Skip to content

MSc projects

LLM-based Assistant For Accessible Video Description

Accessible Video Descriptions (AVD) are narrative audio tracks describing important visual content for blind and low-vision users. The project aims to develop a conversational agent (LLM-based assistant) that interacts with users to generate relevant video descriptions on-demand. The output will be an application demonstrating interactive AVD generation. Example datasets are available at https://www.robots.ox.ac.uk/~vgg/research/autoad/.

Computer Vision Analysis of Heritage Engineering Drawings

Heritage engineering drawings stored in museums and archives contain valuable technical knowledge but are difficult to analyse and search due to noise, ageing, and limited metadata. This project aims to develop computer vision methods to analyse and extract information from historical mechanical drawings. The work may include preprocessing scanned images, detecting components, extracting annotations, and building visual indexing or retrieval tools. The outcomes will support digital preservation, archival search, and engineering heritage research. Example heritage datasets and collections can be accessed via the Museum of English Rural Life (MERL): https://merl.reading.ac.uk/collections/

360-degree imagery enables immersive capture of natural environments but poses challenges for automated ecological analysis due to distortion and wide field-of-view. This project will develop computer vision methods to analyse biodiversity elements visible within 360-degree images, focusing on visual cues that relate to high-biodiversity environments. This includes analysing visual spectrum properties (e.g., amount and distribution of green vegetation) as well as semantic elements (e.g., number of plants, trees, and habitat structures). The work will involve spherical image preprocessing, segmentation or detection modelling, and aggregation of ecological indicators. Example datasets are available via the Synthesis of Systematic Resources (SYN): https://syns.soton.ac.uk/

Prediction and Visualization of User Behavior in 360-degree Video

Understanding how users explore 360-degree video is essential for immersive media delivery and experience design. This project aims to develop models to predict user viewing behavior based on head-movement and viewport datasets. In addition, interactive visualization tools will be created to display attention heatmaps and scanpaths. The work will support research in VR analytics, adaptive streaming, and immersive storytelling. Example dataset is available at: https://gitlab.com/miguelfromeror/head-motion-prediction

Prediction and Visualization of Biodiversity Intactness Index (BII)

The Biodiversity Intactness Index (BII) measures the impact of human activities on biodiversity by assessing species abundance across ecosystems. Using datasets such as PREDICTS, this project will develop predictive models and interactive visualizations to explore biodiversity intactness across spatial and temporal scales. The work will combine environmental data modelling with geospatial visualization to support biodiversity monitoring and policy research. A tutorial is available at: https://adrianadepalma.github.io/BII_tutorial/