A5075611724- Traffic Prediction and Management Techniques
- Semantic Web and Ontologies
- Traffic and Road Safety
- Autonomous Vehicle Technology and Safety
- Advanced Database Systems and Queries
- Natural Language Processing Techniques
- Archaeology and Natural History
- Mobile Learning in Education
- Injury Epidemiology and Prevention
- Urban Transport and Accessibility
- Housing Market and Economics
- Teaching and Learning Programming
- Software Testing and Debugging Techniques
- Vehicle License Plate Recognition
- Vehicle emissions and performance
- Socioeconomics of Resources and Conservation
- Health disparities and outcomes
- Recreation, Leisure, Wilderness Management
- Human Mobility and Location-Based Analysis
- Botany, Ecology, and Taxonomy Studies
- Data Visualization and Analytics
- Wildlife-Road Interactions and Conservation
- Urban, Neighborhood, and Segregation Studies
- Video Surveillance and Tracking Methods
- Data Management and Algorithms
University College London
2024-2025
Trinity College
2017
Hartford Financial Services (United States)
2017
Large language models (LLMs) exhibit emerging geospatial capabilities, stemming from their pre-training on vast unlabelled text datasets that are often derived the Common Crawl corpus. However, content within CC remains largely unexplored, impacting our understanding of LLMs' spatial reasoning. This paper investigates prevalence data in recent releases using Gemini, a powerful model. By analyzing sample documents and manually revising results, we estimate between 1 5 6 contain information...
Sentence transformers are language models designed to perform semantic search. This study investigates the capacity of sentence transformers, fine-tuned on general question-answering datasets for asymmetric search, associate descriptions human-generated routes across Great Britain with queries often used describe hiking experiences. We find that have some zero-shot capabilities understand quasi-geospatial concepts, such as route types and difficulty, suggesting their potential utility...
Creating routes across open areas is challenging due to the absence of a defined routing network and complexity environment, in which multiple criteria may affect route choice. In context urban environments, research has found Visibility Spider-Grid subgraphs be effective approaches that generate realistic routes. However, case studies presented typically focus on plazas or parks with entry exit points; little work been carried out date creating rural settings, are complex environments...
The Common Crawl (CC) corpus is the largest open web crawl dataset containing 9.5+ petabytes of data captured since 2008. instrumental in training large language models, and as such it has been studied for (un)desirable content, distilled smaller, domain-specific datasets. However, to our knowledge, no research dedicated using CC a source annotated geospatial data. In this paper, we introduce an efficient pipeline extract user-generated tracks from GPX files found CC, resulting multimodal...
Global positioning system (GPS) trajectories recorded by mobile phones or action cameras offer valuable insights into sustainable mobility, as they provide fine-scale spatial and temporal characteristics of individual travel. However, the high volume, noise, lack semantic information in this data poses challenges for storage, analysis, applications. To address these issues, we propose an end-to-end pipeline named CycleTrajectory processing high-sampling rate GPS trajectory from cameras,...
Predicting traffic accidents is the key to sustainable city management, which requires effective address of dynamic and complex spatiotemporal characteristics cities. Current data-driven models often struggle with data sparsity typically overlook integration diverse urban sources high-order dependencies within them. Additionally, they frequently rely on predefined topologies or weights, limiting their adaptability in predictions. To these issues, we introduce Spatiotemporal Multiview...
Road traffic crashes cause millions of deaths annually and have a significant economic impact, particularly in low- middle-income countries (LMICs). This paper presents an approach using Vision Language Models (VLMs) for road safety assessment, overcoming the limitations traditional Convolutional Neural Networks (CNNs). We introduce new task ,V-RoAst (Visual question answering Assessment), with real-world dataset. Our optimizes prompt engineering evaluates advanced VLMs, including...
Panoramic cycling videos can record 360{\deg} views around the cyclists. Thus, it is essential to conduct automatic road user analysis on them using computer vision models provide data for studies safety. However, features of panoramic such as severe distortions, large number small objects and boundary continuity have brought great challenges existing CV models, including poor performance evaluation methods that are no longer applicable. In addition, due lack with annotations, not easy...
The “Waitrose effect” captures the notion that presence of stores operated by Waitrose, an upmarket UK grocer, increases value nearby real estate. This paper considers broader relationship between Waitrose store locations and neighbourhood type comparing health wealth neighbourhoods with without access to in England. Whilst we do not seek imply causality, demonstrate better health, wellbeing, falling within a catchment. In those neighbourhoods, median home prices were almost 2.5 times higher...
The Mobile Computer Science Principles curriculum collects data on embedded Quizly programming exercises, which are based the App Inventor version of Blockly. We have recently started mining this to determine whether student performance exercises matches our assumptions about difficulty individual exercises. Various analytic techniques, such as linear regression, used identify those features that most determinative problem difficulty. Our analysis supports number abstractions may be a useful...