed and developed to support modelling of the COVID-19 pandemic. Structured independent commentary on these efforts is synthesized through iterative reflection to develop: evidence of the effectiveness and value of visualization in this context; open problems upon which the research communities may focus; guidance for future activity of this type; and recommendations to safeguard the achievements and promote, advance, secure and prepare for future collaborations of this kind. In describing and comparing a series of related projects that were undertaken in unprecedented conditions, our hope is that this unique report, and its rich interactive supplementary materials, will guide the scientific community in embracing visualization in its observation, analysis and modelling of data as well as in disseminating findings. Equally we hope to encourage the visualization community to engage with impactful science in addressing its emerging data challenges. If we are successful, this showcase of activity may stimulate mutually beneficial engagement between communities with complementary expertise to address problems of significance in epidemiology and beyond. https://ramp-vis.github.io/RAMPVIS-PhilTransA-Supplement/

We present the results of a double-blind phase 2b randomized control trial that used a custom built virtual reality environment for the cognitive rehabilitation of stroke survivors. A stroke causes damage to the brain and problem solving, memory and task sequencing are commonly affected. The brain can recover to some extent, however, and stroke patients have to relearn how to carry out activities of daily living. We have created an application called VIRTUE to enable such activities to be practiced using immersive virtual reality. Gamification techniques enhance the motivation of patients such as by making the level of difficulty of a task increase over time. The design and implementation of VIRTUE is described together with the results of the trial conducted within the Stroke Unit of a large hospital. We report on the safety and acceptability of VIRTUE. We have also observed particular benefits of VR treatment for stroke survivors that experienced more severe cognitive impairment, and an encouraging reduction in time spent in the hospital for all patients that received the VR treatment.

Mild Cognitive Impairment (MCI) is a definition of the diagnosis of early memory loss and disorientation. This study aims to identify people’s symptoms through technology. However, machine learning (ML) can classify Cognitive Normal (CN) and Mild Cognitive Impairment (MCI) and Early Mild Cognitive Impairment (EMCI) using standard assessments from the Alzheimer’s Disease Neuroimaging Initiative (ADNI); Montreal Cognitive (MoCA), Mini-Mental State Examination (MMSE), Functional Activities Questionnaire (FAQ). Consequently, a Multilayer Perceptron (MLP) model was assembled into tables; MCI vs CN, MCI vs EMCI, and CN vs MCI. Additionally, an MLP model was developed for CN vs MCI vs EMCI. As a result, of advanced model performance, a cascade 3-path categorisation approach was created. Similarly, the exploitation of meta-analysis indicated a combination of MLP models (MCI vs CN, MCI vs EMCI, and CN vs MCI) with an overall accuracy within an acceptable limit. In addition, better results were found when assessments were combined rather than individually. Furthermore, applying class weights and probability thresholds could improve the MLP framework by performance achieving a balanced specificity and sensitivity ratio. Altering class weights and probability thresholds when training the MLP neuro network model, the sensitivity and Accuracy could be progressed further. In conclusion, ML, VR and electrodermal activity are constrained. Introducing the possibility of activity-based applications to enhance innovative solutions for cognitive impairment diagnosis and treatment.