Spatio-Temporal Search in Lifelog Data (Bachelor Thesis, Finished)
Author
Description
As digital archives continue to grow, the ability to search and navigate through them has become increasingly important. This challenge is especially visible in lifelogging, an internet trend about the systematic documentation of everyday life with the help of body-mounted cameras, heart-rate sensors, and sleep monitors. The captured information, however, is only useful when it can be retrieved intuitively. A natural way for users to recall and replay past events is through spatio-temporal search queries, yet existing multimedia retrieval systems often lack integrated support for it. This thesis addresses this gap by extending the open-source multimedia retrieval stack vitrivr with spatio-temporal search capabilities. On the backend, spatial and temporal metadata can now be extracted from images, persisted in PostgreSQL/PostGIS, and queried natively through newly developed operators such as the SpatialBooleanQuery, ANDBooleanQuery, and the IntersectionAggregator. On the frontend, a web-based user interface was implemented in Angular, enabling users to define temporal ranges and spatial filters interactively on a map and to explore query results either geographically or in a gallery view. The implementation was evaluated through a user study on the LSC’24 dataset, which showed that participants were able to solve retrieval tasks and generally found the interface intuitive. Feedback highlighted the importance of clear filter handling, result presentation, and query management as areas for small further refinements. Overall, the results demon- strate that spatio-temporal search can be seamlessly integrated into the vitrivr ecosystem, closing a key functional gap and providing an important step toward more user-friendly lifelog retrieval.
Start / End Dates
2025/04/28 - 2025/08/27