Impactful Applied Computing Technologies
IMPACT is a multidisciplinary research team within IDLab that specializes in AI techniques for multimedia data. The group, led by Prof. Steven Verstockt, researches IMPactful Applied Computing Technologies. IMPACT is active in a broad range of applied AI projects in sports data science, cultural heritage, healthcare, and media. The main aim of IMPACT is to assist end-users in these domains with their data hurdles and build innovative applications using their data.
IMPACT’s current research focuses on optimizing data collection workflows, analyzing multimodal data using AI, computer vision, and GEO-ICT. Over the last years, the IMPACT research group has gotten involved in several media-related projects/proposals to share their expertise on spatio-temporal data collection, filtering, classification, enrichment, mapping, and visualization.
The team, comprising around 10 members, includes postdocs Maarten Slembrouck and Joachim Taelman, who contribute to the Sports Data Science and Health domain. The sub-team IMPACT-on-Heritage is led by prof. Dieter De Witte and postdoc Kenzo Milleville.
The IMPACT team is also responsible for various teaching tasks within the faculty of engineering and architecture. They teach courses on applied AI, Big Data technologies, computer vision, sports data science, and digital heritage.
In addition to our substantial research publications, the IMPACT team is widely recognized as influencers in both sports data science and digital heritage, with frequent mentions in the media and invitations to speak as experts at a wide range of events.
Sports Data Science
- Safety: The group investigates smart sensing and data analytics methodologies to enhance athlete safety. Several safety solutions of IMPACT have, for example, been implemented by the Union Cycliste Internationale (UCI) and are currently in use at the World Tour level of race cycling.
- Storytelling: We collaborate with leading broadcasters to explore advanced metadata extraction techniques. By developing novel scene understanding, object detection, and tracking tools, we automatically generate contextual information that empowers the creation of enhanced graphics, immersive storytelling, and augmented reality applications. IMPACT’s goal is to revolutionize the viewing experience by developing innovative storytelling solutions.
- Performance: We design and examine a wide variety of software and hardware tools for collecting and analyzing athlete data. One of our flagship projects is the Wireless Cycling Network deployed at the Eddy Merckx Cycling Center in Ghent. This innovative system integrates data from wearable sensors and timing loops in real time, providing coaches with a comprehensive overview of athlete performance during and after training sessions. By analyzing sensor data, video footage, and timing information, we can offer valuable insights to optimize training strategies and maximize athletic potential.
Heritage
- Enrichment: We enrich the diverse digital collections of GLAM institutions and natural history collections through the strategic application of AI techniques. Our research spans many different collections, from historical maps (Artemis), herbaria and natural history collections (DiSSCo), and museum collections (Hensor). While these collections hold vast amounts of data, a significant portion remains underutilized. Therefore, we fine-tune and customize state-of-the-art AI models to enrich these datasets, unlocking their full potential.
- Search and Retrieval: This additional metadata significantly improves the accessibility of these collections by powering novel interfaces that support both semantic and visual search. By leveraging multimodal vision models and LLMs, users and researchers can explore and query the collections intuitively via natural language. Together with FARO we created a pocket guide that engages in dialogue about heritage given user questions or pictures. At Nerdland we presented a pose-based search engine that allowed children to query art collections by making poses in front of a camera.
FAIR data publication
We also focus on Linked Open Data and Semantic Web technologies to improve interoperability between multiple institutions. Currently, the project MetaBelgica tries to merge data that overlaps between institutions: people who are authors, painters, etc. are fused.
